Ammonium Feedback Control in Wastewater Treatment Plants.pdf

ACTA UNIVERSITATIS UPSALIENSIS Uppsala Dissertations from the Faculty of Science and Technology

104

Linda Åmand

Ammonium Feedback Control in Wastewater Treatment Plants

Dissertation presented at Uppsala University to be publicly examined in Polhemsalen, Ångströmlaboratoriet, Lägerhyddsvägen 1, Uppsala, Tuesday, 29 April 2014 at 09:15 for the degree of Doctor of Philosophy. The examination will be conducted in English. Abstract Åmand, L. 2014. Ammonium Feedback Control in Wastewater Treatment Plants. Uppsala Dissertations from the Faculty of Science and Technology 104. 256 pp. Uppsala: Acta Universitatis Upsaliensis. ISBN 978-91-554-8900-7. The aeration process is often the single largest consumer of electricity in a wastewater treat-ment plant. Aeration in biological reactors provides microorganisms with oxygen which is required to convert ammonium to nitrate. Ammonium is toxic for aqueous ecosystems and contributes to eutrophication. The importance of aeration for the treatment results in combina-tion with the high costs motivates automatic control of the aeration process. This thesis is devoted to ammonium feedback control in municipal wastewater treatment plants. With ammonium feedback control, the aeration intensity is changed based on a meas-urement of the outlet ammonium concentration. The main focus of the thesis is design, imple-mentation, evaluation and improvement of ammonium PI (proportional-integral) controllers. The benefits of ammonium feedback control are established through long-term experiments at three large wastewater treatment plants in Stockholm, Sweden. With ammonium feedback control, energy savings up to around 10 % were achieved compared to keeping the dissolved oxygen concentration constant. The experiments generated several lessons learned with regard to implementation and evaluation of controllers in full-scale operation. The thesis has established guidelines on how to design ammonium feedback controllers for situations when cost-effective operation is the overall aim. Simulations have demonstrated the importance to limit the dissolved oxygen concentration in the process and under what condi-tions the energy saving with ammonium feedback control is large. The final part of the thesis treats improvements of ammonium PI control through minor modifications to the control structure or controller. Three strategies were studied: gain sched-uling control, repetitive control, and a strategy reacting to oxygen peaks in the last aerobic zone. The strategies all had their benefits but the ammonium feedback controller was the key factor to improved aeration control. Keywords: aeration control, ammonium feedback control, biological nitrogen removal, fullscale experiments, gain scheduling control, PI control, process control, repetitive control, wastewater treatment Linda Åmand, Department of Information Technology, Division of Systems and Control, Box 337, Uppsala University, SE-75105 Uppsala, Sweden. © Linda Åmand 2014 ISSN 1104-2516 ISBN 978-91-554-8900-7 urn:nbn:se:uu:diva-219941 (http://urn.kb.se/resolve?urn=urn:nbn:se:uu:diva-219941) Printed by Elanders Sverige AB, 2014

To Johan

ACKNOWLEDGMENTS

The person providing me with the opportunity to become a PhD student with Uppsala University and my supervisor all these years deserves his own chap-ter in this thesis. Bengt Carlsson, I am so very grateful for the years of sup-port, discussions and inspiration.

This project was made possible since Stockholm Vatten, Käppalaförbun-det and Syvab believed in the creation of a joint PhD project with Uppsala University and IVL. I am especially grateful to Christer, Andreas and Kristi-na who supported me, educated me and sometimes during the experiments put up with me over the last four years. The whole foundation of this project is based on their – and their colleagues’ – involvement and dedication.

I would like to thank my extended family of colleagues at IVL Swedish Environmental Research Institute – you know who you are – for all the mo-ments in the IVL kitchen when you made me forget I was at work or had never-ending full-scale experiments running.

To the Systems and Control group in Uppsala: thank you for always mak-ing me feel welcome. It was a pleasure to join your group once a week.

Gustaf Olsson, for being who you are and for sharing your ideas and be-liefs during the course of the project: thank you! Thank you Ulf Jeppsson for providing the benchmark simulation models and for your support.

Special thanks go to all of you who helped revising parts of this thesis over the years: Ulf, Margarida, Jesús, Magnus and Johan. Several master students have written their theses within this research project. Together we have taken important steps towards the final result. Thank you Åsa, Ulrika, Stefan, Elin, Linda and Sofia.

Thank you to family and friends for making me who I am. I wish to acknowledge the support provided by the Swedish Water and

Wastewater Association (Svenskt Vatten) via VA-kluster Mälardalen and the support provided by the Foundation for the Swedish Environmental Re-search Institute (SIVL).

This project has helped the participating treatment plants to save energy and can hopefully inspire other plants to do the same. But the improvements we have produced do not allow you to compensate with an extra five minutes in the shower tomorrow morning. Heating water for domestic use constitute the lion’s share of the energy use within the urban water cycle.

SAMMANFATTNING Ammoniumåterkoppling i avloppsreningsverk

Luftningen av vattnet i det biologiska reningssteget på ett avloppsrenings-verk utgör ofta den enskilt största kostnadsposten för elenergi på ett av-loppsreningsverk. I Sverige motsvarar elenergianvändningen i biosteget knappt hälften av reningsverkets totala energiförbrukning, där den största delen åtgår till luftningen. I ett avloppsreningsverk renas avloppsvatten från syretärande ämnen och ämnen som bidrar till övergödning i recipienten. Kväve är ett av dessa ämnen och kommer i huvudsak in till reningsverket i form av ammonium. Mikroorganismer omvandlar ammonium till nitrat i det biologiska reningssteget, vilket kräver en syrerik miljö.

Denna avhandling behandlar hur man med hjälp av reglerteknik kan göra luftningsprocessen mer energieffektiv. Reglertekniken har idag en given plats i vårt samhälle, och finns i tekniska produkter och processer överallt i vår vardag. Inom processreglering används regulatorer för att styra en pro-cess mot ett uppsatt mål, trots att processen utsätts för yttre påverkan i form av störningar. På ett avloppsreningsverk utgörs den allra största externa stör-ningen av den stora variation i inflöde som sker dagligen. Variationen upp-kommer av att vi invånare har ett dygnsmönster som påverkar när och i vil-ken mängd avloppsvatten kommer till reningsverket. Förutom dygnsvariat-ioner utsätts ett reningsverk även för störningar som är oregelbundna – så som nederbörd – samt långsamma förändringar av t.ex. temperaturen på vattnet in till reningsverket.

Eftersom belastningen in till reningsverket är så pass föränderlig behöver luftningsintensiteten i det biologiska reningssteget anpassas efter processens behov. För mycket luft kostar onödig energi och för lite luft riskerar kvali-teten på utgående avloppsvatten. Reglerteknik är därmed ett användbart red-skap för att anpassa luftningsintensiteten efter förändringar i reningsverks-processen över tid.

Ammoniumåterkoppling är den metod som belyses i detta arbete. Med ammoniumåterkoppling mäts ammoniumhalten kontinuerligt i slutet av det biologiska reningssteget och baserat på dessa mätningar beräknar regulatorn önskad syrehalt i processen. I detta arbete används ammoniumåterkoppling med en enkel PI-regulator – en av de mest använda regulatorerna i process-industrin. Avhandlingen är fokuserad på design, implementering, utvärde-

ring och förbättring av ammoniumåterkoppling med PI-regulator på av-loppsreningsverk.

Den första delen av avhandlingen introducerar ämnet reglerteknik i av-loppsreningsverk, och en litteratursammanställning går igenom olika regler-tekniska metoder som använts för styrning av luftningsprocesser i av-loppsreningsverk. I sammanställningen visar sig ammoniumåterkoppling vara konkurrenskraftig eftersom ammoniumåterkoppling är lätt att imple-mentera och kan bidra till att sänka energiförbrukningen. Dock är de flesta av försöken som publicerats korttidsstudier från ett antal dagar upp till ett par månader.

I tre kapitel i avhandlingen undersöks hur en ammoniumregulator bör de-signas för att både upprätthålla långtgående kväverening i processen men också få processen att arbeta energieffektivt. Den viktigaste enskilda faktorn för energieffektivitet är att sätta en övre gräns på hur högt ammoniumregula-torn får höja syrehalten i processen. Simuleringar med dynamiska process-modeller visar att det inte nödvändigtvis är den snabbaste regulatorn som ger bäst resultat, speciellt om man tillåter höga syrehalter i processen. En snabb regulator tappar snabbt styrförmågan vid höga syrehalter eftersom höga sy-rehalter har en begränsad effekt på kvävereningen.

Under flera års tid har ammoniumåterkoppling testats på Henriksdals re-ningsverk, Käppalaverket och Himmerfjärdsverket. Reningsverken är alla belägna i eller runt Stockholm och renar vatten från omkring 1,5 miljoner invånare. Långtidsförsöken på reningsverken i Stockholm har gett erfaren-heter om hur man på bästa sätt implementerar och utvärderar regulatorer i fullskala på reningsverk. En av dessa erfarenheter är att det är viktigt att utföra långtidsförsök för att få representativa resultat. Under långtidsförsö-ken påvisades energibesparingar på upp till 10 % jämfört med styrning med konstanta syrehalter.

De två viktigaste orsakerna till tidsfördröjningar i implementeringsarbetet var behov av anpassningar i styrsystemet samt givarproblem. Industriella styrsystem kan erbjuda fördelar jämfört med styrsystem där fristående regu-latorer kopplas ihop med ett övervakningssystem. Ammoniumgivaren är nyckeln till att få till en väl fungerande ammoniumåterkoppling, men har fungerat olika bra på de olika verken. Kostnader kopplat till ammoniumåter-koppling som är viktiga att ta i beaktande är kostnaden för ammoniumgiva-ren samt underhåll av densamma.

Två av projektets tre medverkande reningsverk har valt att återkoppla en ammoniummätning från en givare som är placerad efter sedimenteringen i biosteget. Sedimenteringen är det processteg som separerar det biologiska slammet från det renade vattnet. Fördelen med denna placering är att givaren kräver mindre underhåll eftersom miljön är mindre fientlig här än om giva-ren sitter bland mikroorganismerna i luftningssteget. Nackdelen är att sedi-menteringssteget gör att ammoniummätningen blir fördröjd i relation till situationen i luftningsprocessen. Det reningsverk som haft en ammoniumgi-

vare direkt i luftningssteget hade flest problem med att givaren visade felakt-iga värden.

Att utvärdera resultat från fullskalestudier är komplicerat eftersom det finns flera källor till osäkerheter. Jämförelsen av energiförbrukning för olika styrstrategier kompliceras av att parallella linjer i ett reningsverk ofta har kraftigt olika beteenden. Luftflödesmätningarna visade sig vara svåra att lita på i vissa fall. I slutändan blir en utvärdering av energiförbrukningen endast en uppskattning. Det kan vara lockande att både i simuleringsarbeten och vid utvärderingar av fullskaleförsök relatera energiförbrukningen till reningsre-sultatet. Detta har visat sig problematiskt eftersom det kan innebära att man fångar andra korrelationer i data än de man är intresserade av att utvärdera.

Den sista delen av avhandlingen behandlar tre olika metoder som strävar efter att med relativt enkla medel förbättra funktionen hos en ammoniumre-gulator. Parameterstyrning utvärderades på Käppalaverket och visade sig vara ett sätt att spara energi under perioder då ammoniumgivaren överskat-tade ammoniumhalten. En reglerstrategi har testats på Henriksdals renings-verk där syftet var att sänka syrehalterna i den första delen av luftningssteget om ett syreöverskott uppträder i den sista luftade zonen. Den tredje metoden var repetitiv reglering vilket är en metod där regulatorn lär sig hur styrningen bör ske baserat på historiska reglerfel. Samtliga tre metoder hade sina förde-lar, men den viktigaste betydelsen för processen bidrog själva ammoniumre-gulatorn med.

Som ett resultat av forskningsprojektet kommer Käppalaverket och Hen-riksdals reningsverk införa ammoniumåterkoppling i samtliga av verkens linjer. På grund av en kommande stor ombyggnation avvaktar Himmer-fjärdsverket med att ändra styrningen i verket. Ammoniumåterkoppling har mött stor acceptans hos personalen på samtliga tre reningsverk. Avhandlingen är skriven inom ämnesområdet elektroteknik med inriktning mot reglerteknik.

xiii

CONTENTS

Introduction ......................................................................................... 17 1 Motivation ....................................................................................... 17 1.1 Objective ......................................................................................... 18 1.2 Publications ..................................................................................... 18 1.3 Outline ............................................................................................. 19 1.4 Contributions ................................................................................... 21 1.5

Part I Wastewater treatment and aeration control ....................................... 23

Wastewater treatment plants – treatment, modelling and control ........ 25 2 Why wastewater treatment .............................................................. 25 2.1 The wastewater treatment process .................................................. 27 2.2 Control of wastewater treatment plants ........................................... 31 2.3 Introduction to aeration control ....................................................... 37 2.4 Modelling and simulation of wastewater treatment plants .............. 44 2.5

Aeration control – a review ................................................................. 49 3 Introduction ..................................................................................... 49 3.1 Control structures ............................................................................ 50 3.2 Control algorithms .......................................................................... 53 3.3 Control of aeration intensity ........................................................... 53 3.4 Control of the aerobic volume ........................................................ 64 3.5 Comparison between case studies ................................................... 65 3.6 Critical review of aeration control systems ..................................... 70 3.7 Implementing adequate aeration control for full-scale operation ... 73 3.8 Conclusions ..................................................................................... 75 3.9

Henriksdal, Käppala and Himmerfjärden WWTPs ............................. 77 4 Henriksdal WWTP .......................................................................... 77 4.1 Käppala WWTP .............................................................................. 79 4.2 Himmerfjärden WWTP ................................................................... 80 4.3 Similarities and differences ............................................................. 82 4.4

Part II Design of ammonium feedback controllers ..................................... 89

Ammonium PI control and the optimal DO profile ............................. 91 5 Introduction ..................................................................................... 91 5.1 Simulation model ............................................................................ 92 5.2

xiv

Control strategies ............................................................................ 93 5.3 Optimisation problem ..................................................................... 94 5.4 Simulation scenarios ....................................................................... 95 5.5 The optimal DO profile ................................................................... 97 5.6 Comparison to ammonium feedback control .................................. 98 5.7 Discussion ..................................................................................... 107 5.8 Conclusions ................................................................................... 108 5.9

Ammonium PI controller design – benchmark simulations .............. 111 66.1 Introduction ................................................................................... 111 6.2 Model setup ................................................................................... 112 6.3 Simulation settings ........................................................................ 114 6.4 The impact of the controller settings ............................................. 118 6.5 Energy saving for different treatment volumes ............................. 126 6.6 Conclusions ................................................................................... 129

Ammonium PI controller design – plant simulations ........................ 131 7 Introduction ................................................................................... 131 7.1 Model setup ................................................................................... 132 7.2 Control structure and controller settings ....................................... 132 7.3 Aeration system limits .................................................................. 133 7.4 Lambda tuning .............................................................................. 136 7.5 The Henriksdal model ................................................................... 137 7.6 The Käppala model ....................................................................... 141 7.7 The Himmerfjärden model ............................................................ 143 7.8 Comparison between plant models ............................................... 145 7.9

Comparison to lambda tuning .................................................. 149 7.10 Conclusions .............................................................................. 150 7.11

Part III Full-scale ammonium feedback control........................................ 151

Long-term evaluation of full-scale ammonium feedback control ...... 153 8 Introduction ................................................................................... 153 8.1 Control strategies and instrumentation .......................................... 154 8.2 Henriksdal WWTP ........................................................................ 155 8.3 Käppala WWTP ............................................................................ 156 8.4 Himmerfjärden WWTP ................................................................. 156 8.5 Summary of experiments .............................................................. 157 8.6 Evaluation methods ....................................................................... 159 8.7 Ammonium controller performance .............................................. 160 8.8 Estimation of energy saving and cost-benefit analysis ................. 162 8.9

Comparison to simulation results ............................................. 163 8.10 N2O emissions at Käppala WWTP ........................................... 167 8.11 Lessons learnt from controller implementation ........................ 168 8.12 Time aspects ............................................................................. 173 8.13

xv

Energy estimation ..................................................................... 174 8.14 Conclusions .............................................................................. 175 8.15

Part IV Improving ammonium feedback control ...................................... 177

Gain scheduling control ..................................................................... 179 9 Introduction ................................................................................... 179 9.1 Control structure and control strategies ........................................ 180 9.2 Full-scale evaluation at Käppala WWTP ...................................... 180 9.3 Simulations with gain scheduling control ..................................... 184 9.4 Conclusions ................................................................................... 189 9.5

DO deviation control ......................................................................... 191 10 Introduction .............................................................................. 191 10.1 Control structure and control strategies .................................... 192 10.2 Full-scale evaluation at Henriksdal WWTP ............................. 193 10.3 Simulations with DO deviation control .................................... 197 10.4 Conclusions .............................................................................. 198 10.5

Repetitive control............................................................................... 199 11 Introduction .............................................................................. 199 11.1 Control structure and control strategies .................................... 200 11.2 Simulations with repetitive control .......................................... 203 11.3 Conclusions .............................................................................. 206 11.4

Part V Conclusions ................................................................................... 207

Concluding remarks ........................................................................... 209 12 Answering the questions – a summary of the results ............... 209 12.1 Overall conclusions .................................................................. 212 12.2 Future work .............................................................................. 215 12.3

Bibliography ............................................................................................... 217

Appendix ..................................................................................................... 231

A Calibration of activated sludge models .............................................. 233 Objective .................................................................................. 233 A.1 Overall method ......................................................................... 233 A.2 The Henriksdal model .............................................................. 235 A.3 The Käppala model .................................................................. 241 A.4 The Himmerfjärden model ....................................................... 246 A.5 Model settings .......................................................................... 252 A.6 Discussion ................................................................................ 255 A.7

xvi

GLOSSARY

Abbreviations

BOD Biological oxygen demand BSM Benchmark simulation model COD Chemical oxygen demand CSTR Continuous stirred-tank reactor DO Dissolved oxygen ES Energy saving FB Feedback GS Gain scheduling MIMO Multiple input, multiple output MLSS Mixed liquor suspended solids MLVSS Mixed liquor volatile suspended solids MPC Model predictive control PID Proportional - integral - derivative RAS Return activated sludge RC Repetitive control SCADA Supervisory control and data acquisition SISO Single input, single output SP Set-point SV Scheduling variable (in GS) TN Total nitrogen WAS Waste activated sludge WWTP Wastewater treatment plant ZL Zone limit (in GS) Notation

K Controller gain Ki Integral gain KLa Oxygen transfer coefficient N2O Nitrous oxide NH4 Ammonium nitrogen (NH4-N) NO3 Nitrate nitrogen (NO3-N) Qair Air flow rate Ti Integral time

17

1 INTRODUCTION 1

Motivation 1.1Aeration of biological reactors in wastewater treatment plants (WWTPs) is important but costly. Nitrogen and organic substances are removed in the biological treatment process in a municipal WWTP. The microorganisms require oxygen, which is supplied to the water through aeration. The aeration process is often the single largest consumer of electricity at a WWTP.

If the influent load was constant, wastewater treatment would have been a comparatively straightforward task. However, the raw material to a WWTP – the influent wastewater – is constantly changing. The load to a plant has an inherently diurnal pattern as a result of the regular habits of the inhabitants connected to the plant. Despite this daily variation, and despite more irregu-lar disturbances to the process such as rain events, the treatment plant needs to remove harmful substances from the wastewater every day of the week, every week of the year.

Similar to other process industries, methods and tools from process con-trol have improved the performance of wastewater treatment plants the last decades. Process control in combination with new developments within pro-cess technology has enabled the plants to reach more stringent discharge criteria in a more cost-effective manner.

Since the 1970’s when dissolved oxygen (DO) measurements first be-came available, there has been much development within the field of aera-tion control. Aeration control is motivated by the importance of aeration for the treatment process, the high costs of aeration and the large disturbances to the process.

18 1 Introduction

Objective 1.2The overall objective of this work has been to develop and apply methods from automatic control to reduce the energy consumption in the aeration process without compromising nitrogen removal. Implementation and evalu-ation of these methods was to be performed at three WWTPs in Stockholm: Henriksdal WWTP, Käppala WWTP and Himmerfjärden WWTP.

Secondary objectives involved (1) to apply methods that are implementa-ble in the control systems at the participating WWTPs and (2) not to contrib-ute to elevated emissions of nitrous oxide – a strong greenhouse gas.

Early on in the project, ammonium feedback control – determining the DO concentration based on the outlet ammonium concentration – was cho-sen as the main focus of the research. Three reasons were behind this deci-sion: (1) Of the three WWTPs in the project, only one plant had experience of ammonium control prior to the project; (2) a literature review demonstrat-ed the benefits of ammonium feedback control and (3) try simple things first.

Given the overall objective of the project as well as the choice to work with ammonium feedback control, the task became to design, implement, evaluate and improve ammonium feedback controllers.

Publications 1.3The papers which are the basis for this thesis are listed below. In most chap-ters of the thesis various extensions were made to the published material. Journal papers

I Åmand, L., and Carlsson, B., 2012. Optimal aeration control in a nitrifying activated sludge process. Water research 46(7), 2101–10.

II Åmand, L., Olsson, G., and Carlsson, B., 2013. Aeration control – a review. Water Science and Technology 67(11), 2374–2398.

III Åmand, L., and Carlsson, B., 2013. The optimal dissolved oxygen profile in a nitrifying activated sludge process - comparisons to am-monium feedback control. Water Science and Technology 68(3), 641–649.

IV Åmand, L., Laurell, C., Stark-Fujii, K., Thunberg, A., and Carlsson, B., 2014. Lessons learnt from evaluating full-scale ammonium feed-back control in three large wastewater treatment plants. Water Sci-ence and Technology 69(7), doi:10.2166/wst.2014.061.

V Åmand, L., and Carlsson, B. Cost-effective ammonium PI controller design – a simulation study. Submitted for publication.

1.4 Outline 19

Conference papers

VI Åmand, L., and Carlsson, B., 2011. Learning from your mistakes: Repetitive aeration control in continuous wastewater treatment plants, in: 12th Nordic Wastewater Conference. Helsinki, Finland, 14-16 November 2011.

VII Åmand, L., Nygren, J., and Carlsson, B., 2011. Applications of repet-itive control in activated sludge processes, in: 8th IWA Symposium on Systems Analysis and Integrated Assessment. San Sebastian, Spain, 20-22 June 2011.

VIII Åmand, L., and Carlsson, B., 2012. Energy efficient ammonium feedback control, in: New Developments in IT & Water. Amsterdam, The Netherlands, 4-6 November 2012.

IX Åmand, L., and Carlsson, B., 2013. Parameter scheduling in ammo-nium feedback control, in: 11th IWA Conference on Instrumentation, Control and Automation. Narbonne, France, 18-20 September 2013.

X Åmand, L., Laurell, C., Stark-Fujii, K., Thunberg, A., and Carlsson, B., 2013. Long-term evaluation of full-scale ammonium control in three large WWTPs, in: 11th IWA Conference on Instrumentation, Control and Automation. Narbonne, France, 18-20 September 2013.

XI Åmand, L., Laurell, C., Stark-Fujii, K., Thunberg, A., and Carlsson, B., 2013. Full-scale ammonium control in three wastewater treatment plants in Stockholm, in: 13th Nordic Wastewater Conference 2013. Malmö, Sweden, 8-10 October 2013.

XII Åmand, L., and Carlsson, B., 2014. Aeration control with gain scheduling in a full-scale wastewater treatment plant, in: 19th IFAC World Congress. Cape Town, South Africa, 24-29 August 2014.

Outline 1.4 Part I: Wastewater treatment and aeration control 1.4.1

What is aeration control and how does it improve the operation of a WWTP? The first part of the thesis introduces wastewater treatment, modelling and control. The wastewater treatment process is presented, as well as the fun-damentals of automatic control and how control can be applied in a WWTP (Chapter 2). Special attention is given to control of aeration systems through a literature review of recent publications on the topic (Chapter 3, paper II). Henriksdal, Käppala and Himmerfjärden WWTPs are introduced (Chapter 4).

20 1 Introduction

Part II: Design of ammonium feedback controllers 1.4.2How should the settings in an ammonium PI controller be chosen? Ammo-nium feedback control is compared to the optimal DO profile in a simplified model of a wastewater treatment plant in Chapter 5 (paper I, III, VIII). The optimal DO profile is created through mathematical optimisation. In Chapter 6 and 7 (paper V), simulations are performed with a wide range of ammoni-um PI controller settings. The model in Chapter 7 is the Benchmark Simula-tion Model No. 1 Long-term (BSM1_LT). In Chapter 6, models representing Henriksdal, Käppala and Himmerfjärden WWTPs are used to extend the method for the Benchmark model to models describing real-world processes.

Part III: Full-scale ammonium feedback control 1.4.3Implementation and evaluation of ammonium PI control in a full-scale WWTP – what are the challenges and benefits? Part III is devoted to full-scale ammonium feedback control. Part III constitutes one chapter (Chapter 8, paper IV, X, XI), which summarises the experiences gained over several years of working with full-scale ammonium PI controllers at Henriksdal, Käppala and Himmerfjärden WWTPs. Special attention is given to aspects concerning implementation and evaluation of controllers in full-scale opera-tion, including control systems, energy estimation and sensor quality. Statis-tical hypothesis testing and cost-benefit analyses are examples of evaluation methods in Part III.

Part IV: Improving ammonium feedback control 1.4.4Can ammonium PI control be improved through simple modifications? In Part IV, three attempts to improve the performance of a standard ammonium feedback controller are given: Gain scheduling control (Chapter 9, paper IX, XII), DO deviation control (Chapter 10) and repetitive control (Chapter 11, paper VI, VII). For the first two methods, the goal is to decrease energy con-sumption. Repetitive control on the other hand is a method for improved disturbance rejection of the incoming load. Part IV combines simulations with full-scale experiments.

Part V: Concluding remarks 1.4.5In the final part of the thesis, the conclusions from Part I to IV are summa-rised and refined and directions of future research are given.

1.5 Contributions 21

Contributions 1.5This thesis has:

• summarised the state-of-knowledge with regards to aeration con-trol (Chapter 3);

• provided insights into the design of cost-effective ammonium feedback controllers (Chapter 5 to 7);

• further increased the knowledge about how to implement and evaluate controllers in full-scale operation (Chapter 8);

• developed and tested gain scheduling ammonium feedback control in full-scale operation (Chapter 9);

• developed and tested DO deviation control in full-scale operation (Chapter 10) and

• developed and tested repetitive ammonium control in a simulation model (Chapter 11).

PART I

WASTEWATER TREATMENT AND AERATION CONTROL

25

2 WASTEWATER TREATMENT PLANTS – 2

TREATMENT, MODELLING AND CONTROL

HIS CHAPTER INTRODUCES wastewater treatment as a concept in general and biological nutrient removal in particular. It is also an introduction to the methods of this thesis – process modelling and

control – described in the context of wastewater treatment. For readers fa-miliar with either wastewater treatment or process modelling and control the chapter provide an introduction to expressions and terms within the respec-tive fields as well as a brief historical account. Municipal wastewater treat-ment and automatic control are relatively young research fields and emerged during the last century although both research fields has their origins in clas-sical areas such as mathematics, physics, biology and chemistry.

Why wastewater treatment 2.1The water and wastewater system is one of the backbone infrastructures of a modern society. The system includes water treatment and distribution and wastewater collection, treatment and disposal. For many people the system is symbolised by a water tap or a toilet seat. It is often a hidden infrastructure. In a society with well-functioning systems for water and sanitation, it is easy to overlook the importance of clean water and sanitation to the creation of a democratic and prosperous society. Still today, this is one of the key devel-opments that are missing in regions of the world where diseases and hygiene is still an obstacle to a person’s dignity, education and health.

T

26 2 Wastewater treatment plants – treatment, modelling and control

Brief historical outlook – the European perspective 2.1.1The Roman empire had realised the importance of sanitation and already in 500 BC the Cloaca Maxima disposed of wastewater from the city of Rome (Wiesmann et al., 2007). During the Medieval era there were examples of water distribution systems in castles and monasteries, but the cities in Eu-rope did not have a system to handle clean water or human waste products until the mid 1800’s. The increased population in the cities together with the many cholera outbreaks in Europe starting in the 1830’s – killing millions – increased the awareness of the issue of sanitation. This led to the develop-ment of a wastewater collection system, moving the problem from the cities to the outskirts of the urban areas.

In Sweden, basic, mechanical wastewater treatment was implemented in the 1930’s to manage the visual problems wastewater was causing in the receiving water bodies close to the cities (Lindberg, 1997). It is only later that wastewater treatment has been performed under the banner of environ-mental protection. Oxygen depletion in recipients in the 1950’s motivated biological treatment of wastewater. During the 1970’s an increased aware-ness of the negative effects eutrophication had on the receiving water bodies led to the combination of biological and chemical treatment. A reconstruc-tion period of the wastewater treatment plants occurred in Sweden in the 1990’s when many plants were rebuilt to accommodate nitrogen removal. Sweden is on the brim of a third period of reconstruction at the treatment plants when more stringent effluent limits should be met by the plants to reach new goals on the national and international environmental protection agenda. Also, concern is nowadays not only directed towards the classical pollutants carbon, nitrogen and phosphorous, but also priority pollutants such as pharmaceutical residues and chemicals in personal care products.

The production facility 2.1.2In cities, wastewater is collected from large urban areas and treatment is most often performed in a centralised plant. The focus of the treatment plant operation has shifted from treating wastewater to treating wastewater and producing valuable products as efficiently as possible. To generate products from wastewater is not the main motivation behind the operation of a treat-ment plant, but offers opportunities to optimise the overall wastewater sys-tem performance and to close cycles in society.

The treated wastewater is a product in itself and can be recycled and re-used for use on farmland, in the industry or as a source of drinking water (Levine and Asano, 2004). Depending on the use, additional treatment steps complementing the classical mechanical – biological – chemical are re-quired. Examples include removal of particulate matter through ultrafiltra-tion or reverse osmosis as well as methods for disinfection.

2.2 The wastewater treatment process 27

Wastewater contains energy in the form of organic matter. The carbon can be put to use in the treatment process (see Section 2.2.3) but it can also be used to produce energy in the form of biogas. Biogas can be used to fuel a gas motor at the plant or to produce vehicle fuel. Together with other munic-ipal or industrial organic sources, co-digestion can contribute to make treat-ment plants “energy neutral” or even energy producers.

More recent uses of organic matter in combination with microbial activity in wastewater treatment plants are to produce polymers to generate biode-gradable plastics (Satoh et al., 1999) and to use fuel cells to produce elec-tricity (Liu et al., 2004).

The increasing world-wide demand of fertilisers and the declining quality of and increased production costs of phosphorous rock have caused a de-mand for fertilisation products from sludge (Cordell et al., 2009) – one of the by-products from a wastewater treatment plant. There is an increased incen-tive to close the cycle of nutrients from farmland to the urban areas and back again. Phosphorous is not the only constituent in sludge that is valuable for a farmer. Sludge is also a valuable soil improver (Coker, 1983).

Capture of phosphorous in ash from incinerated sludge (Adam et al., 2009) and precipitation of struvite (MAP, magnesium aluminium phosphate) (Doyle and Parsons, 2002) are examples of attempts to create fertilising products from incinerated wastewater sludge.

The wastewater treatment process 2.2 Constituents in wastewater 2.2.1

Wastewater often has a negative feel to its name, representing unwanted material that is wasted from society. This is in some respects true and the wastewater collection system was created in order to protect us and the envi-ronment from harmful constituents in the sewage. But the harmful substanc-es are combined with the previously described resources our society is partly lacking or has a need to preserve, cycle or waste less of.

Wastewater constituents can be divided into physical, chemical and bio-logical constituents (Metcalf & Eddy et al., 2003). Particulate and soluble material containing nitrogen compounds, phosphorous compounds, organic compounds and pathogens are the main focus of a wastewater treatment process with nitrogen and phosphorous removal.

The larger share of nitrogen in incoming wastewater originates from urine and is present in the form of ammonium ions (NH4

+-N). Ammonium is in equilibrium with free ammonia (NH3-N). The distribution of the two species is pH dependent, with free ammonia dominating at high pH values (pK = 9.25). A part of the total nitrogen content is contained in organic molecules, referred to as organic nitrogen (org-N). Summing up the organic nitrogen


and ammonium nitrogen gives the Total Kjeldahl nitrogen (TKN = org-N + NH4

+-N). The total nitrogen (TN) in a sample is the sum of TKN and nitrates (NO3

--N) and nitrites (NO2--N). The convention in this thesis is to refer to

NO3--N as NO3 and NH4

+-N as NH4. TKN is a useful parameter since it gives the nitrogen available for nitrifi-

cation (see Section 2.2.3). There is little or no nitrate in the incoming wastewater. The different nitrogen species in wastewater have different oxi-dation states where ammonium is the most reduced form (-III) and nitrate is the most oxidised (+V). The nitrogen compounds undergo many conversions through nitrogen pathways in soil and water environments. The conversions of nitrogen in a wastewater treatment plant represent some of the pathways in the nitrogen cycle.

Phosphorous compounds are present in aqueous systems as ortophos-phates, polyphosphates and organic phosphate (Metcalf & Eddy et al., 2003). It is the ortophosphates that are available for microorganisms. There are several species of ortophosphate (PO4

3-, HPO44- etc.) and which species

is dominating is pH dependent. Organic material is often analysed as aggregate organic compounds.

Common methods include measuring the biochemical oxygen demand (BOD), chemical oxygen demand (COD) and total organic carbon (TOC). BOD is measured as the oxygen used by microorganisms during degradation of organic material. The oxidation process normally takes place during 5 days (BOD5). In Sweden the analysis is normally performed over a full week (BOD7). A BOD test can be inhibited by toxic substances.

The COD is also a measure of oxygen requirement. COD represents the oxygen equivalent required for chemical oxidation of the organic material using a chemical oxidant, e.g. dichromate. Due to the toxicity of chrome, some plants analyse TOC instead of COD. TOC is measured through con-verting the organic material to carbon dioxide using e.g. heat or chemical oxidants.

The three aggregate measures of organic compounds in wastewater covers different parts of the organics, and therefore TOC > COD > BOD. All organ-ic material is not available for biological degradation. BOD is a common measure to assess the biodegradability but is also the method with the largest uncertainty in the results.

Apart from nitrogen, phosphorous and organic material wastewater also contains pathogens (such as coliform bacteria and viruses), oil and grease and surfactants. During recent years, more and more attention has been put on priority pollutants. Priority pollutants include a vast group of chemicals for which the treatment plants were not constructed to remove, but which are prioritised due to the compounds’ toxicity to the environment. Priority pollu-tants can be organic or inorganic and include among many groups chlorinat-ed solvents, halo acetic acids and pesticides (Metcalf & Eddy et al., 2003). Several of the priority pollutants are known or expected carcinogens.

2.2 The wastewater treatment process 29

General process description 2.2.2Commonly, the first step in a WWTP is primary treatment which involves several mechanical treatment steps including screens, grit chamber and pri-mary clarification (Figure 2.1). Toilet paper and other large objects are re-moved in the screens, and in the grit chamber particles the size of sand are removed. After the water left the primary sedimentation, most particulate matter has been removed.

The biological treatment process removes organic material and nitrogen from the wastewater with the help of microorganisms. Biodegradable organ-ic material is oxidised by microorganisms and carbon dioxide is produced. Biological nitrogen removal is a two-step process involving nitrification and denitrification, see Section 2.2.3. The microorganisms are separated from the water in the secondary settler, the larger part is returned back to the biologi-cal process. After the secondary settler, there are optional polishing steps to add, for instance filters for additional removal of phosphorous (sand filters, disc filters) or disinfection steps (ultraviolet light, peroxide, ozone).

Sludge is separated in the primary and secondary sedimentation steps. The water content in the sludge is reduced through thickening, using gravita-tion, flotation or centrifuge techniques. The sludge is stabilised in a digester. The anaerobic fermentation process reduces the organic content in the sludge, producing biogas and digested sludge. The biogas contains 65 to 70 % methane and the rest of the gas is mostly carbon dioxide. Excessive gas is flared. Sludge can be used in many ways, where the most common alterna-tives in Sweden are to recycle the sludge back to farmland or to use it as landfill.

Figure 2.1. A schematic picture of a generic WWTP, including the water line and the sludge line. Several flows are not included in the figure, such as recycle of sludge liquor from dewatering to biological treatment and management of grit and screen residues.

Screens Grit chamberInlet pump Primary settler Biological treatment Secondary settler

Chemical dosage

Filters/desinfection

Sludge thickening Digester

Sludge dewatering

Raw gas

Stabilised and dewatered sludge

Treated wastewater

WastewaterSludgeGasChemicals


Biological nitrogen removal 2.2.3The core of the nitrogen removal process is the biological treatment where ammonium is oxidised via nitrate to nitrogen gas with the help of microor-ganisms. In the biological treatment one part of the process is aerated and another part has stirring to accommodate the two biological processes re-sponsible for nitrogen removal: nitrification (aerated) converting ammonium to nitrate and denitrification (non-aerated) converting nitrate to nitrogen gas.

Microorganisms need an energy source and a source of carbon for cell growth. Nitrification is a two-step process performed by two groups of bac-teria. Ammonium oxidation is the energy generating process. The carbon source for nitrifying organisms is carbon dioxide. The first group of micro-organisms (Nitrosomonas) oxidises ammonium to nitrite and the second group (Nitrobacter) nitrite to nitrate. Energy is made available when ammo-nium is oxidised and oxygen is reduced. Nitrifying bacteria are therefore aerobic – requiring oxygen – and autotrophic since the ammonium oxidisers can fix carbon from carbon dioxide.

The denitrification process is a series of processes where nitrate is re-duced to form nitrogen gas. Several groups of denitrifying bacteria take part in the transformation. The denitrifying organisms are heterotrophic, meaning they require organic carbon. Denitrification is a respiration process which only takes place when there is no oxygen available, since oxygen is a more favourable electron acceptor than nitrate from an energy point of view. When oxygen is not available, nitrate can be used in its stead for certain groups of bacteria. Conditions where oxygen is not available are called an-aerobic. An environment which is anaerobic but with nitrate available is referred to as an anoxic environment.

Microorganisms are present in the biological treatment step either grow-ing suspended in the liquid or attached to a growth media (plastics for in-stance). The most common process configuration for biological removal of organic matter and nitrogen is the activated sludge process which is a sus-pended growth process. The activated sludge process celebrates its 100th anniversary in 2014. The key characteristic of the process is the separation of the sludge retention time from the hydraulic retention time by returning sludge from the settler. There are several options when configuring the acti-vated sludge process for nitrogen removal. The two most common processes are predenitrification and post denitrification, an overview is given in Figure 2.2.

In a predenitrification process, the denitrifying anoxic process step is placed before nitrification. Wastewater entering the process will pass the denitrification step and the organic material in the wastewater will be used as a carbon source for denitrification. Ammonium is oxidised to nitrate in the final aerobic step, and nitrate rich water will be recirculated back to the deni-trification step.

2.3 Control of wastewater treatment plants 31

In a post denitrification process the configuration is more straightforward since aerobic nitrification is followed by denitrification. Most of the organic material will be oxidised to carbon dioxide by aerobic heterotrophic bacteria in the aeration process. Therefore, an external carbon source is needed in the denitrification process. There are also other process configurations such as simultaneous nitrification and denitrification, two-stage sludge systems and step-feed processes where the influent is distributed to several feed points along the basin.

Figure 2.2. A predenitrification and post denitrification process. RAS = return acti-vated sludge, WAS = waste activated sludge.

The sludge age (also referred to as solids retention time, SRT) is in an acti-vated sludge process the average time a sludge particle stays in the system before it is removed with the WAS flow. The sludge age is often calculated as the aerobic sludge age. The aerobic sludge age is an important process parameter for nitrification. Nitrifiers grow more slowly than denitrifiers and for too low sludge ages they will be washed out of the system.

The nitrification process is also easily disturbed by toxic inhibition, and the process requires an aerobic environment. Nitrobacter are more easily disturbed than Nitrosomonas, meaning there is a risk for nitrite accumula-tion. All bacteria, and especially nitrifiers, are sensitive to low temperatures. At low temperatures, a higher sludge age is required to have the same overall removal rate of nitrogen. Denitrification can be inhibited at high DO concen-trations and when access to organic carbon is limited.

Control of wastewater treatment plants 2.3 Brief historical outlook 2.3.1

The heart of automatic control is the feedback loop. “Feedback is an ancient idea, but feedback control is a young field” (Åström and Kumar, 2014). Since the 1940’s, control is applied in various parts of society – in electric devices, transport systems and in production processes. Automatic control emerged in the 1950’s as a separate research field and has been referred to as the hidden technology (Åström, 1999). Feedback is everywhere around us,


and is central to life itself but seldom something discussed by others than specialists.

Much development was taking place within communications and ana-logue computing in the middle of the previous century which was driven by the many benefits provided by automatic control (Åström and Kumar, 2014). At that time, much of the development was conducted by industry rather than academia. Due to the progress during the Second World War the re-search field was well established in the 1960’s – development that was driv-en by application and needs.

During the 1960’s many new developments were made in the era of the space race and the introduction of the computer and the period is referred to at the Golden Age of automatic control by Åström and Kumar (2014). The progress was not only driven by application, but a new theoretical frame-work was formed including the state space formulation and the Kalman fil-ter. Higher education expanded during this period.

The 21st century is wireless and with an ever increasing access to comput-er power. This has provided opportunities through more complex control systems and controllers and new challenges such as information safety. Au-tomatic control is being integrated into many new developments within other research fields, such as bioengineering.

The feedback loop 2.3.2The main task of the feedback loop in a control system is to keep the con-trolled variable close to its set-point. With perfect knowledge about the sys-tem this would not be a challenge, but the control goal should be met despite disturbances (process disturbances or measurement disturbances). The feed-back loop does that through calculating the value of the manipulated varia-ble. An example of a feedback loop is found in Figure 2.3.

When looking at the top box diagram in Figure 2.3 there are two signals which can be changed from outside the block diagram: the set-point and the disturbance. A change to either of these signals is bound to have an impact on the controlled variable if no change is made to the manipulated variable. The main control problems which the controller can be appointed to solve hence becomes: set-point tracking and disturbance rejection. There are many methods, such as feedback control, which can be used to solve either of these problems. The focus of the work presented in this thesis is with some exceptions disturbance rejection.

The controller in Figure 2.3 is nowadays digitally implemented in soft-ware and is seldom implemented in computers which are located adjacent to the process it is set to control. The actuator on the other hand is the physical unit in the feedback loop which executes the task given by the controller. Actuators can by electrical, pneumatic or hydraulic. At a treatment plant the


actuator is often a pump or a valve. Often the actuator is included within the process box in the feedback loop box diagram.

The sensor is what makes it possible to feedback a measurement of the controlled variable to the controller. The measurement is not perfect but can be noisy and is often to blame if a feedback loop fails. If the controlled vari-able cannot be measured a model can be used to estimate the value of the controlled variable by using a so called observer.

Figure 2.3. A feedback control loop. The actuator and the sensor are included in the process box, which is a common way of describing the loop. A control loop with feedback is referred to as a closed loop system. An example of open loop control is time control.

There are many ways to name the variables in the closed control loop in Figure 2.3. Examples are listed in Table 2.1, with the terminology used in this thesis in bold.

Table 2.1. List of variable names in a closed loop control system. The terminology used in this thesis in bold.

Variable Control science name Other names

Controller set-point Set-point (r) Reference valueControl error Control error (e) DeviationProcess input (output from controller) Input signal (u)

Manipulated variable, control signal

Process output Output signal (y) Controlled variable, process value (PV), measured value (MV)

The most widely used control algorithm in process control is the PID con-troller (Åström and Hägglund, 1995), introduced as a general purpose com-mercial controller in 1931 (Åström and Kumar, 2014). The PID controller consists of three parts:

++= dt

tdeTde

TteKtu d

t

i

)()(

1)()(

0

ττ

(2.1)


where u(t) is the manipulated variable, e(t) the control error, K the controller gain, Ti the integral time and Td the derivation time. The controller can either be used with all its parts or with only the P, PI or PD terms. The proportional part (P) is proportional to the present control error, the integral part (I) sums up previous control errors and the derivative part (D) predicts future control errors by using the derivative of the control error. The integral part provides what is referred to as integral action. Integral action leads to an elimination of steady-state offset. There is much to read about the PID controller in standard textbooks on automatic control.

There are generally speaking three ways to tune a PID controller: Manual tuning, tuning using tuning rules and auto tuning. Not to tune at all, i.e. keep the predefined settings in the controller as determined by the manufacturer, is not an option since all processes have its own dynamics. To tune the pro-cess manually requires good knowledge of the process at hand. It is also important that it is safe to make arbitrary adjustments in the controller set-tings without endangering process stability and performance.

For the PID controller tuning there are several well established tuning rules available, e.g. Ziegler-Nichols tuning, Cohen-Coon tuning and lambda tuning (Åström and Hägglund, 2006). Before a tuning rule can be applied, simple experiments need to be performed on the system, e.g. step response tests. Lambda tuning is a special case of pole-placement design and can be identical to Internal Model Control design (Åström and Hägglund, 2006). Pole-placement design is a common controller design method where the closed-loop poles are determined to achieve desired system properties.

A combination of step-response tests and construction of a process model is often the method behind auto tuning techniques implemented in industrial control systems. Both auto tuning and tuning with tuning rules can provide a first guess of the controller settings, and improvements can be made manual-ly if required.

The purpose of controller tuning is to design the controller in order to meet certain specifications. In a process where set-point tracking is the ob-jective, rise time, settling time and overshoot are commonly used specifica-tions for a controller (Åström and Hägglund, 1995). For controllers where the objective is to attenuate load disturbances, specifications can include integrals of the absolute or squared control errors. Other aspects to consider when tuning the controller are sensitivity to measurement noise and to changes in process dynamics.

Process control in WWTPs 2.3.3In the early 20th century there was a rapid increase in controllers applied in production processes within many industries (Åström and Kumar, 2014). By the use of valves and pumps together with newly developed sensors, control of pressure, temperature and flow was made possible. Today process control


is an indispensable tool in process industry, whether the process is found within pulp and paper, chemical, pharmaceutical or other industries. Com-mon variables to control are still today temperature and flow, but also liquid levels in tanks.

Wastewater treatment is a process industry subject to many disturbances. Like other process industries, a WWTP produces products (e.g. treated wastewater, sludge and biogas) out of a raw material (wastewater). Unlike most other process industries the treatment plants cannot control the raw material to the plant. Despite e.g. high daily influent variations and rain events the plants should consistently treat wastewater at all times.

The aim of using automation and process control in a wastewater treat-ment plant was defined in 1988 by the Swedish Water and Wastewater plant Association as (Svenska Vatten- och Avloppsverksföreningen (VAV), 1988):

• Keep the plant running and avoid large process disturbances; • minimise operational costs but keep the discharge limits and • further improve the effluent water quality.

25 years later the above list is still valid and summarises the combined im-portance of disturbance rejection, resource efficiency and removal results.

External disturbances to a WWTP can be classified into: (1) Seasonal var-iations, (2) diurnal variations and (3) event disturbances (Ingildsen, 2002). Ingildsen (2002) also mentions example of internal disturbances, which of-ten are event disturbances such as breakdown of equipment and backwash of filters. The three disturbance classes require different types of controllers, where seasonal variations can be handled with manual control or slow real-time controllers while diurnal variations need faster real-time controllers. Event disturbances need special attention and can either be handled manual-ly or with the help of controllers. Disturbances are traditionally managed through conservative safety margins in both process design and process con-trol at treatment plants, often at the expense of increased operational costs.

The booklet from1988 (Svenska Vatten- och Avloppsverksföreningen (VAV), 1988) lists many control loops which are still in operation today: flow proportional control of chemical dosing or return activated sludge, pressure control in air mains and cascade DO control. The most important controllers for the quality of the wastewater according to the booklet are DO control and control of the suspended solids concentration in the aeration tanks. Blower control was at that time on-off control either through open loop on-off control (time control) or through closed loop on-off control by feedback of the DO concentration.

Several more manipulated valuables have entered the wastewater treat-ment plant in recent years due to the introduction of nitrogen removal and new process configurations. Some examples include more zones to control (anaerobic, anoxic and aerobic), more sophisticated aeration systems, more


intermittent systems and external carbon addition (Olsson et al., 2014). An overview of actuators in the activated sludge process is given in Figure 2.4.

Figure 2.4. Actuators in the pre and post denitrification processes.

External carbon dosage, internal recycle and the aeration intensity can be adjusted to have a rapid effect on the nitrogen removal process. Apart from the aeration intensity which is governed by the air flow valve in systems with dispersed fine bubble aeration, the aeration volume can also be changed. The RAS and WAS flows are used to change the suspended solids concentration in the tanks, which has an impact on the sludge age. Compared to aeration control, sludge age control is slow and it can take several days up to weeks to change the sludge concentrations in a plant. There are several other control loops than those found in the activated sludge process at a treatment plant. To manage these loops with a unified approach is often re-ferred to as plant-wide control.

The systems in Figure 2.4 are multivariable since there are several manipulated variables and several controlled variables. In a multivariable system one manipulated variable may impact several controlled variables, which is called channel interaction (Halvarsson, 2010). Commonly a control structure for an activated sludge system is created by combining several parallel PID controllers instead of having one multivariable controller. This can be a sufficient solution if the interactions in the open loop system are small. A common name for a multivariable controller is MIMO controller (multiple input multiple output), compared to a SISO controller (single input single output).

A list of cross-couplings between the nitrification and denitrification in a predenitrification process is given by Ingildsen (2002) and includes transfer of oxygen from aerated to anoxic zones, carbon source meant for denitrification which is instead oxidised in the aerobic zones increasing the aeration requirements, simultaneous nitrification and denitrification and reduced availability of nitrate for denitrification if nitrification is disturbed. Performing a relative gain array (RGA) analysis on a simplified model of a predenitrification process, Ingildsen (2002) showed that the couplings between nitrification and denitrification are weak, hence the control loops can be decoupled. Investigated input variables were the DO set-point,

2.4 Introduction to aeration control 37

internal recirculation flow and carbon dosage. Similar results using RGA analysis were found in Machado et al. (2009). Using different interaction measures, Halvarsson et al. (2005) conclude that there are some couplings between nitrification and denitrification since the DO concentration can have an impact on the effluent nitrate concentration.

Another aspect which motivates decoupling of the control loops instead of multivariable control is the broad range of time constants in the process which can be grouped together (Jeppsson, 1996). Processes with slow dynamics (e.g. sludge age) can be controlled separately with slower controllers and fast dynamics (e.g. DO cocentration) can be controlled instantaneously. A full list of cause-effect relaltionships and their dynamics is given in Jeppsson (1996).

Introduction to aeration control 2.4The aeration process if often the single largest consumer of electricity at a plant, it can amount to 45 to 75 % of the total energy cost (Rosso et al., 2008). In Sweden nearly half of the electricity use at the municipal wastewater treatment plants is spent in the biological treatment process (Svenskt Vatten Utveckling, 2013). The larger part of this share is required for aeration. The cost of aeration together with the importance of aeration to sustain the biological treatment process in a WWTP motivates aeration con-trol.

Limitations in DO control 2.4.1The nitrifier growth rate depends on the DO concentration but the relation-ship is not linear. Already in 1965, scientists at the Stevenage site in the UK reported that DO concentrations above 2.0 mg/l had very limited effect on nitrifier growth rates, but there is a wide range of reported effects of DO on maximum nitrifier growth rates (Stenstrom and Poduska, 1980). More about process non-linearities is found in Section 2.4.5.

The DO concentration should not be viewed on its own without consider-ing temperature and aerobic SRT. At lower SRT and lower temperatures, higher DO concentrations might be required to balance a loss in nitrification rate. For processes with denitrification, elevated levels of DO can hamper denitrification performance if DO-rich water is recirculated to the anoxic zones.

Low DO concentrations have been associated with high emissions of ni-trous oxide (N2O) (Kampschreur et al., 2009). Nitrous oxide is a greenhouse gas with the very high global warming potential (GWP) of 310.

Some groups of filamentous microorganisms can compete with floc form-ing organisms during low DO concentrations (< 1.5 mg/l) which could affect


sludge settleability (Martins et al., 2004). This may set a lower bound on the acceptable DO level in an aeration basin.

Oxygen transfer 2.4.2Oxygen transfer is the process where gaseous oxygen dissolves into water. This occurs naturally by the water surface in a lake or in a wastewater treat-ment plant. But due to the low solubility of oxygen in water, additional aera-tion is required to sustain aerobic microbial activities in the aeration tanks.

The oxygen transfer coefficient, KLa is used to evaluate the efficiency of oxygen transfer. KLa is the overall mass transfer coefficient governing the speed of the mass transfer from gas to liquid, and is for a specified volume measured in units of one per time unit (d-1 or h-1). The rate of change of DO concentration in an aerated basin for wastewater treatment can be modelled as:

( ) ( )CCV

QrCCaK

dt

dCinMSL −+−−= (2.2)

KLa = Oxygen transfer coefficient (d-1) C = Liquid DO concentration (g/m3) Cs = DO saturation concentration (g/m3) rM = Respiration rate of microorganisms (g/d) Q = Inflow (m3/d) V = Volume of aeration tank (m3) Cin = Influent DO concentration (g/m3)

The diffusion of DO from gas phase to liquid phase will depend on the speed of transfer from gas to liquid, governed by KLa, and diffusion is slower the closer the DO concentration (C) is to the DO saturation concentration (CS). The concentration gradient will be decreased at higher DO concentrations.

KLa is affected by many physical and chemical factors in the aeration ba-sin. Surfactants (surface active agents) attach themselves to the air bubbles and decrease the oxygen transfer rate. Turbulence in the wastewater as well as tank geometry also has an impact and different aeration systems have different efficiency. The α factor is used to relate the oxygen transfer rate in process water to that in clean water:

cleanL

processL

aK

aK=α (2.3)

Air supply system 2.4.3Diffused aeration involves submerged diffusers placed at the bottom of the tanks. Blowers are used to compress air down through aeration pipes and


subsequently the air is diffused, commonly through rubber membranes. Me-chanical aerators stir up the water and therefore promote oxygen transfer. A surface aerator is a type of mechanical aerator.

For diffused aeration, adequate blower system design is required for effi-cient control of the DO concentration. During the design process it is im-portant to consider that the air flow demand varies over the day, week and year and there is also a variation along the aeration tank. The flexibility of a blower system is crucial for the performance of the aeration system since plants need to handle a large variation in load.

Historically, inlet vanes or outlet dampers have been used to meet a vary-ing demand but not in an energy efficient manner (WPCF & ASCE, 1988). Today, blowers supplied with variable frequency drives (VFD) allow turn-down of the aeration capacity. Centrifugal or positive displacement blowers are the two main types of blowers (Keskar, 2006). Centrifugal blowers – such as turbo blowers – can be controlled at a fixed air flow rate set-point by varying the blower capacity, while positive displacement blowers provide a constant flow independent of the system pressure. Positive displacement compressors with VFD have a nominal efficiency of 50 to 60 %, while cen-trifugal blowers have a higher efficiency (65 to 85 %) (Keskar, 2006). The positive displacement blowers are more common for small installations.

The air passes through a valve before it is diffused into the aeration basin. Butterfly valves, damper valves, globe valves and plug valves etc. have dif-ferent mechanical design. The flow dynamics of the valve describes the flow rate as a function of the valve position. The flow characteristics for a fixed pressure drop over the valve can either be linear (the flow is proportional to valve lift), equal percentage (the flow is proportional to the first derivative of the flow with respect to the valve lift) or quick opening (a small change in valve lift produces a large change in flow) (Seborg et al. 2010). An example of a damper valve with quick opening dynamics is illustrated in Figure 2.5. The valve has an actuator which is commonly pneumatic or electric, setting the valve opening depending on the control signal to the actuator (Keskar, 2006).


Figure 2.5. Example of non-linear valve characteristics with hysteresis (damper valve).

When the aeration control loops call for a certain air flow rate the actuator changes the valve position in the air grid which will cause a change in head-er pressure. Given constant pressure control the change in header pressure will be compensated for by changing the blower capacity or by using in-let/outlet throttling. Apart from blower capacity control, blower control also includes start-stop procedures for multiple blowers as well as safety proce-dures during start and stop and surge control which prevents instability at low flows by maintaining a minimum flow (Keskar, 2006).

One option to minimise the pressure loss over the air supply valves is to use the Most-Open-Valve principle (MOV) (Alex et al., 2002). MOV will vary the pressure in the air headers until the most open valve in the system will be nearly completely open. Another alternative is to relate the pressure set-point to the total air requirement. There is also the option of using Direct Flow control where pressure control is omitted and the blower capacity is adjusted to meet a total air flow requirement. MOV and Direct Flow control are examples of power minimising control strategies, not aeration control strategies.

There are several types of diffusers used in activated sludge basins. One way to categorise them is (1) porous or fine porous diffusers, (2) non-porous equipment and (3) other devices, including jet aerators (Metcalf & Eddy et al., 2003). Another way to categorise diffusers is by bubble size: coarse bub-bles and fine bubbles. It is recognized that coarse bubble aeration has a low-er oxygen transfer efficiency than fine bubble aeration (Groves et al., 1992). Mechanical aerators have values of α in the range of 0.4 to 0.8 while dif-fused aeration systems have values around 0.6 to 1.2 (Metcalf & Eddy et al., 2003).

The oxygen requirement along a plug-flow aeration tank will decrease as the concentrations of organic material and nitrogen decrease. To avoid un-necessary aeration the aeration intensity along the tank should be decreased to balance the requirement. During aeration system design, this is often han-dled by means of tapered aeration. Tapered aeration decreases the diffuser

30 40 50 60 70 80 90 10020

30

40

50

60

70

80

90

100

Valve position (%)

Air

flow

rat

e (%

of m

ax)


density along the tank, but does not satisfactorily compensate for load varia-tions. A more flexible way is to divide the tank into zones and control the air flow rate to the individual zones to be able to compensate for spatial and temporal load variations. An example of the ammonium and DO profiles in a plant with tapered aeration but without individual zone control is illustrated in Figure 2.6. When ammonium is removed the DO concentration rises to very high levels.

Figure 2.6. DO and ammonium profiles along an aeration tank with nitrification and no individual zone control, using one valve to the whole tank which is adjusted to control the DO in the second zone to 2 mg/l.

Sensors 2.4.4When measuring different properties in the activated sludge process there is a range of methods to use. Vanrolleghem and Lee (2003) present state-of-the art on-line measuring equipment in wastewater treatment plants. The authors list general parameters (e.g. suspended solids and temperature) and unit pro-cess specific sensors (e.g. DO and BOD measurements for the activated sludge process).

There are two main ways to measure DO: by electrochemical cells (gal-vanic or polarographic) or by using optical sensors with luminescent tech-niques (Keskar, 2006). Galvanic cells are the dominant electrochemical technology today. With luminescent techniques less maintenance is required compared to membrane sensors since there is no need for membrane clean-ing and maintenance. However the sensor cap needs regular replacement.

In the nutrient removal process there is an option to measure ammonium and nitrate with automated wet chemistry techniques, with in situ ion-selective electrodes (ISE) or with titrimetric sensors (Vanrolleghem and Lee, 2003). The ISE sensor has a faster response time than the other two methods since no sample pretreatment such as filtering is required.

A sensor has certain properties depending on the measurement technique and device such as measurement range, response time and accuracy. Differ-ent response times for sensors were determined in Rieger et al. (2003). Sen-

1 2 3 4 5 60

2

4

6

8

10

Zone number

Con

cent

ratio

n (m

g/l)

DO low load

NH4 low load

DO high load

NH4 high load


sors requiring filtration were estimated to have a total response time of 10 to 30 minutes depending on the speed of filtration, while the ISE sensors and optical sensors without filtration were modelled with a 1 minute response time.

Measurements are often prone to be noisy which can hinder the perfor-mance of a controller. Noisy signals should therefore be filtered. Fast sam-pling may allow for averaging or more sophisticated filtering such as expo-nential filters. Many sensors have implemented an internal filter.

Compared to using a sensor purely for monitoring, a sensor in a control loop can seriously hamper the control performance if the signal is faulty. Rosen et al. (2008) discuss different types of sensor faults in their attempt to model sensor and actuator behaviour. The list of sensor faults include normal noise according to specifications, excessive drift, shift (off-set), fixed value, complete failure (no signal), wrong gain and erroneous calibration. Thomann et al. (2002) present a monitoring concept for on-line sensors having drift, a shift or outlier problems. A detailed method on how to quantify sensor un-certainty is found in Rieger et al. (2005).

Sensor maintenance is an important factor to achieve good performance, and the cost for maintenance work should be included in a cost-benefit anal-ysis when a new control strategy is considered. In situ ISE sensors have been reported to require around 2 hours of maintenance per week and sensor (Kaelin et al., 2008), to mainly take care of dirt around the electrode, par-ticularly chemical film formation on the membrane. The sensor location will impact the need for maintenance. Influent waste streams constitute a more hostile environment for in situ measurements than secondary settler efflu-ents.

The sensor location will also impact what information is available for the controller. There are often large time delays in treatment plants. Time delays are – in a feedback system – not easily managed by a controller. If ammoni-um is to be measured for aeration control the sensor can be placed in the aeration basin in situ. In a plug flow system and especially at larger plants, a sensor placed in the last aerated zone will be delayed in relation to the con-centration in the first aerated zone. An option would be to place the sensor in the middle of the aeration tank providing feedback with respect to the first aerated zones and feedforward action with respect to the last aerated zone.

In theory, plants with a common sludge return could be expected to need less instrumentation than plants with separate sludge return for individual treatment trains. In reality there is often an individual behaviour in treatment lines with a common sludge return due to influent variations and different age of equipment etc.

Since hardware sensors or measurements can be difficult or expensive to handle, there has been a development towards using software sensors (soft sensors) where models are used together with simple measurements in order to calculate a variable that may be more complex or costly to measure direct-


ly. The soft sensor has to be calibrated and validated based on measurement data, typically from a dedicated measurement campaign. During operation the soft sensor relies on information from other hardware sensors. The soft sensor can be used as a “shadowing” sensor to be able to provide infor-mation about estimated sensor faults (Lumley, 2002). As an ordinary sensor, the model in the soft sensors needs to be calibrated at regular intervals to maintain its prediction capability.

Process dynamics 2.4.5The key manipulated variable to the aeration process is either the valve posi-tion – for diffused bottom aeration – or the power input – for surface aera-tion. Several steps are taking place before the actual nitrogen concentration is influenced (Figure 2.7).

Figure 2.7. The multistep process from valve opening to effluent ammonium con-centration in a Bold: manipulated variable. Dashed: variables available for on-line measurements with standard sensors. The steps in the process are considered to be non-linear as schematically depicted in the figure with brief explanations. The equa-tion describes the Monod functions for DO and ammonium. KLa is the oxygen trans-fer rate. KNH and KDO are half-saturation constants.

If the valve is non-linear, the system is non-linear in each of the steps in Figure 2.7. The non-linearities are smooth and monotonically increasing. This makes them readily manageable in control. The origins of the non-linearities are:

• Non-linear valve characteristics, as described above; • decreased oxygen transfer efficiency at higher air flow rates due to

aggregation of bubbles, decreasing the total transfer area towards the water phase as well as increasing bubble rising times;

• saturation of the DO concentration (see Eq. (2.2)) and • growth rate dynamics of nitrifiers. At lower DO concentration the

an increase in DO implies an increase in nitrification rate, while at higher DO concentrations an increase in DO has a limited effect on the growth rate (see further Section 2.5.1).

μmax

Valve(%)

Air flow rate

Air flow rate

KLa

KLa

DO

DO

Growth rate (μ)

Valve characteristics Bubble dynamics Saturation of DO Growth dynamics (Monod)

KDO


The response time in each of the steps in Figure 2.7 increases along the ar-row from a few seconds (change in valve position) to hours (change in efflu-ent concentrations). The aeration system has been modelled to have a re-sponse time of approximately 4 to 5 minutes (including control loops but excluding DO control, rise time of bubbles and delays in air supply system) (Rieger et al., 2006). The response time of the aeration system including DO dynamics is in the order of 30 minutes.

Modelling and simulation of wastewater treatment 2.5plants

There are several benefits of modelling of wastewater treatment plant pro-cesses. Jeppsson (1996) mentions plant design, testing hypotheses in re-search, development and testing of control strategies, forecasting, analysis of total plant performance, and education as general purposes for using models in the field of wastewater treatment. For a history of activated sludge model-ling, see Olsson et al. (2014).

Activated Sludge Model No. 1 2.5.1The International Water Association (IWA) Task Group on Activated Sludge Modelling was initiated in 1982. In 1987, the model that is today considered state-of-the-art for dynamic modelling of activated sludge processes was published as Activated Sludge Model No. 1 (ASM1, Henze et al., 1987). ASM1 is a model with 13 state variables, 19 model parameters and eight process equations (ordinary differential). The COD and nitrogen components in ASM1 are presented in Figure 2.8 and Figure 2.9 and the processes are listed in Table 2.2. Apart from COD and nitrogen state variables, ASM1 also includes dissolved oxygen (SO2) and alkalinity (SALK). The full list of model parameters and their descriptions are given in Appendix A.

There are further developments of the ASM1 model into the ASM2d and ASM3 models (Henze et al., 2000). ASM2d includes biological phosphorous removal. The main difference between ASM3 and ASM1 is that ASM3 models storage polymers in heterotrophic COD removal. The ASM models are COD-based, meaning concentrations (carbon, oxygen) are given in units of gCOD/m3.

2.5 Modelling and simulation of wastewater treatment plants 45

Figure 2.8. COD components in ASM1. From Jeppsson (1996). Notation according to Corominas et al. (2010), previously used nomenclature in parenthesis.

Figure 2.9. Nitrogen components in ASM1. From Jeppsson (1996). Notation accord-ing to Corominas et al. (2010), previously used nomenclature in parenthesis.

Total COD

Biodegradable COD

Nonbiodegr. COD

Active biomass COD

HeterotrophsXOHO (XB,H)

AutotrophsXANO (XB,A)

SolubleSB (SS)

ParticulateXCB (XS)

SolubleSU (SI)

ParticulateXU (XI)

Total Kjeldahl NTKN

Free, saline ammoniaSNHx (SNH)

Organically bound N

N in active biomass

Soluble organic N

Particualte organic N

Nonbiodeg. NNonbiodeg. N

Nitrate, nitrite NSNOx (SNO)

Biodeg. N

SB,N XCB,N

(SND) (XND)


Table 2.2. Processes in ASM1. Parameters are explained in Appendix A.

No. Process Components Parameters

1 Aerobic growth of heterotrophic biomass XOHO, SB, SO2 µOHO, Max, KSB,OHO,

KO2,OHO

2 Anoxic growth of heterotrophic biomass XOHO, SB, SO2, SNOx ηµOHO, Ax, µOHO, Max, KSB,OHO, KO2,OHO,

KNOx,OHO

3 Aerobic growth of autotrophic biomass XANO, SNHx, SO2 µANO, Max, KNHx,ANO, KO2,ANO

4 Decay of heterotrophic biomass XOHO bOHO

5 Decay of autotrophic biomass XANO bANO

6 Ammonification of soluble organic nitrogen SB,N qa

7 Hydrolysis of entrapped organisms XCB, XOHO, SO2, SNOx qXCB_SB,hyd, KXCB,hyd, KO2,OHO, ηµANO, Ax, KNOx,OHO

8 Hydrolysis of entrapped organic nitrogen XCB, XOHO, SO2, SNOx, SB,N, XCB,N

qXCB_SB,hyd, KXCB,hyd, KO2,OHO, ηµANO, Ax, KNOx,OHO

By combining the eight processes in Table 2.2 the full set of ordinary differ-ential equations in ASM1 is created. As an example, heterotrophic growth is described by combining the three processes aerobic and anoxic heterotrophic growth and heterotrophic decay:

dt

dX OHO +++

= OHO

OOHOO

O

BOHOSB

BMAxOHO X

SK

S

SK

S

2,2

2

,

,μ

(2.4) −+++

+ OHO

NOxOHONOx

NOx

OOHOO

OHOO

BOHOSB

BMaxOHOAxOHO X

SK

S

SK

K

SK

S

,2,2

,2

,

,, μημ

OHOOHO Xb−

The factor SK

S

sMax +

μ is an example of a Monod equation (Monod, 1942).

This equation is a non-linear function describing biomass growth limited by access to a substrate. In the case of heterotrophic growth, the growth rate is limited by the availability of biodegradable carbon. The last part of the Monod equation (S/ (KS + S)) acts like a continuous switching function. The switch is 50 % on (the function has a value of 0.5) when the substrate con-centration equals the half-saturation constant (KSB,OHO).

The Monod function is used in many of the model equations in ASM1. Aerobic growth of autotrophs is described by the two processes for aerobic growth and decay and contains two Monod:

ANOANOANO

OANOO

O

NHxANONHx

NHxMaxANO

ANO XbXSK

S

SK

S

dt

dX −++

=2,2

2

,

,μ (2.5)

2.5 Modelling and simulation of wastewater treatment plants 47

Due to the DO Monod function a change in DO concentrations will not have a large impact on the growth of autotrophs at high DO concentrations. More about process non-linearities is found in Section 2.4.5.

The benchmark simulation models 2.5.2A commonly used tool for evaluating control strategies in a model is the IWA/COST Simulation Benchmark Model No. 1 (BSM1) (Copp, 2002), developed to provide a unified framework for control strategy evaluation. The benchmark procedure defines the plant layout, a simulation model, the influent to the plant model, test procedures and evaluation criteria. BSM1 represent a predenitrification plant with two anoxic and three aerobic zones. There is the possibility to control the nitrate recycle through feedback of the nitrate concentration in the last anoxic zone and to control KLa in the last aerated zone with feedback of the DO concentration. There are three defined influent files in the BSM1 setup: dry weather file, rain weather file and storm weather file.

ASM1 is the bioreactor model in BSM1, and the settler model is a 10 lay-er Takács model (Takács et al., 1991). There are several sensor classifica-tions with response times up to 30 minutes. Most control handles in the model are assumed to be ideal, but KLa has a response time of 4 minutes to reflect the delay generated by the aeration system.

The BSM concept is continuously being developed. One important exten-sion to BSM1 is BSM2 (Nopens et al. 2010), covering not only the activated sludge process but also primary treatment and sludge handling (Jeppsson et al., 2006). Jeppsson and Vanrolleghem (2014) present publications related to the Benchmark Simulation Model. The BSM models are available in several simulation platforms (Copp, 2002).

49

3 AERATION CONTROL – A REVIEW 3

HIS REVIEW COVERS automatic control of continuous aeration systems in municipal wastewater treatment plants. The review focus-es on published research from the years 2000 to 2014 and describes

research into various methods to decide and control the DO concentration and to control the aerobic volume. Special focus is put on plants with nitro-gen removal, and alternating or intermittent aeration systems, sequencing batch reactors (SBRs) and industrial applications are only mentioned briefly when applicable. Important aspects of control system implementation are discussed, together with a critical review of published research on the topic. The review is supplemented with a summary of comparisons between con-trol strategies evaluated in full-scale, pilot-scale and in simulations.

Introduction 3.1Earlier published material on the topic include the annual literature reviews published by Water Environment Research (e.g. Sweeney and Kabouris, 2011), and text books like Olsson and Newell (1999) and Olsson et al. (2005) where different aspects of ICA (instrumentation, control and automa-tion) within the wastewater and water industries are presented. Weijers (2000) has documented a detailed list of control laws for wastewater treat-ment control up to then, including aeration control. Another overview of different control systems is found in Vanrolleghem (2001). Jeppsson et al. (2002) provide an overview of ICA from a European perspective and con-clude that PI control or variations thereof were the most common strategies in full-scale at the turn of the last century.

T

50 3 Aeration control – a review

Control structures 3.2Control structure design is about how to set up the control system, namely which variables to control, which variables to manipulate and how to com-bine these two sets of variables to create control loops. Depending on the process at hand different types of controller structures can be considered in a process control scheme.

There is no unique way to categorise control structures for aeration con-trol. In this chapter we have chosen the following four categories:

A. DO cascade control B. Ammonium-based supervisory control

B1. Feedback control B2. Feedforward-feedback control

C. Advanced SISO and MIMO controllers D. Control of the aerobic volume

The motivation behind this categorisation is that each of the levels requires a different level of complexity (programming, sensors) in the control system. The block diagrams for each of the strategies are presented in Figure 3.1. The controllers in A and B are SISO controllers. Advanced controllers have often a MIMO structure, but they can also be SISO. To be precise, advanced SISO and MIMO controllers (structure C) are not necessarily part of a unique control structure. An advanced controller can be a part of control structure A or B. But in C, we consider control strategies that are typically not included in a basic course in automatic control.

3.2 Control structures 51

Figure 3.1. Categorisation of controller structures for aeration control. SP = set-point, C = controller, FF = feedforward controller. For control structure C, the ex-ample is a model-based controller with constraints and cost-function controlling the DO set-point. Control structure A is included in structure B to D.


A. DO cascade control 3.2.1There are several levels of control in an aeration control structure. The in-nermost controllers are generally simple feedback controllers, often set up in cascade. Air flow rate control and DO control benefit from cascade control, since it is a non-linear process with increasing response times and an inter-mediary measured variable. The intermediary variable (air flow rate) is measured and controlled in an inner (slave) loop, and the outer (master) loop controls the controlled variable (DO). The benefit of a cascade control sys-tem is that non-linear dynamics of the elements in the slave control loop can be compensated for by the slave controller, meaning the master controller “can see” a more ideal behaviour which simplifies controller tuning.

B. Ammonium-based supervisory control 3.2.2The DO set-point in the DO cascade controller is decided by the operator. To improve the control performance an externally calculated set-point can be used. The DO set-point can be calculated based on the measured ammonium concentrations in the outlet of the activated sludge process or from an in situ sensor (structure B1). This is nothing else but a triple cascade controller.

Another way to calculate a supervisory set-point is by feedforward control for improved disturbance rejection (structure B2, including feedback ammo-nium control). Commonly, the key disturbances are the influent ammonium concentration and influent flow rate to the plant. Feedforward control has the ability to react faster to a disturbance, since it will predict the impact of the disturbance before it affects the process, by using a feedforward model. The accuracy of the prediction will depend on the model quality. A perfect pre-diction never occurs, which is why feedback control should be added to feedforward control in order to make the final correction based on the true measurement.

C. Advanced SISO and MIMO control 3.2.3With advanced control we here refer to different model-based and optimal controllers. Model-based controllers include a large group of control algo-rithms which all make use of a process model in the control law. The model can be either black-box or be based on “physical” process equations. Often the model can be used to find a controller output that is optimal in some sense. Optimal control, as defined here, assumes a cost function to be math-ematically minimised and attempts to find the best solution to the minimisa-tion problem given constraints on the system.

3.3 Control algorithms 53

D. Control of aerobic volume 3.2.4Additional to adjusting the aeration intensity, parts of the aerobic volume can be switched on and off. The control is often feedback or feedforward, as can be seen in Figure 3.1, and the output of the controller is a decision on whether or not a zone should be aerated. Therefore the controller commonly needs a defined rule on whether or not aeration should operate.

Control algorithms 3.3The task of a controller is to keep the process value at the set-point. The most common controller to fulfil this task in process control is the PID con-troller, as mentioned in Section 2.3.2. A large group of controllers can be joined under the name rule-based con-trol. The most simple form of a rule-based controller includes if...then rules to determine for instance set-points of DO based on a feedforward or feed-back signal. Control of the aerobic volume is commonly based on rule-based control.

Rule-based control in the form of fuzzy logic control (FLC) was tradition-ally applied to alternating systems and batch reactors; see for instance Traoré et al. (2005) and Fiter et al. (2005) The trend has been to expand the applica-tion towards continuous operation. Historically, fuzzy logic is an extension of Boolean logic where not only 0 and 1 are considered as alternatives but also the continuous interval in between. Membership functions are used to “fuzzify” the controller and apply rules. At the end the fuzzy controller has to be “defuzzified” and the end product is a nonlinear controller. Fuzzy logic control is appreciated for its transparency and the possibility to include pro-cess knowledge (such as operator experience) in the controller.

There are many types of model-based controllers used in advanced con-trol strategies. Both feedback model-based control, such as Linear Quadratic Control (LQC), and predictive control, such as Model Predictive Control (MPC), minimise a cost function. MPC has become popular within many industries for its ability to handle constraints and to include multiple varia-bles. MPC has been a research topic for WWTPs since the mid 1990’s. There are many developments of the classical MPC method, such as robust MPC, adaptive MPC and non-linear MPC, see Weijers (2000).

Control of aeration intensity 3.4 A. DO cascade control 3.4.1

DO control has been common practice in process control in WWTPs for many decades, and DO control was first implemented more than 40 years


ago. Ingildsen (2002) reproduce a table originally published in Andersson (1979) with results on energy savings at seven Danish WWTPs in the 1970’s due to implementation of DO control. The total power savings ranged from 2.5 to 60 % with an average of 27 %, emphasising that the gain achieved from implementation of a new control strategy very much depends on the situation before the upgrade.

An example of a full-scale evaluation of individual zone control of DO at the Käppala WWTP (Stockholm, Sweden) is reported in Thunberg et al. (2009). The goal was to distribute the air according to the oxygen demand along the length of the bioreactor, and to avoid high air flow rates in the beginning of the basin. The original control strategy is based on a linear air flow distribution and makes use of two DO sensors: one in the beginning and the other at the end of the four aerated zones. The first sensor decides the total air flow to the reactor and the second sensor the slope of the step-like air flow profile. With individual control of each zone a saving of 26 % of air flow rate was achieved over a one year period.

Air flow distribution was also investigated by Sahlmann et al. (2004) where four different DO set-point combinations were compared for three aerobic zones in an A2O (anaerobic-anoxic-oxic) process. The zone distribu-tion of air flow and standardised oxygen transfer efficiency (αSOTE, meas-ured by off gas method) were analysed for different loads. An air flow rate saving of 15 % was achieved by using a DO profile of 1.2/1.2/1.5 mgO2/l compared to 2/2/2 mg O2/l. There is no information on variations in nitrogen removal performance, other than that the effluent concentrations met the discharge criteria.

Classical PID control has been investigated and developed further. As an example, Tzoneva (2007) evaluated two standard PID tuning methods (Zieg-ler-Nichols and a relay tuning method) using the BSM1. The paper presents a method on how to perform real time tuning for the purpose of adaptive control. Adaptive controllers use rules for updating controller parameters with the purpose to adjust to changes in process dynamics or disturbances.

Gerkšič et al. (2006) evaluate gain scheduling of DO PI control in BSM1 and in a pilot-plant MBBR (moving bed biofilm reactor) based on a model-based estimation of the respiration rate. Gain scheduling change the control-ler parameters depending on the value of a scheduling variable, in this case the estimated respiration rate. The goal was to compensate for process non-linearities. The pilot-plant results show a slight improvement in DO control performance.

Another example of adjusting controller parameters is presented by Han et al. (2008) who simulate a fuzzy DO controller. The simulation study em-ploys a piecewise linearised relationship between the air flow rate and the DO. The PI parameter values are tuned for each linear section. A blending of the output of the PI controllers is performed based on Gaussian membership

3.4 Control of aeration intensity 55

functions. The result is a controller acting similar to gain scheduling which changes the controller parameters based on the DO set-point.

A recent example of a full-scale implementation of DO cascade control is found in Lazić et al. (2012). A small treatment plant was rebuilt with new aeration equipment, new blowers and a new control system with ammonium feedback control. Of a total energy saving of 65 %, 8 percentage points could be explained by the improved control system, which included DO cascade control and most-open-valve logic. The reference lane had DO con-trol without an air flow controller as well as older equipment.

Many DO cascade controllers have been investigated in full-scale and in simulation studies during the 21st century. Most often they are used as a ref-erence strategy when evaluating more advanced controllers as can be seen in the following sections. The case studies presented in this section are com-pared in Table 3.1.

B. Ammonium-based supervisory control – simulation 3.4.2studies

In Krause et al. (2002) the determination of the DO set-point using feedfor-ward-feedback control is combined with rule-based control. The feedforward controller compares the nitrification load to the nitrification capacity and use rules to switch aerated compartments on and off and to step the DO set-point up and down. The set-points are compared with the set-points of a feedback controller which base the set-point on measured ammonium in the aeration tank outlet. The feedforward controller excels at reducing ammonium peaks due to an early increase of aeration during peak load. This is important, par-ticularly in Germany since the German effluent standards never allow the plant to exceed the effluent limits in grab samples (15 minutes composite sample).

Rosen (2001) describes challenges involved in monitoring and control of wastewater treatment operation and outlines the possibilities for multivariate monitoring and control. With respect to control of aeration systems, the the-sis covers set-point adjustments based on clustering to make the process return to its preferred process state and a multivariate feedback controller that calculates appropriate set-points for lower level controllers. The tech-niques are applied to DO set-point control by simulations of a reduced order ASM1.

Serralta et al. (2002) present simulations where DO in the last aerobic tank and nitrate are controlled with supervisory and fuzzy control in a model of a Bardenpho process using ASM2. Pressure control in air mains, air flow control, DO control, ammonium control and nitrate control by control of internal recycling all had fuzzy controllers.


A rule-based feedforward methodology was developed by Shen et al. (2010) for an A2O process, based on the cumulative frequency distribution of the influent ammonium and the C/N (carbon/nitrogen) ratio. Optimal set-points for the feedforward rules were created based on steady state simula-tions and validated in dynamic simulations. An 8.5 % reduction in air flow rate was found for similar total nitrogen concentrations in the effluent.

In Murphy et al. (2009), DO set-points are determined with a feedforward rule-based approach depending on the influent ammonium load. The DO set-points (between 1 and 2.5 mg/l) were calculated for the four aerated zones at the Mangere WWTP in New Zealand using steady-state Monte-Carlo simu-lations searching for the lowest DO profile for a chosen ammonium load which satisfies the DO limitation and the discharge criteria. Other data such as temperature are not included in the calculations.

Using a hierarchical control structure, Machado et al. (2009) compute the most economical set-point for a number of decentralised controllers, control-ling for instance the ammonium in the effluent of an A2O process. The top-most controller in the hierarchy was a so called Cost controller, constituting three PI controllers (representing ammonium, nitrate and phosphorous set-point manipulation) with the total operating cost as the controlled variable. The three PI controllers were designed based on first-order-plus-deadtime (FODT) models created from step-response tests. The cost controller is not really control structure B (supervisory DO control); it is rather a supervisory ammonium controller.

The energy consumption for different controller settings in an ammonium feedback controller was investigated in Åmand and Carlsson (2012a). Simu-lations were performed in a BSM1 model with one aerobic compartment and with only daily influent variations. Different ammonium controllers were compared to an optimised KLa vector, where all simulations reach the same daily ammonium concentration in the effluent. The optimal solution is 1-4 % more efficient in terms of KLa than constant DO control depending on the variation in load, and close to optimal performance can be reached with su-pervisory ammonium control. Further developments of this approach is pre-sented in Åmand and Carlsson (2013a).

In Rieger et al. (2012b), the authors review important aspects of ammoni-um control in general and ammonium feedforward in particular, including disadvantages of feedforward controllers and selection criteria. The authors discuss the limitations to ammonium removal created by the mass of nitrifi-ers in the system. The limitation of the nitrifier mass cannot be compensated for by an increased aeration above DO concentrations of about 2 mg/l, since the average mass is based on the average ammonium load removed. To miti-gate the effect of high nitrogen load when DO is high, the options remaining are increased aerobic volume (swing zones) or looking at load buffering.

Ammonium feedforward is not always beneficial, as exemplified by Rieger et al. (2012a). Using simulations for control system design of the


Nansemond treatment plant, the authors test different ammonium feedback and feedforward controllers, including continuous and discrete (on-off) de-termination of the DO set-point. See Table 3.1 for savings at different water temperatures. Continuous ammonium feedback was considered the prefera-ble option compared to discrete feedback mainly due to less wear and tear on the equipment. Feedforward control (comparing incoming ammonium load to nitrification capacity) was evaluated compared to feedback ammonium control for dry weather flow and for an ammonium peak. The impact of the feedforward controller was limited since it lost its control authority quickly at DO concentrations above 2 mg/l.

B. Ammonium-based supervisory control – full-scale and 3.4.3pilot-scale case studies

Rule-based feedback control of outlet ammonium was evaluated in full-scale by Husmann et al. (1998) during warm and cold temperatures. DO concen-trations were changed in steps and a facultative aerated zone was controlled. The controller managed to reduce effluent ammonium and nitrate concentra-tions with an aeration energy reduction of 16%, and the controller main-tained the ammonium concentration below the effluent permit.

The control of the DO set-point was also reported in a study by Suescun et al. (2001). The DO set-point was adjusted every four hours to compensate for the deviation in actual effluent ammonium compared to the ammonium set-point. The DO was controlled with conventional feedback. There are facultative zones which can be aerobic/anoxic depending on operational aspects. The simulations demonstrated an air flow rate reduction of 11 %. The controller was eventually combined with a similar control loop for sus-pended solids and verified in full-scale at the Galindo-Bilbao WWTP (Ayesa et al., 2006; Galarza et al., 2001).

Meyer and Pöpel (2003) performed simulations and pilot plant testing of a predenitrification system controlling the DO set-point and the ratio of aero-bic and anoxic zones using a fuzzy controller. The system combined feed-forward of influent ammonium with feedback of the outlet ammonium and nitrate concentrations as well as the outlet ammonium time variation. Com-pared to a fixed set-point of DO with relay control alternating between 0 (meaning prolonged denitrification) and 2 mg/l with constant zone division, the fuzzy controller air flow was decreased by 24%.

A similar approach to control was made by Yong et al. (2006), where, apart from the DO set-points, the external carbon dosage was controlled. Inlet and outlet ammonium concentrations were used as inputs to the control-ler which was tested in simulations and in pilot-scale testing. Pilot-plant testing showed an increased removal of ammonium of 16 % and an air flow rate decrease of 10 %. No graphs were shown of pilot-plant performance.


In full-scale experiments, Ingildsen (2002) concluded that in situ nutrient measurements in combination with simple control strategies can improve the plant performance significantly. An important advantage of the in situ in-struments was the fact that different sensor locations could readily be tested and compared. Different controllers based on feedback of ammonium and feedforward of ammonium load were tested, see Table 3.1. The best perfor-mance was achieved by a feedback ammonium controller.

In Vrečko et al. (2006), the air flow per kg of ammonium removed was reduced by 45% with a combined feedforward and ammonium cascade con-trol method, compared to PI control with a constant DO set-point. Using only ammonium feedback control reduced the air flow per kg of ammonium removed by 23%. The results came from a pilot-plant MBBR experiment. The oxygen set-point in the reference case was higher than the average DO concentration during the evaluation of the ammonium-based controllers, nevertheless the effluent ammonium concentration was substantially higher during constant DO control (see Table 3.1).

Baroni et al. (2006) present a full-scale implementation of a fuzzy logic system in a predenitrification system. The DO set-point and the air supply were controlled through fuzzy logic. The installations were running for ap-proximately a year and produced long term and short term process stability as well as energy savings.

Two examples of full-scale implementations of the Bioprocess Intelligent Optimization System (BIOS) were presented in Liu et al. (2005) and Walz et al. (2009). The BIOS uses feedforward control to update the DO set-point and internal recycle flow in a predenitrification plant based on on-line influ-ent measurements and process data. In Liu et al. (2005), energy savings of around 19 % were achieved with improved nitrogen removal, while Walz et al. (2009) demonstrated a 15 % energy reduction with maintained nitrogen removal. It is not specified in Liu et al. (2005) whether the total nitrogen reduction from implementing BIOS originate from improved aeration con-trol or from improved control of the internal recycle.

Yoo and Kim (2009) performed full-scale testing where different autotun-ing methods for PID DO control were evaluated in an industrial WWTP. Together with estimation of the oxygen transfer function (KLa) and respira-tion rate (R) proposed by Lindberg (1997) and a DO set-point decision law based on the estimated R, the study demonstrated more stable treatment re-sults for COD (chemical oxygen demand) and an energy saving potential of 5 %. The control structure is not identical to that depicted in Figure 3.1 (B2), since it is a supervisory DO controller without ammonium measurement.

Slow ammonium feedback control together with feedback of DO in the last aerated zone to quickly counteract oxygen peaks was evaluated for 1 week in full-scale in Thunberg et al. (2009). Compared to the air flow distri-bution method with two DO sensors described before, the strategy saved 18 % of air flow rate for similar treatment performance. The larger part of the


energy saving can be explained with the introduction of individual zone con-trol of DO.

Thornton et al. (2010) investigated a feedforward controller in a full-scale plant in the south of UK. The feedforward model used information about ammonium in the first anoxic zone, flow into the first aerobic zone, water temperature, settled sewage suspended solids and COD concentrations and the ammonium set-point. The model was based on the ASM1 model and the controller caused the DO in zone 1 to vary between the minimum allowed level of 0.75 mg/l up to about 1.2 to 1.4 mg/l. A reduction in air flow rate of about 20 % along with an increase in effluent ammonium was achieved compared to fixed DO set-points of 0.5 to 2.1 mg/l depending on zone.

Extensive evaluations of different rule-based feedforward and feedback controllers is presented in Rieger et al. (2012c), simulating and performing full-scale testing at three Swiss WWTPs. The authors use rules to determine switching points for the DO set-points. Energy savings and improved total nitrogen removal are achieved through reducing aeration thereby improving denitrification and at the same time allowing for operation closer to the am-monium set-point. The new controllers are compared to the full-scale base cases with constant DO control, sometimes limited by i.e. blower con-straints. Energy savings in full-scale operation amount to up to 20 %, com-pared to fixed DO set-point control, see Table 3.1.

Full-scale trials with ammonium feedback control performed at Henriks-dal, Käppala and Himmerfjärden WWTPs are described in detail in Chapter 8 in this thesis and the results are published in Åmand et al. (2014).

C. Advanced control – simulations 3.4.4Steffens and Lant (1999) describe simulation results based on different mod-el-based controllers in comparison with classical PI control with fixed or variable DO set-points using ammonium feedback control. The DO set-point, internal recycle flow rate, return activated sludge flow rate and exter-nal carbon dosing are used as control handles. The model-based multivaria-ble controllers included were LQC, DMC (Dynamic Matrix Control) and non-linear predictive control (NPC). When simulating a real influent, the NPC controller display tighter process control with respect to a never-to-exceed total nitrogen license limit compared to the other controllers, and improvements were experienced using DMC and LQC compared to feed-back control. All model-based controllers showed an increase in total costs of 16-25 % compared to feedback ammonium and nitrate control. Aeration costs and license costs were reduced while carbon costs increased signifi-cantly (factor 10). It is unknown how the ammonium and nitrate concentra-tions were affected since only total nitrogen was reported. The model-based controller with the lowest aeration cost (LQC) is presented in Table 3.1. The


model in the NPC is ideal, since it is the same model as the simulation mod-el.

Later on, Shen et al. (2009) compared DMC, Quadratic DMC (QDMC) and various nonlinear MPC algorithms. According to their simulations, a nonlinear MPC improves performance (measured as EQI) but at the cost of increased energy consumption. QDMC is not reported to outperform DMC. Also, feedforward for disturbance rejection is investigated in combination with DMC. Ammonium feedforward brings better results than only measur-ing flow rate and a combination of the two signals is even better. Perfor-mance is improved through feedforward but with increased energy consump-tion. A similar study published by Shen et al. (2008) the year before com-pared QDMC, QDMC with feed-forward and nonlinear MPC. The authors conclude that non-linear MPC handles disturbances best and with acceptable energy consumption.

Weijers (2000) develops a methodology for an improved control system design approach. The author searches to construct a method for control goal formulation and argues in support of a mathematical approach to set up the design problem. The case studies of the thesis where the proposed control system design approach is applied are exclusively examples of model-based control. Both a predenitrification system and a carousel system are evaluated by use of simulations with the ASM1.

Several studies report on the oxygen concentration tracking. Brdys and Konarzcak (2001) investigated a non-linear SISO MPC based on the oxygen dynamics. The method was improved through increased computational effi-ciency by a fuzzy predictive controller in Brdys and Diaz-Maiquez (2002). A nonlinear MPC and an Adaptive Model Reference Controller (DMRAC) were compared in Chotkowski et al. (2005). In Piotrowski et al. (2008), the nonlinear MPC is supplemented by a model of the blower system. The DO controllers above fit into the hierarchical control structure set up by Brdys et al. (2008), which is divided into the supervisory control layer, optimising control layer and follow-up control layer. The structure involves integrated control of a treatment plant and sewer system and the optimising control layer involves slow, medium and fast time scales. The optimising control layer contains a MIMO (Multiple Input Multiple Output) robust MPC and other advanced methods. In this layer the DO trajectory is produced to the lower level DO controller in the follow-up control layer. In Brdys et al. (2008), the low-level DO controller in the follow-up layer consists of a sim-ple proportional controller, however, the authors argue that a much better solution would be to apply an MPC for this purpose. The simulations in the report are based on real data from the Katurzy UCT (University of Cape Town) treatment system in Poland.

The set-point of DO was determined with MPC in Sanchez and Katebi (2003). Sub-space identification is used to create models for DO. The au-thors compare three different MPC controllers with a single PI controller


with constant set-point. The evaluation is performed mainly through compar-ing system overshoot and settling time, criteria which are not as important in DO control as they are in servo systems.

The thesis by Holenda (2007) investigates methods to optimise the aera-tion length in an alternating system, but also develops an MPC controller for the purpose of DO set-point tracking. The MPC is based on a linearised ver-sion of the ASM1.

Stare et al. (2007) compare several control strategies in BSM1: constant manipulated variables, DO control, nitrate and ammonium feedback control (with and without feedforward control) and MPC. The lowest aeration cost is required by the ammonium and nitrate feedback controllers. When opera-tional costs (including effluent fines) are included the MPC and feedfor-ward-feedback controller performs the best, with slightly better performance by the MPC. The MPC model is the benchmark model with perfect meas-urement, meaning the MPC model is ideal since it incorporates perfect pro-cess knowledge which is never achievable in a real process. The authors conclude that improvements by process control are more related to control structure than to choice of control algorithm.

In Ekman (2008) a bilinear discrete time model is estimated using a re-cursive prediction error method. Data from a simulated activated sludge process with post-denitrification were used. A bilinear model predictive con-trol algorithm was derived and applied to the simulation model. The results reveal that even though the identified bilinear model describes the dynamics of the activated sludge process better than linear models, bilinear model pre-dictive control only gives moderate improvements of the control perfor-mance compared to linear model predictive control laws.

Zarrad et al. (2004) and Vilanova et al. (2009) compare decentralised PI controllers to multivariable model-based controllers in the BSM1. In the first paper, a nitrate recycle PI loop and an air flow rate PI loop are compared to two model-based controllers (LQC, disturbance accommodation controller (DAC)). The PI controllers demonstrated better results (measured as EQI) than the model-based controllers with respect to EQI and aeration energy. This was motivated by the fact that the dynamics of the processes at hand are very different and can therefore be controlled with decentralised controllers. Vilanova et al. (2009) compare the performance of a multiloop PI controller to a multivariable controller in a single aerated reactor. DO and substrate concentration are considered. The results when analysing step responses are comparable for the two controllers.

DO control by MPC with a process model incorporating classical DO dy-namics is verified by Holenda et al. (2008). The effect of sampling time is investigated, indicating improved controller performance with decreased sampling time but this does not impact plant performance or operating costs. The MPC controller compared to standard PI control shows marginal im-provements on EQI and a small increase in aeration energy when comparing


simulation results in BSM1. The controller was also tested on an alternating process. DO control with MPC is also found in Ostace et al. (2011), who also consider ammonium MPC which performs better than only DO control. Simulations are performed in BSM1, where only the last aerobic compart-ment is controlled. Effluent nitrogen is increased with ammonium MPC while aeration energy is decreased by 15 %.

Model-based set-point optimisation has been investigated by Guerrero et al. (2011). The optimisation searched for the set-points of e.g. ammonium using a pattern search method. The optimisation minimised total operational costs, including effluent fines, pumping energy etcetera. Due to this, all the control strategies did not treat the same amount of ammonium (effluent am-monium varied from 0.6 to 7 mg/l) making comparisons difficult if only looking at aeration control performance. Several control strategies are evalu-ated, including using constant DO of 4 mg/l and optimal set-points of am-monium and nitrate, which are fixed or updated in different intervals. The best strategies with regard to the operating costs were using separate but fixed nutrient level set-points for weekdays and weekends respectively.

Genetic algorithms (GA) have been used to calculate model-based con-trollers. Genetic algorithms have search strategies inspired by the process of natural evolution. Yamanaka et al. (2006) evaluate a cost minimisation con-trol scheme using BSM1. The DO set-points are calculated using genetic algorithms and a simplified process model. Multiobjective genetic algo-rithms were used by Beraud et al. (2009) to find the best set-point in three consecutive aerobic zones and obtain energy reductions of 10-20 % com-pared to the original benchmark performance.

Early developments of advanced controllers simulated in wastewater treatment models are given above. Many examples are given where DO ra-ther than ammonium is the controlled variable. To test advanced controllers for DO control has continued to be popular since ASM1 or simplified ver-sions of it are complex non-linear models. Some recent examples include self-organising radial basis function (RBF) neural network model predictive control (H.-G. Han et al., 2012), adaptive neuro fuzzy inference system (ANFIS) inverse control (Gaya et al., 2013) and fuzzy model-based predic-tive control (Yang et al., 2013). As in Sanchez and Katebi (2003), the studies often compare closed-loop overshoot and settling time.

C. Advanced control – full-scale and pilot-scale case 3.4.5studies

The STAR control system (Superior Tuning and Reporting) (Thornberg et al., 1992; Önnerth et al., 1996) is an early example of an advanced controller in full-scale systems. Ingildsen (2002), Chapter 3, presents a summary of benefits from implementations of the STAR control system in Denmark and


Sweden. The summary was originally presented at a seminar in 2001 ar-ranged by the Society of Civil Engineers in Denmark. The benefit-cost ratio ranged from approximately four to up to around ten. The savings were large-ly due to avoided capital costs from plant extensions.

STAR is a supervisory model-based control system using on-line meas-urements and process data. It was primarily developed for intermittent aera-tion of the Biodenitro process, and is put on top of the plant’s SCADA (Su-pervisory Control and Data Acquisition) system. Today, STAR consists of several modules for treatment plant control. Full-scale examples of STAR are described by Nielsen and Önnerth (1995) and Thornberg et al. (1998). STAR was published before the main period of interest for this review and to the authors’ knowledge there are only published results on alternating pro-cesses. STAR is nevertheless mentioned since it is an early example of an advanced controller.

Stare et al. (2006) developed reduced order non-linear models based on mass balances and subsequently performed simulations of an MPC controller for an MBBR. The results demonstrated tighter process control with a non-linear model than with a linear model. For this reason, a non-linear MPC was evaluated in pilot-scale in Vrečko et al. (2011), where the MPC controller was compared with the feedforward and feedback controllers previously published in Vrečko et al. (2006). The MPC controller performed better than conventional feedback control, but compared with the present MPC model the feedforward controller used 16 % less air flow per kg ammonium re-moved.

Kandare and Nevado Reviriego (2011) used adaptive predictive expert control to keep the DO concentration at a specified level and to achieve an air pressure level that minimizes power consumption. The controller has been evaluated in a pilot-plant, showing smoother DO control compared to PID DO control with fixed parameters. The reference PID controller oscil-lates and does not appear to be performing as a well-tuned state-of-the-art PID controller. The study also looked at pressure set-point optimisation.

O’Brien et al. (2011) present full-scale results from a MPC for aeration control. The previous on/off control (0.5/1.5 mgO2/l) of the surface aerators in an activated sludge process for BOD-removal was improved by using a MPC based on a black-box model. The model uses a feedforward term from measured incoming BOD. The BOD was measured with a spectral analyser (“spectrolyser”). The MPC had aerator powers as manipulated variables and DO in the 2 lanes as controlled variables. The constraints in the MPC were minimum power levels and minimum required DO concentrations for carbo-naceous removal. Compared to the reference on-off controller the MPC saved 20 % of the energy by keeping the DO concentration closer to the set-point of 1 mg/l. The on-off controller operated in the DO range of 1-2 mg/l.

Two of the studies mentioned above (Kandare and Nevado Reviriego, 2011; O’Brien et al., 2011) do not involve ammonium control, but the con-


troller structure is closer to control structure A (DO control). The model-based controllers are categorised as advanced control since the controllers incorporated an advanced control strategy.

Control of the aerobic volume 3.5 D. Volume control – simulation studies 3.5.1

Suescun et al. (2001) incorporated a facultative zone in the control which could be made aerobic if the DO set-point was at its maximum level. The facultative control strategy was needed to meet ammonium and nitrate efflu-ent requirements.

Samuelsson and Carlsson (2002) controlled the ammonium concentration in the last compartment in a plant model through a model-based feedforward strategy which changed the aerobic volume. The approach was based on on-line estimation of the reaction rate of ammonium and combined feedforward with feedback control.

As mentioned, Krause et al. (2002) simulated the control of the DO set-point and aerated volume in simulations. Through a feedforward model sev-eral switching points were determined for the three aerated compartments in a predenitrification plant. The improved handling of ammonium peaks was due to the control of aerated volume together with an early increase in aera-tion intensity created by the feedforward controller.

Meyer and Pöpel (2003) used fuzzy logic to determine the DO set-point and the fraction of aerobic volume. Compared to using on-off control based on effluent ammonium to decide the aerated volume, the fuzzy controller could reduce the nitrogen peaks.

D. Volume control – full-scale and pilot-scale studies 3.5.2Brouwer et al. (1998) is an early example where a feedforward model-based approach was used to determine the aerobic volume needed for complete nitrification. A simple process model together with estimation of biokinetic parameters through batch respirometric measurements in one of the plant compartments decided the necessary aerobic volume. Evaluations were per-formed in a pilot plant.

Baeza et al. (2002) varied the total aerobic volume in a pilot plant A2O process fed with synthetic wastewater, leading to a 10 % increase in nitrogen removal compared to operation with a fixed volume. Estimation of COD was performed by turning off the aeration for short periods, hence enabling OUR (oxygen uptake rate) calculation. The COD estimations served as inputs to a neural network model which determined the total volume to be aerated. From the description in the paper, it is not obvious if the fixed volume is

3.6 Comparison between case studies 65

smaller or equivalent in size to the average volume in the case with a varied volume.

Svardal et al. (2003) used the measured air flow rate and DO concentra-tion to decide how to adjust the aerobic volume to the ammonium load. The method maximises the anoxic volume given the goal of complete nitrifica-tion. The OUR is in this study a good indicator of nitrification requirement at low ammonium concentrations, and is approximately proportional to the air flow rate. The method is based on increasing the aerobic volume when the total air flow passes certain thresholds adapted to the changes of the monthly temperature ranges over the year in moderate climatic zones. The paper pre-sents full-scale results from the Linz-Asten WWTP in Austria which uses an oxidation ditch for simultaneous nitrification and denitrification.

Ekman et al. (2006) developed a method for aerobic volume control that only required measurements of the DO concentration. The method made use of supervisory control where two out of three zones could be either aerobic or anoxic depending on the DO concentration in all three zones, creating a disturbance rejection effect. The strategy was evaluated in the BSM1 and in the large pilot-plant facility Hammarby Sjöstadsverk in Stockholm, Sweden, treating municipal wastewater. The pilot plant evaluation suggested that volume control can give lower nitrogen concentrations in the effluent with less energy consumption compared to constant DO control.

Comparison between case studies 3.6Table 3.1 is aimed to guide through the research within aeration control of activated sludge processes. It summarises comparative studies between con-trol strategies and covers full-scale, pilot-scale and simulation results. The table includes research that compares one or several strategies to some refer-ence case. Some papers are listed more than once to simplify reading, since they cover several comparisons.

In order to judge the efficiency of a control strategy or control structure in a case study the reader would need some quantitative indicator, such as en-ergy consumption in combination with a measure of the treatment perfor-mance. Other important data include whether or not nitrate is controlled (through carbon addition or internal recycle control), changes in carbon addi-tion, the water temperature, the duration of the evaluation and the type and level of effluent permit. As can be seen in Table 3.1, not all data are availa-ble from all the studies. Even if the publications were carefully read there is no guarantee that all available data was found. If available, the control goals are listed in the order mentioned in the paper. If internal recycle is used, improved nitrogen removal could be a result of nitrate control rather than aeration control. This is not always clear from the presentations in the papers in Table 3.1.


If effluent concentrations are listed as intervals in Table 3.1 the numbers have been read from a graph, otherwise averages are listed. Maximum con-centrations were often estimated from graphs and are not necessarily repre-sentative for the whole evaluation period.

Energy consumption is reported in various ways in the research papers. In Table 3.1, the reference controller’s energy consumption is given if reported by the paper, together with the type of energy measure given (air flow rate, aeration power or simply aeration energy without any specific unit). Air flow rate is recalculated to Nm3/d and aeration power to kWh/d if reported in other units. The energy consumption of the investigated control strategy is reported as saving in percent compared to the reference case.

In Figure 3.2 the number of simulation, pilot-scale and full-scale studies mentioned in this review is summarised. The publications are divided based on the control structure (A-D) and the two categories method development study and case study. A method development study does not compare two control strategies to each other, but looks at the method of the controller itself. The papers with case studies are often counted more than once, one time for the reference strategy and one or more times for the experimental strategies.

Figure 3.2 shows that as the control structure becomes more advanced, simulation studies and method development studies becomes more domi-nant.

(a) (b)

Figure 3.2. Number of publications in the literature review categorised as (a) simu-lation studies or (b) full-scale and pilot-scale studies.

0

5

10

15

20

25

A B C D

No.

pub

licat

ions

Method development Case studies

0

5

10

15

20

A B C D

No.

pub

licat

ions

Method development Case studies

Tab

le 3

.1. C

ompa

riso

n be

twee

n co

ntro

l str

ateg

ies.

Typ

e of

stu

dy: f

= f

ull-

scal

e, p

= p

ilot-

scal

e, s

= s

imul

atio

n st

udy.

Typ

e of

pro

cess

: A

SP

= a

ctiv

ated

slu

dge

proc

ess,

Pre

-DN

= p

rede

nitr

ific

atio

n, A

2 O =

ana

erob

ic-a

noxi

c-ox

ic, M

BB

R =

Mov

ing

bed

biof

ilm r

eact

or,

ML

E =

Mod

ifie

d L

udza

ck-E

ttin

ger

proc

ess,

UC

T =

Uni

vers

ity

of C

ape

Tow

n pr

oces

s. T

ype

of c

ontr

ol s

trat

egy:

A. D

O c

asca

de c

ontr

ol, B

1.

Am

mon

ium

-bas

ed f

eedb

ack

cont

rol,

B2.

Am

mon

ium

-bas

ed f

eedf

orw

ard-

feed

back

, C. A

dvan

ced

SISO

and

MIM

O c

ontr

olle

rs, D

. Con

trol

of

the

aero

bic

volu

me.

Oth

er a

bbre

viat

ions

: Exp

= E

xper

imen

t, R

ef =

Ref

eren

ce c

ase,

TN

=T

otal

nitr

ogen

, SP

= s

et-p

oint

, AF

R =

air

flo

w r

ate,

A

P =

aer

atio

n po

wer

, AE

= a

erat

ion

ener

gy. *

The

fee

dfor

war

d co

ntro

ller

is n

ot s

uppo

rted

by

feed

back

con

trol

.

Aut

hors

Proc

ess

Stru

ctur

e

Con

trol

goa

l N

H4

NH

4SP

N

H4 m

ax

NO

3 T

N

DO

E

nerg

y T

emp

°CE

val.

time

NO

3 ct

rl

Com

men

t/ ef

fluen

t lim

it

Exp

R

ef

E

xpR

efE

xpE

xpR

efE

xpR

efE

xpR

efE

xpR

ef

Exp

(%)

Ref

(uni

t)

H

usm

ann

et

al. (

1998

) f

Step

-fee

d B

1+D

A

L

ow N

H4

Ene

rgy

redu

ctio

n 0-

2 0-

6 2.

0 10

14

1-

12

6-20

0.5-

2.0

2.5

-16

1 Y

dat

a (A

E)

16

1 W

N

N

H4<

10

TN

<18

H

usm

ann

et

al. (

1998

) f

Step

-fee

d B

1+D

A

M

aint

ain

nitr

ific

atio

n 0-

6 2-

18

6 18

4-

12

4-14

0.5-

2.0

2.5

-16

1 Y

dat

a (A

E)

7 1

W

N

NH

4<10

T

N<

18

Stef

fens

and

L

ant (

1999

) s

Pre-

DN

C

B

2 O

pera

ting

cos

ts

Lic

ense

vio

latio

n D

efer

red

cap.

exp

.

3-

15

4-22

0-

4.5

0.5-

5 -3

3 45

3 (A

UD

/d)

4.

5 D

Y

T

N<

10 (

GS)

+

1015

% c

arbo

n

Gal

arza

et a

l. (2

001)

s

RD

N/D

RD

N

B1

A

Red

uce

cost

s D

enitr

ific

atio

n pe

rf.

0.5-

2 0.

5-2

2.

5 4

1-2.

5 2.

5 -1

6 (A

FR)

8

M

Y

T>

13: N

H4+

NO

3<13

Sue

scun

et a

l. (2

001)

s

Pre-

DN

B

1+D

A

N

H4+

NO

3 ct

rl

Ene

rgy

redu

ctio

n 0-

2 0-

4 2.

0 10

14

0-

12

0-12

1.0-

2.25

2.

0 -1

1 (A

FR)

1

Y

Y

T<

13: N

H4<

3,

NO

3<15

, T>

13:

NH

4<2,

NO

3<10

Se

rral

ta e

t al.

(200

2)

s B

arde

npho

B

1 A

T

N b

elow

lim

it

Min

imis

e N

O3

0.5-

1.5

1.5

<

0.2

<0.

4

1.

3-1.

6 2.

0 -1

0 (e

stim

ate,

AE

)

5 D

Y

Kra

use

et a

l. (2

002)

s

Pre-

DN

B

2+D

B

1+D

E

fflu

ent p

erm

it

Ene

rgy

redu

ctio

n R

educ

ed C

sou

rce

0.8

2

6 7.

5

6-

16

6-16

0.

5-2.

0

+6

(AFR

) 11

-16

17 D

N

T

>12

: NH

4<5

(GS)

Ingi

ldse

n et

al

. (20

02)

f Pr

e-D

N

B2*

A

E

fflu

ent p

erm

it

Ene

rgy

redu

ctio

n S

impl

est c

ontr

olle

r

0.5

0.4

2.

5 3.

5 5.

6 5.

6

0.

25-2

.5

-5

to

-15

(est

imat

e, A

E)

35

D

N

Ingi

ldse

n (2

002)

f

Pre-

DN

B

2 A

Se

e ab

ove

1.6

1.1

2.0

6 6

4.1

5.6

0.25

-2.5

2/

1.8

-11.

7 31

968

(A

FR)

24

D

Y

Ingi

ldse

n (2

002)

f

Pre-

DN

B

1 A

Se

e ab

ove

1.9

2.4

1.5

5

0.25

-4

-1

3.5

37 2

96 (

AFR

)

17 D

Ingi

ldse

n (2

002)

f

Pre-

DN

B

1 A

Se

e ab

ove

3.3

3.

0

0.

25-3

-27.

5 42

912

(AFR

)

41 D

Mey

er &

Pö

pel (

2003

) s

Pre-

DN

B

2+D

A

0.

5

1.7

3.1

5.7

9.1

6.6

9.6

2.

0 -2

5 32

3 (A

FR)

3

D

N

Mey

er &

Pö

pel (

2003

) s

Pre-

DN

B

2+D

A

+D

0.

5

1.7

3.9

5.7

7.7

6.6

8.5

2.

0 -1

2 28

2 (A

FR)

3

D

N

Vrečk

o et

al.

(200

3)

s B

SM1

B2

A

0.

5-3.

20.

5-4

1.0

3.2

4

12

-22

12-2

20.

2-5.

0

1

W

N (

Y)

Yon

g et

al.

(200

5)

s B

SM1

B1

A

2.

4 2.

8 2.

0 7.

9 8.

4

17

.7

18.5

3.

2 3.

5 -9

.6

1556

9 (A

P)

1

W

N

Yon

g et

al.

(200

5)

s B

SM1

B2

A

2.

1 2.

8

6.9

8.4

17.3

18

.5

3.0

3.5

-9.3

15

569

(AP

)

1 W

N

Aut

hors

Proc

ess

Stru

ctur

e

Con

trol

goa

l N

H4

NH

4SP

NH

4 max

N

O3

TN

D

O

Ene

rgy

Tem

p °C

Eva

l. tim

eN

O3

ctrl

Com

men

t/

efflu

ent l

imit

E

xp

Ref

Exp

Ref

Exp

Exp

Ref

Exp

Ref

Exp

Ref

Exp

Ref

E

xp (%

) R

ef (u

nit)

Yon

g et

al.

(200

5)

p P

re-D

N

B2

A

4.

2 4.

9 4.

0 9.

2 21

A

s re

f.

0-6

2.5

Up

to

-9.6

(A

E)

20

60 D

N

Liu

et a

l. (2

005)

f

ML

E

B2

A

4.9

7.7

2.

75/ 2

/0.5

-1

9 (A

E)

11

W

Y

Vrečk

o et

al.

(200

6)

p P

re-D

N

(MB

BR

) B

1 A

1.1

3.5

1.0

3.7

7.8

5.5-

8.5

6.8-

7.8

-23

1098

(A

FR/k

gNH

4)

15

5 D

per

st

rat.

N

Vrečk

o et

al.

(200

6)

p P

re-D

N

(MB

BR

) B

2 A

1.0

3.5

1.0

2.2

7.8

5-8

6.8-

7.8

-45

1098

(A

FR/k

gNH

4)

15

5 D

per

st

rat.

N

Ekm

an e

t al.

(200

6)

p P

re-D

N

A+

D

A

4.

5 6.

7

2.2

6.7

6.7

13

-37

348

(AFR

) 22

/19

5 W

N

Bar

oni e

t al.

(200

6)

f P

re-D

N

B1

A

Proc

ess

eff.

+st

ab.

Ene

rgy

redu

ctio

n 5.

4 3.

0

11.6

10.1

0.3-

1.7

2.2

-4

1650

6 (A

P)

1

Y

N

Aye

sa e

t al.

(200

6)

f D

RN

/DR

DN

B

1 A

Pr

oces

s st

abili

ty

Eff

luen

t qua

lity

Cos

t

1.0

1.5-

4 1.

5 -1

5 to

-20

(A

E)

1

Y

Y

Sta

re e

t al.

(200

7)

s B

SM

1 C

B

1 O

pera

ting

cos

ts

Tre

atm

ent p

erf.

1.

5-3

1-6

2.0

3.0

6.0

8-16

8-

18

0.2-

2.5

0.2-

3 -0

.2

538

(€/d

)

1 W

Y

E

fflu

ent

fine

s +

2.7

% c

arbo

n S

tare

et a

l. (2

007)

s

BS

M1

B2

A

Ope

rati

ng c

osts

T

reat

men

t per

f.

1.5-

4 0.

5-8

2.0

4.0

8.0

8-18

11

-16

0.2-

3 0.

5 +

5.7

512

(€/d

)

1 W

Y

E

fflu

ent

fine

s -5

3 %

car

bon

Zha

ng e

t al.

(200

8)

s R

educ

ed

BS

M1

B2

A

Eff

luen

t sta

ndar

ds

Min

imal

cos

t 3.

15

3.23

7.5

8 10

.8

11.3

0.5-

3

-4.7

47

794

(AP

)

2 W

N

Wal

z et

al.

(200

9)

f M

LE

B

2 A

E

nerg

y re

duct

ion

As

ref.

As

ref.

2/

1.3

2.5/

2 -1

5 (A

P)

48

D

N

Thu

nber

g et

al

. (20

09)

f U

CT

B

1 O

ther

E

nerg

y re

duct

ion

Mai

ntai

n tr

eatm

ent

0.2

0.2

0.2

3.5

3.6

1.8-

1.2/

11.

8/1

-18

(AFR

)

8 D

N

Thu

nber

g et

al

. (20

09)

f U

CT

A

O

ther

E

nerg

y re

duct

ion

Mai

ntai

n tr

eatm

ent

0.2

0.4

8.

1 6.

4

-2

6 28

0 80

0 (A

FR)

1 Y

N

Tho

rnto

n et

al

. (20

10)

f B

arde

npho

B

2 A

0-1

<0.

51.

0 1.

5 0.

5 6-

10

7-12

0.75

-1.4

2/

2.1/

1.

6/0.

5 -2

0 (A

FR)

1

W

Y

-50

% c

arbo

n

She

n et

al.

(201

0)

s A

2 O

B2*

+D

A

E

fflu

ent c

rite

ria

Ene

rgy

redu

ctio

n ~4

~4

~12

~14

0-1/

2-3/

2-3/

1 1/

3/3/

1 -8

.5

228

392

(AFR

)20

N

Impr

oved

P-

rem

oval

K

anda

re a

nd

Rev

irie

go

(201

1)

f A

SP

C

A

DO

con

trol

Pr

essu

re c

ontr

ol

0.

5-3

0.5-

6 -2

7 23

5 (k

W)

1

D

N

Incl

udes

pre

ssur

e

opti

mis

atio

n

O’B

rien

et a

l. (2

011)

f

ASP

C

O

n/ o

ff

Ene

rgy

redu

ctio

n D

O c

ontr

ol

1-

1.5

2-3

-20

(AP)

2 W

N

Vrečk

o et

al.

(201

1)

p P

re-D

N

(MB

BR

) C

B

2

1.1

1.0

1.0

2.8

2.2

6-8

+

19

602

AF

R/k

gNH

4

3 D

N

Rie

ger

et a

l. (2

012

b)

s P

re-D

N

B1

A

Impr

oved

TN

rem

. E

nerg

y re

duct

ion

Red

uced

C s

ourc

e

1.0

~5

~5

0-2.

5 2.

5/2/

1 -5

(A

E)

12

1 W

Y

T

N<

8, T

P<

2 -5

to -

10 %

car

bon

Aut

hors

Proc

ess

Stru

ctur

e

Con

trol

goa

l N

H4

NH

4SP

NH

4 max

N

O3

TN

D

O

Ene

rgy

Tem

p °C

Eva

l. tim

eN

O3

ctrl

Com

men

t/

efflu

ent l

imit

E

xp

Ref

Exp

Ref

Exp

Exp

Ref

Exp

Ref

Exp

Ref

Exp

Ref

E

xp (%

) R

ef (u

nit)

Rie

ger

et a

l. (2

012

b)

s P

re-D

N

B1

A

Impr

oved

TN

rem

. E

nerg

y re

duct

ion

Red

uced

C s

ourc

e

1.0

~5

~5

0-2.

5 2.

5/2/

1 -2

0 (A

E)

20

1 W

Y

T

N<

8, T

P<

2 -2

0 to

-40

% c

arbo

n

Rie

ger

et a

l. (2

012

b)

s P

re-D

N

B1

A

Impr

oved

TN

rem

. E

nerg

y re

duct

ion

Red

uced

C s

ourc

e

1.0

~5

~5

0-2.

5 2.

5/2/

1 -3

0 (A

E)

30

1 W

Y

T

N<

8, T

P<

2 -2

0 to

-40

%

carb

on

Rie

ger

et a

l. (2

012

c)

s P

re-D

N

B1

A

Incr

ease

N r

em.

Ene

rgy

redu

ctio

n

+32

%

rem

2.

0 -2

5 ($

/Y)

15

N

E

fflu

ent t

ax, i

ncl.

supe

rnat

ant c

trl

-50

% c

arbo

n R

iege

r et

al.

(201

2 c)

s

Pre

-DN

B

2 A

E

nerg

y re

duct

ion

0-1

1-2

2.

5 3.

5

+

48%

re

m

0-

2 0.

5-4

-30

($/Y

)

N

Rie

ger

et a

l. (2

012

c)

f A

2 O

B1

A

Red

. op.

cos

ts

Hig

her

N r

emov

al

Hig

her

P re

mov

al

1.0

2.

0 4.

0

+40

%

rem

2.

0 -1

7 ($

/Y)

4

W

(B1)

N

E

fflu

ent t

ax

Åm

and

et a

l. (2

014)

f

pre-

DN

B

1 A

E

nerg

y re

duct

ion

Mai

ntai

n tr

eatm

ent

1.3

1.4

1.0

2.5

2.9

-7

(est

imat

e, A

FR)

17

1 Y

N

Åm

and

et a

l. (2

014)

f

pre-

DN

B

1 A

E

nerg

y re

duct

ion

Mai

ntai

n tr

eatm

ent

0.7

0.8

0.8

1.4

2.0

-10

(est

imat

e, A

FR)

12

8 M

N

Åm

and

et a

l. (2

014)

f

Nitr

ific

atio

n B

1 O

ther

E

nerg

y re

duct

ion

Mai

ntai

n tr

eatm

ent

2.2

2.0

2.0

1.7/

2.2

2.2/

4.9

-19

(est

imat

e, A

FR)

14

1 Y

N


Critical review of aeration control systems 3.7 Full-scale considerations 3.7.1

It is challenging to draw conclusions about the superiority of one control algorithm over another in full-scale operation, given the many uncertainties in operation with regards to load, equipment and other local conditions, such as plant capacity.

It is important to recognise limited control authority due to constraints of the system such as blowers (min/max air flow), sensor signal quality or sen-sor location. Another reason for limited control authority is that plants have a limited mass of nitrifiers which limits the possibility to treat peak loads. Also when operating at a long sludge age, the nitrification capacity will depend on the average load of ammonium that is treated, as discussed in Rieger et al. (2012b).

Controller performance has to be related to the legislative framework which the plant has to comply with, which in turn will determine the control goal for the process. Internationally, there are different levels of total nitro-gen or ammonium limits. There are also different averaging periods when assessing effluent limit compliance. If the plant nitrifies more than is re-quired by the effluent permit there is additional room for improvement by reducing aeration. Grab sampling and effluent fines on ammonium will re-quire limitations of ammonium peaks and it might be worth an increased energy consumption to satisfy these demands. Similarly, if denitrification is limited it may be worthwhile reducing aeration and thereby increasing the total nitrogen removal while perhaps emitting more ammonium. Whether this is a good thing or not depends on plant specific factors and effluent lim-its and thereby on how the control goal is defined. This is the reason why not all control structures will be useful at all plants. There is not a single evalua-tion criterion in place, but the evaluation is a plant-specific multi-criteria task.

Another aspect of judging the performance of a controller is the choice of the reference controller. Nitrification can only be improved if the reference case is limited in some way. Similarly, the energy saving with a new control-ler can be exaggerated if compared with a poor-performing reference case.

The time scale in a comparative study is important, especially in full-scale. Most full-scale studies are, due to the reality, not very long. By chance, it is possible to reach results that are not representative for the plant or for the control structure under study. In Table 3.1, five of the fourteen full-scale comparisons are made during longer periods than approximately 1 month, and three of these look at periods of about one year. It is not obvious from the papers in Table 3.1 that the investigated full-scale controllers are still operating at the plants.

3.7 Critical review of aeration control systems 71

Modelling 3.7.2Simulation is a valuable tool for evaluation and development of controllers as discussed in Section 2.5. Modelling can provide next to unlimited flexibil-ity and opportunities in control and process choices. However, there is a gap between real world limitations, and the opportunities that a simulation model can offer. Simulations are a step ahead with regards to development and testing of new controllers, but it is important that the gap is not allowed to become too large.

The limitations and quality of the models used are critical in modelling and simulation, in the design of model-based control algorithms as well as in control system evaluation. It is said that “All models are wrong, but some are useful” (the statistician George E.P. Box, 1979). The models used for wastewater treatment modelling are constantly evolving (Jeppsson et al., 2011). It is encouraging that more efforts are put into modelling e.g. sensor failure (Rosen et al., 2008), trying to add more realism to models used for controller comparison and development.

The BSM1 model and the BSM2 model differ in the design of the plant in relation to the load. Essentially, BSM1 is highly loaded. Ammonium control-lers in BSM1 which are evaluated without adjusting the load or the volume of the modelled reactor has little control authority, rendering evaluations of aeration control strategies difficult. Several early investigations in the BSM1 model investigating feedforward-feedback control tolerates high DO concen-trations above 2-3 mg/l (Vrečko et al., 2003; Yong et al., 2005). Despite this they cannot decrease the ammonium peaks to a large extent.

Ammonium-based control 3.7.3Ammonium-based control in the form of feedforward and feedback control can add two advantages to the process: the potential to limit aeration during periods with low effluent ammonium (and possibly limit complete nitrifica-tion) and the possibility to increase aeration intensity to limit ammonium peaks during peak load. There are several examples from the last decade on well-performing supervisory controllers.

Feedforward control should be used with care, since it adds more sensors and extra complexity to the control system. The controller also has to be provided with a feedforward model. Using feedforward control can be moti-vated by discharge criteria where it is never allowed to be above the limit. Adding feedforward control can use more energy than using only feedback control (Krause et al., 2002; Stare et al., 2007), which could be justified by the effluent criteria. Using feedforward without a feedback loop is not rec-ommended since feedback contributes with a more robust performance in the light of feedforward model uncertainty and can compensate for non-modelled disturbances.


A thorough discussion on feedforward ammonium-based control is given in Rieger et al. (2012b). Two of the points made are that (1) feedforward should only be used when there is a benefit of reacting fast to a disturbance (such as “never-to-exceed” effluent limits) compared to pure feedback con-trol and (2) feedforward control can easily lose control authority if nitrifica-tion capacity is limited.

Volume control 3.7.4Volume control can offer three benefits. Firstly, volume control can add control authority at high loads to be able to decrease ammonium peaks. Un-like ammonium feedback control, volume control can rapidly increase nitri-fication capacity at times when feedforward control is limited by the lack of nitrifier mass and the DO concentrations should not be further increased. Secondly, volume control can be a tool to save energy when a plant is low loaded and parts of the volumes are not required for nitrification. Conse-quently the additional aeration merely contributes to endogenous respiration. Finally, if denitrification is limited then volume control can provide extra anoxic volume. Volume control can therefore be a means to balance total nitrogen removal. The efficiency of volume control is much improved if walls are limiting the zones that are switched on and off.

The possibility to control the aerobic volume demonstrates the important coupling between design and operation. Many plants are not designed to use available volumes in the best possible way. For example, the volumes for denitrification and nitrification are not always matched. Volume control can be used to better utilize the plant capacity for both organic removal, in-creased energy efficiency by good use of the denitrification volume, and for nitrification.

Advanced control 3.7.5The wastewater treatment process is often referred to as being complex, non-linear, with a range of time constants and never being in steady-state. In the literature, it is possible to find arguments supporting decentralised SISO controllers, as well as MIMO controllers. Some argue that given the complex nature of the system, simple control cannot guarantee good performance under a full range of conditions (Brdys et al., 2008). Simple SISO controllers are not sufficiently robust and control of an activated sludge system must be considered a multivariable control problem and should thereby best be han-dled with model-based control methods, such as MPC (Steffens and Lant, 1999; Weijers, 2000). On the other hand, other researchers argue that con-ventional well-tuned control algorithms are sufficient to achieve acceptable system performance under most conditions (Ingildsen, 2002; Stare et al., 2007). It is possible to find sub-processes that can be controlled with linear

3.8 Implementing adequate aeration control for full-scale operation 73

controllers that demonstrate little coupling with other processes (Vrečko et al., 2002). That is particularly true for intermediate process variables – such as the DO concentration.

The DO concentration is under normal conditions well controlled with a properly implemented PI-controller as is demonstrated in several of the full-scale studies in Table 3.1. Early attempts to use a self-tuning controller of higher order in a full-scale process resulted in a controller converging to-wards a PI-controller performance (Olsson et al., 1985). Despite this, at-tempts are still made to control the DO concentration with advanced control algorithms.

The benefit of an optimal controller such as MPC is the possibility to in-clude constraints in combination with handling a MIMO system. If DO con-trol is the goal, constraints in air flow rate/aerator power and limitations of DO concentrations is readily manageable with a simple controller. Hence, the benefit from MIMO and optimal controllers is not on the lower level aeration controllers, but in higher levels of control. As with feedforward control, having a never-to-exceed limit on ammonium could motivate ad-vanced control since a predictive capacity together with handling of con-straints could benefit treatment performance. Looking at Table 3.1, there are so far no examples of advanced full-scale controllers outperforming supervi-sory controllers in continuous aeration where ammonium control is consid-ered. Since the control strategy should be kept as simple as possible, conven-tional feedforward and feedback controllers should be used whenever this is adequate.

Advanced controllers could provide set-points of ammonium and other ef-fluent concentrations, while lower level controllers can be allowed to be feedforward and feedback controllers. This concept would be similar to a plant-wide control approach, which is not within the scope of this review.

Implementing adequate aeration control for full-3.8scale operation

This review has provided a walk-through in recent developments within control of aeration in wastewater treatment plants. The process to implement a new control strategy is systematic, and involves defining the control goal and looking into process and system limitations and possibilities as well as evaluation criteria. To summarise, important aspects for aeration control success in full-scale operation are listed below. These should be considered irrespective of the control structure chosen.


Important conditions of aeration control system success are: • Properly designed aeration system: Proper design of blowers, pip-

ing, valves and diffusers is crucial for good aeration performance. Limitations in maximum and minimum air flow capacity are often limiting the controller performance. Without a blower with possi-bilities for turndown during low load there is no help to be given from a well-tuned and well-working controller further up in the aeration system. Process simulators can be a help in aeration sys-tem design.

• Individual line and zone control: This is the next step in adding flexibility. If the reactor is a plug-flow system, the varying oxygen demand along the reactor is best managed by separate DO control of zones along the basin, using separate actuators in each zone. The same concerns with the individual treatment lines which should be separately controlled to handle individual behaviour.

• Adequate instrumentation and adequate maintenance: The main cause for controller failure is instrumentation. Sensors and actua-tors must always be maintained adequately, and before choosing the final control strategy this must be taken into account in the cost-benefit analysis.

• Sensor location: This is part of the control structure and should be carefully examined.

• Wise controller implementation: Basic requirements and safety nets such as maximum and minimum limits for all controller out-put signals, anti-windup (Åström and Hägglund, 1995) of control-lers with integration, a sufficient range in controller parameters and fall-back strategies are sometimes overlooked. The sampling times used in the control systems have to be carefully considered. It should primarily be selected based on response time of the con-trolled variable. Appropriate filtering of signals should be per-formed taking the sampling time into account.

• Adjustable controller implementation: It is important that the end user can get access rights to make necessary adjustments in the control system. Depending on the possibilities and needs for the plant, the system vendor or the plant operating team can assume different levels of responsibility to the future developments of the system.

• Always consider plant specific aspects: Given the size, load and location of a certain plant, additional aspects may be considered in control structure design and implementation. Such aspects may include effluent taxes, the level and averaging period of the efflu-ent criteria, the variability of the load over the day, energy cost structure, peak demand charges, design characteristics, alkalinity of the wastewater and how close the plant is to its design load.

3.9 Conclusions 75

• Plan for control system maintenance: To achieve successful im-plementation of a new control strategy as well as long term stable operation, operators need to get sufficient training to understand and trust the basics of the controllers. Today there are several vendors offering PID controllers with autotuning capabilities (Åström and Hägglund, 2006), but PID controllers can also be tuned with simple tuning rules. Process simulators can be helpful in operator training.

Conclusions 3.9More than ten years into the 21st century, the state-of-the-art within aeration control is changing. Ammonium feedback control is evolving towards state-of-the art in real application and there are several examples of energy sav-ings and improved nitrogen removal from full-scale case studies. More ad-vanced algorithms including model-based control have their success stories, but so far the main efforts have been focused on method development using simulation models. This review has not found any advanced controllers for continuous ammoni-um control tested in full-scale or pilot-scale studies which outperforms con-ventional feedforward-feedback controllers. Generally, there is an increase of publications from full-scale case studies during the last ten to fifteen years, even though there is a lack of long-term studies. This development will hopefully continue to better establish the benefits of improved process control. Looking at ammonium feedback control, the energy saving of published research is in the range of 5 to 25 %, partly depending on what is the control strategy for comparison.

77

4 HENRIKSDAL, KÄPPALA AND 4

HIMMERFJÄRDEN WWTPS

HIS THESIS HAS three case studies where full-scale experiments have taken place, and simulation models were calibrated to fit the plant behaviour. The three wastewater treatment plants are situated in

and around Stockholm and together they treat wastewater from close to 1.5 million inhabitants. This chapter gives an introduction to the treatment plants and summarises the similarities and differences between them.

Henriksdal WWTP 4.1The plant is the most centrally located treatment plant in Stockholm and treats wastewater from around 780 000 inhabitants. It is in terms of size the largest treatment plant in Sweden. In terms of flow, Henriksdal (/hɛnriksdɑːl/) is second after Rya WWTP in Gothenburg. The plant is oper-ated by Stockholm Vatten AB.

The plant was originally constructed in 1941 and has since then been re-built at several occasions, the latest reconstruction being the improvement of nitrogen and phosphorous removal that occurred from 1992 to 1997. The next reconstruction phase will take place the coming years to meet an in-creased load and more stringent discharge criteria. The present plan involves the construction of the world’s largest membrane bioreactor (MBR) for sol-ids separation. Henriksdal WWTP is one of the largest underground plants in the world, a schematic overview of the plant is found in Figure 4.1.

T

78 4 Henriksdal, Käppala and Himmerfjärden WWTPs

Figure 4.1. An overview of Henriksdal WWTP. The plant covers approximately 300 000 m2. The biological treatment is situated underground while the control building, mechanical treatment and gas facilities are above ground.

Henriksdal has two inlets, the Sickla inlet and the Henriksdal inlet and wastewater arrives from Stockholm and the southern municipalities Hud-dinge, Haninge, Nacka and Tyresö. The treatment process involves mechan-ical treatment with screens, preaeration, a grit chamber and primary sedi-mentation (Figure 4.2). The biological process is a predenitrification process. The chemical precipitation of phosphorous involves preprecipitation with ferrous sulphate in the preaeration tanks and secondary precipitation where iron is dosed prior to the sand filters.

Figure 4.2. The treatment steps in the water line at Henriksdal WWTP. The sludge treatment line includes a centrifuge, digester, a sludge tank, thickener and final cen-trifuge.

The biological treatment plant involves 209 000 m3 of volumes divided into seven parallel treatment lines. Each line is divided into six equally sized main zones, three aerated and three anoxic, with a mixing zone and deox zone in the beginning and end of the line, respectively. Each treatment line has its own return sludge flow. A detail of the predenitrification process at Henriksdal is given in Figure 4.3.

4.2 Käppala WWTP 79

Figure 4.3. The predenitrification process at Henriksdal WWTP. The only wall between the zones is the division between the anoxic and aerobic zones.

Käppala WWTP 4.2Käppala (/ɕæpɑːla/) WWTP is the third largest treatment plant in Sweden and treats wastewater from around 450 000 inhabitants. Käppala WWTP is situated on the island Lidingö in the Stockholm inner archipelago. The plant was constructed in 1969 and a reconstruction was made from 1994 to 2000 to improve the nitrogen removal and increase the volumes. The plant is still today partly operated as two separate plants, the old plant and the new plant representing the volumes that were available before and after reconstruction. The plant capacity is today 700 000 people. Similar to Henriksdal WWTP the plant is an underground facility and the schematic picture is found in Figure 4.4.

Figure 4.4. An overview of Käppala WWTP. The old part of the plant (left) has more narrow basins and the basins have an U-shape compared to the straight and wider basins in the new part of the plant (right).

Nitrate recycle

Return activated sludge


Käppala WWTP has a 65 km long tunnel system connecting the 11 munici-palities in the Käppala Association to the treatment plant. Most of the munic-ipalities are situated north of central Stockholm. The water treatment line is similar to that of Henriksdal WWTP (Figure 4.5). One difference is that fer-rous sulphate is dosed for simultaneous precipitation in the activated sludge process, in combination with secondary precipitation with iron dosing prior to the sand filters. The biological treatment step in the new plant is according to the UCT (University of Capetown) configuration with biological phospho-rous removal (bio-P), and the old plant has the predenitrification process. In the future, Käppala WWTP plans to combine predenitrification with postde-nitrification to reach even lower total nitrogen concentrations in the effluent.

Figure 4.5. The treatment steps in the water line at Käppala WWTP. The sludge treatment line includes a centrifuge, digester, chemical conditioning, and mechanical dewatering.

The biological treatment volume is 143 850 m3 divided into eleven parallel treatment lines – six in the old part and five in the new part. Each part has a combined return sludge flow for all its lines. A detail of the UCT process in the new plant – operating without bio-P – is given in Figure 4.6.

Figure 4.6. The predenitrification process at Käppala WWTP.

Himmerfjärden WWTP 4.3Himmerfjärden (/hɪmɛrfjæːrden/) WWTP is situated in a fjord south of Stockholm and treats wastewater from around 300 000 people. The treatment plant was taken into operation in 1974. Since 1997, the plant has accommo-dated nitrogen removal and during the years 2007 until today, the plant has

4.3 Himmerfjärden WWTP 81

continuously been improved through construction of a disc filter hall, diges-tate supernatant treatment through a DeAmmon process and ozone dosage to prevent bulking sludge. An overview of the treatment process is found in Figure 4.7. Similar to Henriksdal WWTP, the future plans of the plant in-volves an MBR process.

Figure 4.7. An overview of Himmerfjärden WWTP.

Himmerfjärden WWTP receives wastewater from the western and southern parts of the greater Stockholm area. The plant is owned by the municipal company Syvab. The treatment process is somewhat different to the other two plants. Mechanical treatment involves two types of screens and a grit chamber, and the activated sludge process only accommodates nitrification (Figure 4.8). Denitrification is performed in a fluidised bed process incorpo-rating a drum sieve, a fluidised bed reactor and a sand trap. The fluidised bed reactor has a total nitrogen removal of 80 to 90 % during normal operation (Harri and Bosander, 2004). The DeAmmon® process reduces the internal nitrogen load to the plant by reducing the nitrogen content in the nitrogen-rich reject water. The DeAmmon® process makes use of anammox bacteria and was the world’s second full-scale DeAmmon® process.

Figure 4.8. The treatment steps in the water line at Himmerfjäden WWTP. Carbon (C) and phosphorous (P) are dosed to the denitrification process.


The nitrification process has eight parallel treatment lines divided into two blocks, block A and block B. Block A and block B have separate sludge systems. One treatment line with nitrification is presented in Figure 4.9.

Figure 4.9. The nitrification process at Himmerfjärden WWTP.

Similarities and differences 4.4A summary of process data, discharge limits and treatment results for Hen-riksdal, Käppala and Himmerfjärden WWTPs is presented in Table 4.1.

Table 4.1. Data for the three treatment plants. One person equivalent (P.E.) equals 70 gBOD7/p. TN = total nitrogen, DN = denitrification. Averaging periods in the permits are annual (a), quarterly (q) or from July to October (s). Data from 2012 (Käppalaförbundet, 2013; Stockholm Vatten, 2013; Syvab, 2013).

Henriksdal Käppala Himmerfjärden

Connected persons 782 600 454 409 294 419

Industrial load (P.E.) 35 800 38 000 50 – 60 000

Nominal capacity (P.E.) - 700 000 -

Inflow (m3/d) 284 000 160 000 114 800

Influent TN (mg/l) 41 46 33

TN prim. settler (mg/l) 36 42 27

Aerobic sludge age (d) 5 7 10 – 15

Total volume (m3) 204 000 143 850 21 680

Aerobic volume (m3) 94 500 80 296 21 680

Settler volume (m3) 58 000 65 120 21 600 + 24 000

Effluent limits (mg/l) BOD7: 8q, TN: 10a,

TP:0.3q, NH4-N: 3s

BOD7: 8q, TN: 10a,

TP: 0.3a , NH4-N: 3s

BOD7: 8a, q, TN: 8a,

TP: 0.4a

Treatment 2012 (mg/l) BOD7: 5, TN: 8.1,

TP: 0.27, NH4-N: 2.1

BOD7: < 3, TN: 8.7,

TP: 0.2, NH4-N: < 1

BOD7: 7.7, TN: 8.8,

TP: 0.36, NH4-N: 2.0 N removal process Pre-DN Pre-DN Nitrif. + post-DN

Reject water treatment No No DeAmmon®

Return activated sludge

4.4 Similarities and differences 83

The treatment permits 4.4.1Länsstyrelsen (the county administration) is responsible for issuing and su-pervising permits according to Miljöbalken (the Environmental Code). The present permits are similar for the three plants (Table 4.1) and have been issued 2000 (Henriksdal), 2003 (Käppala) and 2012 (Himmerfjärden). In 2012 the effluent limit of total nitrogen was decreased for Himmerfjärden WWTP from 10 to 8 mg/l. All the plants are facing new and more stringent effluent criteria from 2016 to 2019. The exact criteria are not decided by the authorities yet but the plants expect decreased limits on total nitrogen, BOD and phosphorous.

Not all limits are limit values – some are target values. There is also a limit on the amount of discharged nitrogen in tons, but the concentration limits are more challenging to reach.

The control programme for assessing the compliance to the permits con-sists of self-monitored sampling. Composite weekly samples are taken on the effluent water and averaged to assess the compliance. Sludge quality and discharges to air are also included in the self-monitoring programmes. The self-monitoring programmes should be organised according to directive 1998:901.

The loads and treatment capacity 4.4.2Henriksdal and Käppala WWTPs have similar nitrogen load to aerobic vol-ume ratios, 0.12 kgTN/m3,d and 0.09 kgTN/m3,d respectively, while Him-merfjärden has a higher ratio of 0.17 kgTN/m3,d. Looking at the concentra-tions entering the biological treatment the load is 0.11 kgTN/m3, 0.08 kgTN/m3,d and 0.14 kgTN/m3,d for Henriksdal, Käppala and Him-merfjärden WWTPs, respectively. This means that Himmerfjärden WWTP has the smallest volumes for performing nitrogen removal. This is partly compensated for by a longer sludge age at Himmerfjärden WWTP (Table 4.1).

The three plants have different locations in relation to the city and the ur-ban areas (Figure 4.10). The different positions and lengths of the tunnel systems generate different inflow profiles (Figure 4.11).


Figure 4.10. The location of the three WWTPs. All plants have their outlet points in a basin which is part of the Baltic Sea. Lake Mälaren is found west of the centre of Stockholm, and is the capital’s main source of fresh water.

Figure 4.11. Normalised dry weather inflow at Henriksdal, Käppala and Him-merfjärden WWTPs (hourly data).

The largest variation in inflow and the earliest peak load is found at Hen-riksdal WWTP which is situated closer to the city centre. At Himmerfjärden WWTP the distance from the city results in the latest peak load around 8 to 10 hours after the water left the households. Käppala WWTP has the lowest variation in inflow of the three plants.

The inflow profile, the available plant capacity and treatment process de-termines the profile of the effluent nitrogen concentrations. When working with aeration control, this profile will determine how the control problem should be formulated and what process-related constraints that has to be taken into consideration. The dry and wet weather ammonium removal is described in Table 4.2. More information about the ammonium profiles is found in Appendix A.

0 5 10 15 200.4

0.5

0.6

0.7

0.8

0.9

1

1.1

1.2

1.3

Hours

Nor

mal

ised

Qin

Henriksdal WWTP

Käppala WWTP

Himmerfjärden WWTP

Source: Google Maps


Table 4.2. Ammonium removal at Henriksdal, Käppala and Himmerfjärden WWTPs.


Dry weather conditions

During most of the year the influent variation propagates to the effluent, and effluent ammonium varies from zero to 2 to 4 mg/l. During summer ntirification is often complete.

Most of the year Käppala reaches complete nitrifica-tion, and only occa-sionally during au-tumn and winter has daily effluent varia-tions. Complete nitrification during summer.

Daily ammonium variations from zero up to 10 mg/l. Lower or no ammonium in the effluent during summer.

Wet weather conditions

Ammonium peak at most rain events and snow melting.



To summarise, Henriksdal WWTP is the plant with the most periodic treat-ment performance while Käppala’s removal pattern is more of the on-off type. Himmerfjärden is similar to Henriksdal but with a less regular pattern and with larger variations in the effluent concentration.

Control systems, control structure and control strategies 4.4.3Henriksdal and Käppala WWTPs had the industrial control system ABB 800xA at the start of the project. In 2011 Henriksdal changed to Siemens Simatics PCS 7. The control systems make use of industrial PID controllers integrated into the control system. The PID controllers have standard fea-tures including bumpless transfer when switching control modes and anti-windup.

At Himmerfjärden WWTP a SCADA system (Citect) is used for process supervision. Process control is performed via stand-alone PID controllers from Beijer Electronics which communicate with the SCADA system. The controllers have basic anti-windup where integration is interrupted when the control signal saturates.

At the start of the project Henriksdal WWTP had constant DO control (Figure 4.12). Käppala WWTP operated with constant DO control and am-monium control in some treatment lines and another control strategy creating a slope in air flow rate from the first to the last aerated zone depending on two DO sensors (Figure 4.13). This control strategy, known as the “stair strategy”, provides much air in the beginning and little towards the end of the treatment line. This leads to costly aeration and DO peaks in the last aerated zone. At Himmerfjärden WWTP the air flow rate was controlled via one valve per treatment line, based on PI control of the DO concentration in the second aerated zone, i.e. DO zone control was not used (Figure 4.14).


Figure 4.12. The original control structure at Henriksdal WWTP (only aerobic zones). Constant DO control in three aerated zones. The last aerated zone is manual-ly switched off in periods.

Figure 4.13. The original control structure at Käppala WWTP (only aerated zones). Above: Ammonium feedback control with a DO deviation controller present in parts of the older part of the plant. Below: The “stair” control strategy calculating the air flow rate to the four aerated zones. K is a bias. The third aerated zone is manually switched off in periods.


Figure 4.14. The original control structure at Himmerfjärden WWTP. There is one air flow valve for each treatment line.

Other control loops 4.4.4The two nitrogen removal processes nitrification and denitrification are inti-mately connected since the more nitrification, the higher the denitrification demand. The total performance of nitrification and denitrification gives the effluent total nitrogen concentration. The two control handles for denitrifica-tion control are the internal nitrate recycle flow for plants with predenitrifi-cation and external carbon addition at plants with post-denitrification (see Section 2.3.3). At Henriksdal and Käppala WWTPs there is no automatic control of the internal recycle flow pump, the flow is set at a constant value. At Himmerfjärden WWTP the external carbon dosage to the post denitrifica-tion is controlled automatically.

At the three plants the sludge age is controlled over the year by manually adjusting the return and waste activated sludge flows. The sludge age is longer in the winter than in the summer to compensate for temperature ef-fects on the nitrification rate.

Aeration control is not only about DO control, but also about the control loops managing the blowers and the pressure in the air mains. At Käppala WWTP there is a controller for the most-open-valve technique (see Section 2.4.3) while the pressure is kept constant at Henriksdal and Himmerfjärden WWTPs.

DO set-point

DO cascade control

Qair

DO

PART II

DESIGN OF AMMONIUM FEEDBACK CONTROLLERS

91

5 AMMONIUM PI CONTROL AND THE 5

OPTIMAL DO PROFILE

HE AIM OF this chapter is to illustrate what the daily optimal DO profile looks like in an aerated activated sludge system with three aerated zones, and how this optimal profile relates to what is achiev-

able with ammonium PI control. Optimality is here defined as the minimum air flow rate required to reach a specified daily flow proportional ammonium concentration in the effluent. By simulating ammonium PI controllers with different settings, the chapter addresses how to best use ammonium control in the model under study. To be able to perform the optimisations there are a few limitations to the model. The influent includes daily variations only, and the model does not include denitrification.

Introduction 5.1Optimisation of activated sludge models is often based on steady state mod-els (e.g. De Araujo et al., 2011) or if dynamical optimisation is considered, the aerated period in intermittently aerated processes was previously investi-gated (Balku and Berber, 2006; Fikar et al., 2005). Dynamic optimisation of aeration intensity has been considered in Luo and Biegler (2011). Luo and Biegler (2011) minimise a 24-hour KLa vector in the standard BSM1 where the KLa value in the last aerated basin was manipulated. The cost function was the Aeration Energy (AE) calculation specified by the BSM1 protocol combined with a measure of the effluent ammonium and total nitrogen con-centration compared to the set-point. The result was an hourly KLa profile where KLa was 0 d-1 at low load and had a daily maximum of 250 d-1, hence promoting aeration to be switched on and off during the day. The profile only partly resembles the load profile. Compared to the work presented in this chapter, Luo and Biegler (2011) use a different optimisation procedure

T

92 5 Ammonium PI control and the optimal DO profile

(a primal-dual interior point nonlinear programming algorithm) and a differ-ent process model (the standard BSM1 benchmark set-up where only the KLa in the last aerated zone was controlled). Therefore the results in Luo and Biegler (2011) are limited to the time variation of aeration and does not in-clude the space variation in the different aerated zones.

Simulation model 5.2The simulation model in this study is depicted in Figure 5.1. The model is a modified version of the MATLAB/Simulink® BSM1 implementation. The Benchmark Models do not model air flow rate. Instead, air flow rate is indi-rectly given as KLa, which in a real process is a function of the air flow rate. The evaluation criterion Aeration Energy is calculated in the BSM procedure as a function of KLa in the Benchmark Model evaluation procedure. Since the relationship between KLa and air flow rate is non-linear, air flow rate was modelled in this study by including the air flow model presented by Dold & Fairlamb (2001) in BSM1. More information about this model is found in Section A.2. The default model parameters were used in the air flow model. No zones for denitrification were included to speed up the optimisation pro-cedure. The total aerobic sludge age was around 5.2 days in all simulations. Each aerobic zone had a volume of 2000 m3.

Figure 5.1. Simulation model with supervisory ammonium feedback control with the ammonium sensor in zone 3. Qin = inflow, WAS = waste activated sludge (475 m3/d), RAS = Return activated sludge (0.7Qin), PI = proportional integral con-troller, Qair = air flow rate, SP = set-point.

The influent flow rate, total COD (chemical oxygen demand) and total nitro-gen concentration were created on the basis of the model presented in Lang-ergraber et al. (2008). The fractionation of nitrogen and carbon compounds was performed using factors in the influent generator model published in Gernaey et al. (2011). The load of 70 000 P.E. (population equivalents) rep-

5.3 Control strategies 93

resented the design load to the aerated basin according to design standards (ATV, 2000). The average total COD was 300 mg/l and total Kjeldahl nitro-gen 40 mg/l. All days had an identical influent load, an example representing the design load is presented in Figure 5.2. Simulations were run for 150 days in steady-state before dynamic simulations were performed for 8 weeks. The last day was used in the evaluations.

Figure 5.2. Nominal influent flow rate and ammonium concentration used in the study. The average flow is 14 000 m3.

Control strategies 5.3Ammonium feedback control was implemented and evaluated using differ-ent controller parameters in the PI controller (Eq. (2.1)). Several values on K and Ti were tested in the simulations. The longest integration time used was 4 d, and the shortest 0.01 d. The controller gain ranged from -0.05 to -4. The choice of controller gain varied depending on simulation scenario. The DO controllers were PI controllers (K = 30 000, Ti = 0.01 d) and were the same in all simulation scenarios. The upper limit of the DO set-point was varied to find the value yielding the lowest energy consumption. The lower limit of the DO set-point was fixed at 0.4 mg/l. This was based on the results from the optimal DO profile, see further Section 5.4. An iterative strategy in MATLAB was used to find the ammonium set-point required to reach the desired flow-proportional outlet concentration. All simulations were com-pared to a reference strategy with constant DO control and all simulations reached the same ammonium concentration in the effluent.

The BSM1 PI controllers have anti-windup through back-calculation (Åström and Hägglund, 1995), see Figure 5.3. Back calculation anti-windup has a feedback loop comparing the difference between the calculated con-troller output (v) and the actual actuator output (u) given possible actuator saturation. This error signal (es) is fed back to the integrator multiplied with 1/Tt. es is zero when u = v. When the actuator saturates the controller output

00 06 12 18 24

0.8

1

1.2

1.4

1.6

1.8

2x 10

4

Time (hr)

Flow

rate

(m3 /d

)

Flow rate

35

40

45

50

55

60

65

Am

mon

ium

con

cent

ratio

n (m

g/l)

Ammonium concentration


u is constant, but the back-calculation loop will drive the integrator input towards zero, thereby preventing the integral term from winding up. The tracking time constant Tt determines how fast the integrator input is reset. If Tt is small, integrator reset is fast. In the simulations Tt was chosen to 0.8Ti (Åström and Rundqwist, 1989).

Figure 5.3. Block diagram of the PI controller with back-calculation anti-windup used in the benchmark models.

In Chapter 8, master controller windup in the presence of cascade control without tracking between the controllers is discussed (Section 8.12.1). Mas-ter controller windup arises when the slave controller (e.g. a DO controller) is saturated and the integrator in the master controller (e.g. an ammonium controller) keeps integrating the control error in spite of the fact that the slave controller lost its control authority. The discussion in Chapter 8 refers to master controller windup in the DO controller. This situation cannot ap-pear in the BSM1 model since there is no air flow rate controller. It could however appear in the ammonium controller if the DO controller reaches an air flow rate limit. Since there are no air flow rate limits in the model under study, this is not an issue.

Optimisation problem 5.4The ammonium PI controllers with different settings were compared to the optimal DO profile. The optimal DO profile was created through mathemati-cal optimisation in MATLAB. The objective was to minimise the daily air flow rate, with the constraint that the daily average flow proportional ammo-nium concentration should be less than a certain value – the same value as was achieved in the ammonium PI simulations. The daily air flow vector was optimised in MATLAB using the function fmincon, solving the following optimisation problem, discretised in 36 steps per day:

5.5 Simulation scenarios 95

Min = =

36

1

3

1

)(j k

k jQair (5.1)

s.t. 0

)()(36

14

≤−

= αmean

j

Q

jQinjNH (5.2)

,0)( ≥jQair k j=1, ..., 36, k= 1, 2, 3 (5.3)

where Qairk, is the air flow rates in zone 1 to 3 respectively, NH4 is the am-monium concentration in the outlet from zone 3, Qin is the flow rate to the plant, Qmean is the average daily flow rate and α is the desired daily flow proportional ammonium concentration. The initial guess of the air flow rate is a vector with 108 variables, representing the air flow rate in the three aer-obic zones from the reference simulation with constant DO control.

Simulation scenarios 5.5The simulations performed in the study are presented in Table 5.1. All sce-narios are compared to a reference control strategy of constant DO control. For all scenarios there was a reference simulation, simulations with ammo-nium feedback controllers (Section 5.3) and an optimal solution (Section 5.4). The daily effluent flow proportional concentration of ammonium (α) in the reference strategy with constant DO control determined the ammonium concentration that all other simulations for that scenario must achieve.

Scenario 1 was the standard scenario with a constant DO concentration in the reference strategy of 1.5 mg/l, design load and the default BSM1 model parameters. Scenario 2 and 3 had the same reference strategy and optimal DO profile as scenario 1, but the ammonium sensor used for ammonium control was placed in zone 2 for scenario 2 and zone 1 for scenario 3.

In scenario 4 and 5 the level of the load to the plant was changed and in scenario 6 and 7 the amplitude of the influent was varied. A description of the influents for scenario 4 to 7 is given in Table 5.2.

The DO concentrations in the reference strategy were changed in scenario 8 and 9 which led to different levels of α in these scenarios compared to α in scenario1. Thereby, the ammonium set-point in the PI controllers was differ-ent in scenario 1, 8 and 9.

The ammonium profiles in the last zone for scenarios 1 to 9 are given in Figure 5.4. Despite a different variation in load in scenario 6 and 7 the am-monium profile is similar to scenario 1.


Table 5.1. Simulation scenarios. The changes made compared to scenario 1 in bold. The DO set-point refers to the reference strategy simulation with constant DO con-centrations. D = design, L = low, H = high, SP = set-point, FB = feedback.

Scenario No. 1 2 3 4 5 6 7 8 9 10

Mean load D D D L H D D D D D

Load amplitude D D D D D L H D D D DO SP (mg/l) 1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.0 2.0 1.5 NH4 FB zone 1 x NH4 FB zone 2 x NH4 FB zone 3 x x x x x x x x Default model x x x x x x x x x Changed model x

Table 5.2. The load settings in scenario 1, 4 to 5 (different mean load) and 6 to 7 (different influent amplitude). This table explains the notation D, H and L in Table 5.1.

Scenario No. 1 4 5 6 7

Mean load (P.E.) 70 000 50 000 90 000 70 000 70 000

Mean inflow (m3/d) 14 000 10 000 18 000 14 000 14 000 Maximum inflow (m3/d) 18 064 13 040 23 032 16 976 19 151 Minimum inflow (m3/d) 8 273 5 796 10 795 9 059 7 348

(a) (b) (c)

Figure 5.4. Ammonium profiles of the reference simulation (constant DO control) for (a) scenario 1, 4 and 5, (b) scenario 1, 6 and 7 and (c) scenario 1, 8 and 9. Sce-nario 2 and 3 have the same profile as scenario 1.

Finally, scenario 10 considered the case of changed model parameters in the equation for aerobic growth of autotrophs in ASM1 (Eq. (2.5)). In this sce-nario, the oxygen half-saturation constant for autotrophs (KO2,ANO) and the maximum growth rate for autotrophs (µANO,Max) were changed from 0.4 mg/l to 0.8 mg/l and from 0.5 d-1 to 0.6 d-1, respectively. This change yields ap-proximately the same value of the DO Monod function in Eq. (2.5) at 1.5 mg/l, but the slope of the Monod curve around 1.5 mg/l is different (Figure 5.5). The other model parameters have the default BSM1 values.

00 06 12 18 240

1

2

3

4

Effl

uent

am

mon

ium

(m

g/l)

Hour

00 06 12 18 240

1

2

3

4

Effl

uent

am

mon

ium

(m

g/l)

Hour

00 06 12 18 240

1

2

3

4

Effl

uent

am

mon

ium

(m

g/l)

Hour

Scenario 4Scenario 5Scenario 1



5.6 The optimal DO profile 97

Figure 5.5. The DO Monod function at different DO concentrations multiplied with the maximum specific growth rate for autotrophs (µANO,Max) for model settings in scenario 1 to 9 (bold black) and in scenario 10 (black).

The optimal DO profile 5.6Figure 5.6 shows the daily variation in DO and outlet ammonium for the optimal solution of scenario 1. The average effluent ammonium concentra-tion is 1.02 mg/l. The peak-to-peak amplitudes in DO concentration are 1.27 mg/l, 1.60 mg/l and 1.37 mg/l in zones one, two and three, respectively. The peak ammonium concentration in zone 3 is 2.12 mg/l.

The optimal DO profile has a lower DO concentration in the first aerobic zone compared to the second and third zones (Figure 5.6). Because the ammonium concentration is much higher in this zone, the overall removal rate is still the highest in the first zone since the product of the Monod functions of ammonium and oxygen in Eq. (2.5) is high. By reducing the DO concentration in the first zone, a larger part of the ammonium removal takes place towards the two last zones, compared to the case with constant DO in all zones.

The peak DO concentration arrives first in zone 1 and last in zone 3, which follows the same trend as the peak ammonium concentrations in each zone. However, the peak DO concentration arrives before the peak ammonium concentration. This implies that it is beneficial to react early to a load disturbance.

0 0.5 1 1.5 2 2.5 3 3.5 40

0.1

0.2

0.3

0.4

0.5

DO concentration (mg/l)

DO

Mon

od fu

nctio

n *

μA

KO2,ANO

= 0.4, μANO,Max

= 0.5

KO2,ANO

= 0.8, μANO,Max

= 0.6


Figure 5.6. DO and ammonium profiles in zone 1 to 3 for the optimal solution (sce-nario 1).

Comparison to ammonium feedback control 5.7The ammonium controllers in scenario 1 are compared to the reference strat-egy of constant DO and the optimal solution in Figure 5.7. The air flow rate is weighted with the air flow rate with constant DO control and plotted as a function of the variance of the DO concentration. The DO variance is an indirect measure of the speed of the controller. Figure 5.7(a) shows the simu-lation results without an upper DO set-point limit in the ammonium control-ler while Figure 5.7(b) has an upper DO limit of 1.8 mg/l. Figure 5.7 clearly shows the importance of having an upper DO limit in the controller.

The lowest energy requirement for each choice of controller gain always corresponds to having the longest possible integration time (Ti = 4 d in this study). The simulations with varying K but fixed Ti = 4 d are shown in Fig-ure 5.8. When the controller gain is increased the energy consumption is first reduced until a minimum is reached. After the minimum, the energy re-quirement increases for increased values on K. If the DO set-point is not limited (Figure 5.7(a)), the energy consumption will be higher than that of constant DO control for controller gains below -1. In these simulations, the maximum DO concentration reaches above 2.5 mg/l.

With an upper DO set-point limit in the ammonium controller of 1.8 mg/l there is a minimum air flow consumption achieved with ammonium feed-back control at K = -1.25, Ti = 4 d. When using a maximum DO set-point limit a faster controller can be used. Again, the lowest air flow requirement

00 06 12 18 24

0

0.5

1

1.5

2

2.5

DO

(m

g/l)

Simulation time (hr)

00 06 12 18 24

0

5

10

15

NH

4 (m

g/l)


Zone 1

Zone 2

Zone 3

Zone 1

Zone 2

Zone 3

5.7 Comparison to ammonium feedback control 99

for each choice of ammonium controller gain always corresponds to having the longest possible integration time. If using a proportional controller only, there is approximately 0.1 % more to gain with respect to energy consump-tion, compared to using Ti = 4 d.

When the integral time is increased from Ti = 4 d for a fixed value on the controller gain the energy consumption is increased. Integral action adds a delay to the DO profile compared to when only proportional control is used. When K < -0.7 there is a decrease in the energy consumption for very short integral times (Ti = 0.03 d or Ti = 0.01 d). This phenomenon occurs when the DO concentration hits the lower DO limit of 0.4 mg/l.

A summary of the results from scenario 1 to 10 is presented in Table 5.3. The table includes information about the daily ammonium concentration (α), air flow rate for the reference strategy (constant DO), controller settings in the best ammonium controller and the energy saving for the best ammonium controller and for the optimal solution, compared to constant DO control.

In the ammonium PI simulations, four out of five possible controller pa-rameters in the PI controller were changed in order to find the best settings: K, Ti, upper DO set-point limit and ammonium set-point. The lower DO set-point limit was fixed. Extensive simulations have shown that by changing all five settings in the controller, near identical results can be achieved by dif-ferent combinations of settings. For example, the best controller for Scenario 1 had K = -1.25 for an ammonium set-point of 0.54 mg/l when the upper and lower DO set-point limits are 1.8 and 0.4 mg/l, respectively. Similar results in terms of energy consumption can be achieved with K= -1.05, ammonium set-point = 0.69 with upper and lower DO limits of 1.83 and 0.96 mg/l, re-spectively. Hence, the simulations results are dependent on the constraint that the lower DO set-point limit was fixed.


(a) (b)

Figure 5.7. Relative comparison between air flow demand for constant DO control, NH4 feedback controllers with different settings (K = [-0.1 -2.5], Ti = [0.01 4] d) and optimal air flow rate. (a) No maximum DO limit and (b) Maximum DO = 1.8 mg/l.

Figure 5.8. Relative comparison between air flow demand for constant DO control, NH4 feedback controllers with different settings (K = [-0.1 -2.5], Ti =4 d) and opti-mal air flow rate.

0 0.5 1 1.5 2 2.5 3 3.5 40.9

1

1.1

1.2

1.3

1.4

1.5

1.6

1.7

1.8

Variance of DO (mg/l)2

Air

flow

rat

e co

mpa

red

to c

onst

ant D

O

Surface ammonium feedback controller

Constant DO=1.5 mg/l

Optimal solution

Simulations with different K, varying Ti

0 0.05 0.1 0.15 0.2 0.25 0.3 0.350.94

0.95

0.96

0.97

0.98

0.99

1

1.01

1.02

1.03

Variance of DO (mg/l)2

Air

flow

rat

e co

mpa

red

to c

onst

ant D

O

Surface ammonium feedback controller

Constant DO=1.5 mg/l

Optimal solution

Simulations with different K, varying Ti

0 0.2 0.4 0.6 0.8 1

0.94

0.96

0.98

1

1.02

1.04

1.06

Variance of DO concentration (mg/l)2

Air

flow

rat

e co

mpa

red

to c

onst

ant D

O

NH4 feedback, no DO limit

NH4 feedback, max DO=1.8 mg/l

Constant DO (1.5 mg/l)

Optimal DO profile

K=−0.1

K=−2.5

K=−1.5

Tab

le 5

.3. S

umm

ary

of s

imul

atio

n re

sults

. The

am

mon

ium

con

cent

ratio

n (α

) is

the

daily

ave

rage

flo

w p

ropo

rtio

nal c

once

ntra

tion

in th

e ae

ra-

tion

tank

out

let.

The

bes

t con

trol

ler

refe

rs to

the

amm

oniu

m c

ontr

olle

r w

ith th

e lo

wes

t ave

rage

air

flo

w r

ate.

SP

= s

et-p

oint

.

Sce

nari

o N

o.

1 2

3 4

5 6

7 8

9 10

α (N

H4

conc

entr

atio

n, m

g/l)

1.

021.

021.

020.

791.

201.

00

1.04

1.67

0.

78

1.06

A

ir f

low

rat

e co

nsta

nt D

O (

m3 /d

) 59

451

5945

159

451

3992

179

986

5943

7 59

414

5227

3 66

791

5940

8 m

3 air

/kg

NH

4 re

mov

ed c

onst

ant D

O

109

109

109

102

115

152

85

97

122

114

NH

4 se

t-po

int b

est c

ontr

olle

r 0.

542.

647.

030.

510.

590.

51

0.61

1.48

0.

39

0.56

K

bes

t con

trol

ler

-1.2

5-0

.25

-0.1

5-1

.25

-1.2

5-1

.5

-0.9

-0.2

-2

.0

-2.4

M

ax D

O s

et-p

oint

bes

t con

trol

ler

(mg/

l)

1.8

1.8

1.8

1.8

1.8

1.8

1.8

1.4

2.3

1.8

Ene

rgy

savi

ng b

est c

ontr

olle

r (%

) -2

.5-2

.8-3

.5-1

.9-2

.8-2

.8

-2.1

-1.2

-3

.3

-2.4

E

nerg

y sa

ving

opt

imal

sol

utio

n (%

) -5

.2-5

.2-5

.2-4

.9-5

.3-5

.3

-5.1

-3.3

-6

.9

-4.4


Comparisons between the energy consumption of ammonium feedback control, the optimal solution and constant DO control for scenarios 1 to 9 are given in Figure 5.9. The DO profile of the optimal solutions are compared to the best ammonium feedback controller in Figure 5.10 to Figure 5.13.

The pattern in Figure 5.9 is similar for all scenarios with a minimum point where the energy consumption is the lowest for the removed amount of ammonium. The optimal solutions are approximately the same for all scenarios apart from those where the DO level is changed. Varying the load either in terms of changing the average load or changing the influent variability does not impact the optimal solution to a large extent (Figure 5.9(b), (c)). If the DO concentration operated around 1.5 mg/l the DO profile should, optimally, be the same. Even though the scenarios with different mean loads result in different ammonium concentrations in the effluent, the optimal profile is approximately the same.

If the average DO concentration in the reference simulation is changed, so is the optimal solution (Figure 5.9(d)). When operating at an in average higher DO concentration there is more to gain by varying the DO concentration, compared to keeping the concentration constant.

In all four plots in Figure 5.9 there is an impact on the potential energy saving with ammonium PI control when changes are made to scenario 1. Figure 5.9(a) shows that there is energy to gain from moving the ammonium sensor from the first to the last aerated zone. By doing so, the ammonium controller becomes a combined ammonium feedback-feedforward controller, but the ammonium set-point is harder to determine. The controller is feedback with respect to the zone the sensor is placed, and feedforward with respect to the subsequent zones. The DO profile from ammonium control approaches that of the optimal controller when the sensor is moved earlier along the basin (Figure 5.10). The required controller gain is, as expected, lower when the sensor measures higher ammonium concentrations earlier in the process (Table 5.3). The influent amplitude has an impact on the best choice of controller gain. At higher influent variations the best controller requires a lower gain and when the variation is lower, the gain is lower (Ta-ble 5.3).

As discussed the impact on the optimal solution is small when changes are made to the load, but changes do have an impact on the potential energy saving from using ammonium control (Figure 5.9(b), (c)). If the load is higher or the influent variation lower the best performing ammonium feedback DO controller reaches closer to the optimal solution (Figure 5.11, Figure 5.12).


(a) (b)

(c) (d)

Figure 5.9. Comparison between air flow consumption for NH4 feedback control-lers, constant DO control and the optimal solution. (a) Scenario 1 to 3, (b) scenario 1, 4 and 5, (c) scenario 1, 6 and 7 and (d) scenario 1, 8 and 9.

0 0.05 0.1 0.15 0.2 0.250.94

0.95

0.96

0.97

0.98

0.99

1


Air

flow

rat

e co

mpa

red

to c

onst

ant D

O

NH4 FB from zone 3

NH4 FB from zone 2

NH4 FB from zone 1


Optimal solution

0 0.05 0.1 0.15 0.2 0.250.94

0.95

0.96

0.97

0.98

0.99

1


Air

flow

rat

e co

mpa

red

to c

onst

ant D

O

NH4 FB, 50 000 P.E.

NH4 FB 70 000 P.E.

NH4 FB, 90 000 P.E.


Optimal solution 50 000 P.E.



0 0.05 0.1 0.15 0.2 0.250.93

0.94

0.95

0.96

0.97

0.98

0.99

1


Air

flow

rat

e co

mpa

red

to c

onst

ant D

O

0 0.05 0.1 0.15 0.2 0.250.94

0.95

0.96

0.97

0.98

0.99

1


Air

flow

rat

e co

mpa

red

to c

onst

ant D

O

NH4 FB, DO level=1.0 mg/l



Constant DO (1.0/1.5/2.0 mg/l)

Optimal solution, DO level=1.0 mg/l



NH4 FB low influent amplitude

NH4 FB design influent amplitude

NH4 FB, high influent amplitude


Optimal solution low amplitude

Optimal solution design amplitude

Optimal solution high amplitude


(a) (b) (c)

Figure 5.10. DO profiles for the optimal solution and the best NH4 feedback control-ler measuring NH4 in (a) zone 1, (b) zone 2 and (c) zone 3. FB = feedback. Note: the optimal solutions in the three plots are the same.

(a) (b) (c)

Figure 5.11. DO profiles for the optimal solution and the best NH4 feedback control-ler for (a) 50 000 P.E., (b) 70 000 P.E. and (c) 90 000 P.E. FB = feedback.

00 06 12 18 240

0.5

1

1.5

2

2.5


00 06 12 18 240

0.5

1

1.5

2

2.5


00 06 12 18 240

0.5

1

1.5

2

2.5


DO

(m

g/l)

Optimal DO profile zone 1



NH4 FB zone 1 (K=−0.15, Ti=4 d)




NH4 FB zone 3 (K=−1.25, Ti=4 d)




NH4 FB zone 2 (K=−0.25, Ti=4 d)

NH4 sensor zone 1 NH4 sensor zone 2 NH4 sensor zone 3

00 06 12 18 240

0.5

1

1.5

2

2.5


DO

(m

g/l)

00 06 12 18 240

0.5

1

1.5

2

2.5


00 06 12 18 240

0.5

1

1.5

2

2.5





NH4 FB (K=−1.25, Ti=4 d)




NH4 FB (K=−1.25, Ti=4 d)




NH4 FB (K=−1.25, Ti=4 d)

50 000 P.E. 90 000 P.E.70 000 P.E.


(a) (b) (c)

Figure 5.12. DO profiles for the optimal solution and the best NH4 feedback control-ler for (a) low influent amplitude, (b) design influent amplitude and (c) high influent amplitude. FB = feedback.

(a) (b) (c)

Figure 5.13. DO profiles for the optimal solution and the best NH4 feedback control-ler with different DO concentrations in the reference strategy: (a) 1 mg/l, (b) 1.5 mg/l and (c) 2 mg/l. FB = feedback. Note: the y-axis scale is different in the three plots.

00 06 12 18 240.5

1

1.5

2

2.5


DO

(m

g/l)

00 06 12 18 240

0.5

1

1.5

2

2.5


00 06 12 18 240

0.5

1

1.5

2

2.5





NH4 FB (K=−1.25, Ti=4 d)




NH4 FB (K=−0.9, Ti=4 d)




NH4 FB (K=−1.5, Ti=4 d)

Design load amplitude High load amplitudeLow load amplitude

00 06 12 18 240

0.5

1

1.5


DO

(m

g/l)

00 06 12 18 240

0.5

1

1.5

2

2.5


00 06 12 18 240

0.5

1

1.5

2

2.5

3





NH4 FB (K=−1.25, Ti=4 d)




NH4 FB (K=−2, Ti=4 d)




NH4 FB (K=−0.25, Ti= 4 d)

DO level 1 mg/l DO level 1.5 mg/l DO level 2 mg/l


The key to how much there is to save from varying the DO concentration is summarised in Figure 5.14. The larger the variation in DO in the controller, the more there is to gain compared to keeping the DO concentration constant (Figure 5.14(a)). A large variation in DO concentration is achieved when there is a large difference between the actual daily ammonium concentration (α) and the ammonium set-point (Figure 5.14(b)). In the simulations where the required ammonium set-point was low compared to the average concentration in the effluent, the energy saving due to ammonium control was high. The required ammonium set-point in the ammonium controller was decided through an iterative procedure in MATLAB where the set-point was changed to achieve the same ammonium concentration in the effluent as in the reference simulation with constant DO control.

(a) (b)

Figure 5.14. The energy saving with the best ammonium controller in scenario 1 and 4 to 9. (a) In relation to the variance of the DO concentration and (b) in relation to α divided by the ammonium set-point.

The optimal DO profile in the three aerobic zones for changed values on the nitrification rate parameters are found in Figure 5.13. The optimal DO profile has a smaller variation in scenario 10 compared to scenario 1.

Figure 5.15. Optimal DO profile for default benchmark settings (scenario 1, KO2,ANO = 0.4 mg/l, µANO,Max = 0.5 d-1) and for changed settings (scenario 10, KO2,ANO = 0.8 mg/l, µANO,Max = 0.6 d-1).

0 0.1 0.2 0.31

1.5

2

2.5

3

3.5


Ene

rgy

savi

ng (

%)

1 1.5 2 2.51

1.5

2

2.5

3

3.5

α/NH4 set−point

Ene

rgy

savi

ng (

%)

00 06 12 18 240

0.5

1

1.5

2

2.5

DO

(m

g/l)


DO zone 1, scenario 10






5.8 Discussion 107

Discussion 5.8When operating at higher DO concentrations there is a smaller impact on the nitrifier growth rate when the DO concentration is changed than at lower DO conentrations. This is because the plant operates in the more saturated region of the DO Monod function in Figure 5.5. Therefore a larger variation in DO concentration can be allowed without affecting the nitrification rate much. This allows the DO concentration to be reduced more compared to when the average DO concentration is lower, and the saving compared to constant DO control will be larger.

The impact of the slope in the Monod function was also studied in scenario 10 where the model parameters in the Monod function were changed. In scenario 10 the oxygen half saturation constant is increased, meaning the slope in Monod function was higher around the DO operating point of 1.5 mg/l (Figure 5.5). In scenario 8 the variation in DO concentration was less than in scenario 1 and the advantage from varying the DO concentration compared to having a constant DO was smaller in scenario 10.

If the plant due to the load situation, the plant design and/or the effluent limits needs to operate at higher DO concentrations, it is more worthwhile to implement control strategies which vary the DO concentration based on the load or the treatment performance. As expected, it is more costly to operate at higher DO concentration when considering the air flow rate per removed ammount of ammonium (Table 5.3).

Constant DO control is only 1 to 3.5 % more costly than ammonium feedback control in this study. However, ammonium feedback control can be a method to counteract non-periodic disturbances and it adds robustness to the operation compared to using constant DO concentrations. Furthermore, with constant DO control, it can be difficult to set the DO set-point a priori.

This study has solely targeted energy efficiency together with a daily treatment performance of ammonium assuming a 24 h periodic influent. In reality, there are several more aspects which needs to be considered when implementing ammonium feedback control. As mentioned in Section 2.4.1, there is a risk of triggering nitrous oxide emissions at low DO concentrations (Kampschreur et al., 2009). It has been suggested that the DO concentrations in the first aerobic zone should not be below 2.0 mg/l to avoid emissions of nitrous oxide emissions (Lotito et al., 2012). These results are in conflict with the optimal DO profile in this study which in all scenarios has a lower DO concentration in the first aerobic zone. Nitrous oxide formation as well as sludge quality aspects (Section 2.4.1) need to be adressed when ammonium feedback control is implemented, and could motivate a higher DO set-point limit in the controller.

The optimisation does not take the instantaneous treatment performance into consideration, since the constraint on the ammonium concentration is


based on a daily average concentration. This makes the results relevant for countries or regions which do not use grab sampling when evaluating compliance to the effluent criteria. The cost function is also limited to only looking at aeration costs rather than treatment performance. Even though the daily ammonium concentration is a constraint, there are several other aspects which the optimisation procedure does not take into account, such as the size of the control signal and other limitation such as limits in air flow rate.

The simulations in this study are not limited by sensor data quality. In a real process sensors are noisy, could be ill-placed in the basin or give the wrong value. Having access to the same information about the ammonium and DO concentrations in a real process as in this simulation study is not possible, and it is not possible to create an optimal DO profile. The results in this study should therefore be seen as a theoretical guideline on how to reason about the DO profile and the application of ammonium feedback control.

Conclusions 5.9This chapter compares ammonium PI controllers to an optimal DO profile in a model with three aerated zones and daily influent variations. Compared to constant DO control a variation in DO concentration over the day is more economical. The optimal DO profile was around 3 to 7 % more energy efficient than constant DO control. The DO concentration was lower in the first aerobic zone compared to the two last zones and resembles the ammonium concentration profile, but the peak DO concentrations arrived before the ammonium peak. This motivates feedforward control. The optimal DO profile reached a similar energy saving irrespective of changes to the influent load as long as the level of the DO concentration in the reference simulations was the same. If the load was kept the same but the ammonium set-point was changed, this had an impact on the saving achieved with the optimal solution. At higher DO concentrations the DO Monod curve was more saturated and there was more to gain from varying the DO concentration. The best ammonium PI controller could reach close to optimal performance, and was approximately 1 to 3.5 % more energy efficient than constant DO control. It is important to limit the maximum DO concentration in the ammonium controller. The ammonium controller with less energy demand is a controller with a long integral time.

5.9 Conclusions 109

There was more to gain from ammonium feedback control if the controller could be allowed to generate a large variation in DO concentration. The variation in DO concentration was large when the difference between the mean effluent ammonium concentration and the ammonium set-point was large.

111

6 AMMONIUM PI CONTROLLER DESIGN – 6

BENCHMARK SIMULATIONS

MMONIUM FEEDBACK CONTROL has been proven to save aeration energy in the range of 5 to 25 % according to the literature review in Chapter 3. However, when considering only daily varia-

tions in the influent, the optimal performance was just up to 7 % less costly than constant DO control, as discussed in Chapter 5. Through long-term simulations in one of the benchmark models, this chapter provides insights into how to design cost-effective ammonium PI controllers. Three aspects of ammonium PI design have special attention: The PI controller settings (K, Ti), the ammonium set-point and the upper DO set-point limit.

6.1 Introduction In ammonium control the process dynamics is slow. It can take several hours before the ammonium concentration settles after a step change in the DO set-point. Before the ammonium concentration settles, the process will be in a different state caused by the change in influent load. This makes step-response tests for the purpose of tuning ammonium PI controllers difficult, which makes tuning based on step-response tests or similar methods prob-lematic.

Since the load disturbance to the treatment plant is continuously changing the ammonium concentration is rarely at its set-point during ammonium feedback control. Normally, the ammonium controller does not have enough control authority to change the aeration intensity to balance the load disturb-ance. At low load, nitrification is often complete and ammonium in effluent is close to zero, no matter if the DO set-point is reduced. At high load on the other hand, the ammonium controller quickly lose control authority since

A

112 6 Ammonium PI controller design – benchmark simulations

increasing an already high DO concentration have a limited effect on nitrifi-cation. If a plant was to always achieve near complete nitrification, the re-quired treatment volumes would have to be enormous. In this respect, the plant is limited by capacity in terms of mass of microorganisms, rather than limited by aeration intensity, as discussed in Section 3.7.

As discussed by Ingildsen (2002), the effluent permit of a WWTP is de-fined differently in different countries. The time frame of sampling, the in-clusion of extreme events or not and the compliance assessment method (e.g. type of average calculation) can vary. If it is important to never exceed the effluent limit, savings can be achieved through variance reduction. If the permit is assessed based on relatively long averages – which is so far the case for Sweden – variance reduction is less important but the control goal should be formulated to achieve the required effluent criteria over time. In the case of never-to-exceed limits, integration of the control error or similar design specifications could be useful to evaluate controller performance, while in the case of long averages when assessing the compliance to criteria these specifications would not capture the goal of the controller.

Due to the present legislation in Sweden, the point of view in this and the following chapter is that the effluent criteria should be achieved as an aver-age over time – hence instantaneous disturbance rejection is not a priority and commonly used specifications for disturbance rejection control are not suitable to evaluate controller performance. Also, the ammonium removal process is slow enough not to allow for easy access to the use of tuning rules or autotuning, which leaves manual tuning based on process knowledge as the most attractive alternative.

The effluent ammonium concentration is one of the key quality parame-ters at a WWTP, but ammonium removal is achieved through the single largest energy consuming process in the WWTP. The purpose of the ammo-nium controller is therefore to perform cost-effective ammonium removal. In this context, Chapter 6 and 7 look at how the design of ammonium PI con-trollers influence the performance of the nitrification process both in terms of treatment results and energy consumption. Simulations are performed over a large range of controller settings, and ammonium PI control is com-pared to constant DO control which is a well-established and commonly used control strategy in many WWTPs.

6.2 Model setup Simulations were performed in the BSM1 extension BSM1_LT. LT stands for long-term, and the BSM1_LT model is similar to BSM2 but only encom-passes the activated sludge process (Gernaey et al., 2014). Unlike BSM1, BSM1_LT and BSM2 are simulated for a full year instead of 14 days and temperature effects on the biological processes are included. There is also

6.2 Model setup 113

the possibility to model nitrifier inhibition, but this was not included in this work.

The simulation protocol includes simulating the model for 609 days, where days 245 to 609 are used for evaluation. A summary of the model setup and influent characteristics is given in Table 6.1. The zone volume was 1333 m3 in the default settings which was increased to 2000 m3 to achieve a more reasonable load situation. This is a deviation from the BSM1_LT pro-tocol.

Table 6.1. Model setup and average influent concentrations for BSM1_LT. TKN= Total Kjeldahl Nitrogen.

Model and influent setup

Anoxic volumes (m3) 3000

Aerobic volumes (m3) 6000

Average flow (m3/d) 20 514

TKN (mg/l) 41.7

NH4 (mg/l) 23.1

COD (mg/l) 350

Average temperature (°C) 14.3

Aerobic sludge age (d) 7.0

BSM1_LT includes actuator and sensor models. The time delay models for the actuators and sensors were used, but not the noise in the sensors. The air flow model by Dold & Fairlamb (2001) was implemented in BSM1_LT, as in Chapter 5. More information about this model is found in Appendix A. The default model parameters were used in the air flow model.

An example of the effluent ammonium profile in the model using a con-stant DO set-point is found in Figure 6.1. Since the ammonium concentration is the controlled variable the concentration profile is important for how the control signal output is calculated.

Figure 6.1. Effluent ammonium concentration with constant DO set-point (2 mg/l).

Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun0

2

4

6

8

10

12

NH

4 (m

g/l)


6.3 Simulation settings 6.3.1 Control structure The default BSM1_LT control structure includes constant DO control through manipulation of KLa in the last aerobic zone as well as optional ni-trate control through manipulation of the internal recycle flow based on a nitrate sensor in the last anoxic zone. In this study DO control was extended to all three aerated zones, the model setup and aeration control structure for this simulation study are depicted in Figure 6.2. The internal recirculation had a fixed value three times the average inflow. The RAS and WAS flows are the same as in the default BSM1_LT settings, with an average flow rate of 14 600 m3/d and 350 m3/d respectively. The RAS flow is constant, while the WAS flow is changing to have a variation in sludge age over the year to compensate for temperature effects on the nitrogen removal rate. As in Chapter 5 the controllers have back-calculation anti-windup with Tt = 0.8Ti

(see Figure 5.3). There are no air flow rate limits in the model in order to look at the behaviour of the ammonium controllers when aeration intensity was not limited. Aeration limits are considered in Chapter 7. The DO con-troller was tuned to be fast (K = 15 000, Ti = 0.005 d).

Figure 6.2. BSM1_LT model setup and aeration control structure. SP = set-point, Qr = nitrate return, RAS = return activated sludge, WAS = waste activated sludge.

6.3.2 Controller settings Three sets of simulations were performed. For each set of simulations the PI controller settings (K, Ti) were varied to find the best performing controller. The first set of simulations investigated the combined effect of the choice of ammonium set-point and the choice of the upper DO set-point limit. The second set of simulations looked at the effect of the ammonium set-point in combination with a varying aerobic treatment volume in the model. The third set of simulations looked at how the upper DO set-point limit impacts the results, also for a varying treatment volume.

6.3 Simulation settings 115

For the first set of simulations, each scenario had a certain ammonium set-point and certain control signal limits (Table 6.2). For each scenario, simulations were performed with a broad range of PI controller settings, presented in Table 6.3. All the listed integral times were simulated for each choice of controller gain in Table 6.3 The method of performing a number of simulations with different controller settings is similar to using tuning maps, as described in Åström and Hägglund (2006). Tuning maps can be used to look at how the controller settings impact the response to a load disturbance or the shape of the Nyquist plot.

Table 6.2. Controller settings in the first set of simulations: different NH4 set-points and DO set-point limits.

NH4 set-point (mg/l) Min/max DO set-point (mg/l)

Scenario 1 2 3 1/2 1/2.5 1/3

1 x x

2 x x

3 x x

4 x x

5 x x

6 x x

7 x x

8 x x

9 x x

Table 6.3. PI controller settings simulated for each scenario in Table 6.2.

Controller gain (K) -0.01 -0.025 -0.05 -0.075 -0.1 -0.15 -0.25 -0.5 -0.75 -1 Integral time, d (Ti) 0.01 0.02 0.05 0.075 0.1 0.2 0.5 1 2

The second set of simulations was performed with a broad range of ammo-nium set-points for three aerobic treatment volumes. The influent was the same for all simulations but the aerobic zone volume was either 2000 m3 (same as in Table 6.2), 2400 m3 or 1800 m3. The volumes were chosen to achieve a variation in effluent ammonium of ± 0.7 mg/l compared to V = 2000 m3 for constant DO control (2 mg/l). The DO set-point limits were 1 to 2.5 mg/l in all simulations and the controller gain was the same as in Table 6.3 but the integral time was always 2 d. A summary of the settings in the second set of simulations is given in Table 6.4. The ammonium set-point was changed in steps of 0.2 mg/l. The integral time was 2 d due to results from the first set of simulations.

The third set of simulations was similar to the second set of simulations, but with varying upper DO set-point limit (Table 6.5). The upper DO set-point limit was changed in steps of 0.2 mg/l. The ammonium set-point was chosen low based on results from the second set of simulations. No other


changes were made to the model apart from the treatment volumes and set-tings according to Table 6.4 and Table 6.5. The diffuser density was not changed when the volume was changed, but since the WAS outtake was not adapted to the volumes the simulations are not entirely comparable. The fixed upper DO set-point limit in Table 6.4 and ammonium set-point in Ta-ble 6.5 were decided to try and maximise the energy saving.

Table 6.4. Controller and model settings in the second set of simulations: different NH4 set-point and aerobic treatment volume.

Aerobic volume (m3)

Min. NH4 set-point (mg/l)

Max. NH4 set-point (mg/l)

Min/max DO set-point(mg/l) K Ti (d)

5 400 0.6 3.2 1/3.0 See Table 6.3 2

6 000 0.4 3.2 1/2.8 See Table 6.3 2

7 200 0.2 1.6 1/2.8 See Table 6.3 2

Table 6.5. Controller and model settings in the third set of simulations: different upper DO set-point limit and aerobic treatment volume.

Aerobic volume (m3)

Min. upper DO set-point limit (mg/l)

Max. upper DO set-point limit (mg/l)

NH4 SP K Ti (d)

5 400 1.6 3.6 1.2 See Table 6.3 2

6 000 1.6 3.2 1.0 See Table 6.3 2

7 200 1.6 3.4 0.6 See Table 6.3 2

The influent is more realistic than in Chapter 5 since it covers a full year instead of one single day and includes temperature effects and rain events. However, given the long simulation time the iterative strategy to change the ammonium set-point for each simulation in order to fulfil a specified mean daily ammonium concentration was not used. Therefore the average effluent nitrogen concentration varies in the simulations.

6.3.3 Performance measures The best performing controller was selected for each scenario. The selection was made based on a performance measure calculating the energy saving compared to constant DO control with the same effluent ammonium concen-tration:

DO

FBNHDO

Qair

QairQairES 4100

−= (6.1)

ES = Energy saving compared to constant DO control (%) QairDO = Mean annual air flow rate with constant DO (m3/d) QairNH4FB = Mean annual air flow rate with NH4 feedback control (m3/d)

6.3 Simulation settings 117

When calculating the energy saving according to Eq. (6.1) the ammonium feedback controller was compared to a constant DO controller with the same mean annual flow-proportional effluent ammonium concentration. The an-nual air consumption for the constant DO controller (QairDO) was calculated from a polynomial fit to simulation data where the constant DO set-point was varied in steps, see Figure 6.3. Hence, interpolation was used instead of matching each ammonium controller to a simulation with constant DO con-trol where the DO set-point would have taken a long time to iteratively de-cide to match the effluent ammonium concentration in each ammonium PI simulation. A third order polynomial fitted to the points in the graph made it possible to calculate QairDO for a certain effluent ammonium concentration.

Figure 6.3. Example of a curve from simulations with several constant DO set-points.

For comparison, a second performance measure was used where it was as-sumed that it was not possible to calculate the energy saving in comparison to constant DO control with the exact same average effluent concentration. Instead the energy saving was calculated as the saving in specific air flow rate (Eq.(6.2)) with ammonium feedback control compared to constant DO control, where the constant DO controller could have a different effluent ammonium concentration than the ammonium feedback controller, see Eq. (6.3).

( )outin

sp NHNHQin

QairQair

,4,4

1000−

= (6.2)

Qairsp = Mean annual specific air flow rate (m3air/kgNH4 removed) Qair = Mean annual air flow rate (m3/d) Qin = Mean annual inflow rate (m3/d) NH4,in = Mean annual influent NH4 concentration (mg/l) NH4,out = Mean annual effluent NH4 concentration (mg/l)

1.6 1.7 1.8 1.9 2 2.1 2.2 2.3 2.4 2.56.8

7

7.2

7.4

7.6

7.8

8

8.2

8.4x 10

4

Mean annual ammonium concentration in effluent (mg/l)

Mea

n an

nual

air

flow

rat

e (m

3/d)


DOsp

FBNHspDOsp

sp Qair

QairQairES

,

4,,100−

= (6.3)

ESsp = Energy saving compared to constant DO control based on specific air flow rate (%) Qairsp,DO = Mean annual specific air flow rate with constant DO control (m3air/kgNH4 removed) Qairsp,NH4FB = Mean annual specific air flow rate with NH4 feedback control (m3air/kgNH4 removed)

3

A water quality parameter is a natural choice of performance measure for a WWTP. However, in this study the total nitrogen concentration, EQI (efflu-ent quality index) or similar parameters were not used as a performance measure. The reason was that the nitrogen concentration or EQI was not sensitive enough to changes in the ammonium PI controller settings. Within a scenario with controller settings according to Table 6.3 the difference be-tween the highest and lowest total nitrogen concentration was < 0.1 mg/l. The difference in ammonium concentration could be larger, but since a high-er ammonium concentration is partly compensated for by lower nitrate con-centrations the total nitrogen concentration was relatively constant and so was EQI. Since a performance measure should be sensitive to changes in the variables under study – in this case the ammonium controller gain and inte-gral time – water quality parameters were only reported for comparison.

6.4 The impact of the controller settings The results from the first set of simulations (Table 6.2) are presented in con-troller maps where the energy saving calculated from Eq. (6.1) measures the performance of the controllers. For each controller the energy saving was plotted on the y-axis as (1 - ES)/100. The x-axis represents the level of inte-gral action in the controller. The PI controllers in BSM1_LT are on standard form:

+=

τ

ττ0

)(1

)()( deT

teKtui

(6.4)

Eq. (6.4) implies that when K is changed, this has an impact on the integra-tion term. To normalise the integral action in the controller the integral gain, Ki, is used in the plots as a measure of integral action instead of Ti. Rewriting of (6.4) gives

6.4 The impact of the controller settings 119

+=τ

ττ0

)()()( deT

KtKetu

i

+=τ

ττ0

)()( deKtKe i (6.5)

where Ki = K/Ti

In Chapter 5 the DO variance was used as a measure of controller speed on the y-axis in the graphs (see Section 5.7). DO variance becomes a much more blunt tool to describe controller speed when a full year is simulated, compared to when a single day is in focus as in Chapter 5. This is why the integral gain is plotted on the y-axis instead.

In the controller maps, |Ki| is plotted since K < 0. Dark grey colours repre-sent small controller gains and simulations to the left on the x-axis represent simulations with little integral action, i.e. small |Ki|. Figure 6.4, Figure 6.5 and Figure 6.6 display the controller maps for scenario 1 to 3 (upper DO set-point limit 2 mg/l), scenario 4 to 6 (upper DO set-point limit 2.5 mg/l) and scenario 7 to 9 (upper DO set-point limit 3 mg/l), respectively. The trend is the same for all the maps. For very slow controllers (dark colours to the left on the x-axis) the energy saving is small or even negative. Increased integral action or larger gains (|K|) increases the energy saving for the slow control-lers. When |K| is increased there comes a point when an increased integral action decreases the energy saving. The best controller is always a controller with little integral action.

There are several conclusions that can be drawn from the graphs in Figure 6.4 to Figure 6.6:

1. The lower the ammonium set-point or the higher the upper DO set-point limit, the larger the energy saving;

2. the lower the ammonium set-point or the higher the upper DO set-point limit, the larger the controller gain in the best performing controller and

3. the lower the ammonium set-point or the higher the upper DO set-point limit, the larger the span in achievable energy saving.

A higher upper DO set-point limit or a lower ammonium set-point increases the average DO concentration in the simulation. The two first trends in the above list can be explained by the same phenomenon: When the plant oper-ates at higher DO concentrations there is more energy to save from a change in DO concentration, but less to lose in terms of nitrification rate. When the plant operates at higher DO concentrations the plant is operating in a more saturated region of the Monod curve (Figure 5.5), and there is less to lose when varying the DO concentration in terms of nitrification rate compared to


when the slope of the Monod curve is higher (see Section 5.7). Hence the controller gain can be allowed to be higher. Another way to express it is that the process gain is lower at higher DO concentration, i.e. the ammonium response to a change in DO concentration is small, and this can allow for a larger controller gain. See further Section 7.10.

The span in energy saving between the controller with the highest and lowest energy saving for each scenario depends on the ammonium set-point and on the upper DO set-point limit. The connection to the ammonium set-point is explained by the same phenomenon as the two first points in the list. At lower ammonium set-points the DO concentration is higher and faster controllers can be allowed to decrease the energy consumption without com-promising the ammonium removal. It can be imagined as “folding” the con-trollers with high |K| values in Figure 6.4(c), Figure 6.5(c) and Figure 6.6(c) downward in the graphs when the ammonium set-point is decreased in Fig-ure 6.4(a), Figure 6.5(a) and Figure 6.6(a). This increases the span in the energy saving towards lower energy savings.

The span in energy saving is increased towards higher energy savings when the upper DO set-point limit is increased. A fast controller forces the DO concentration towards high levels without being hindered by a DO limit, which is why the energy saving is negative for some controllers with short integral times when the upper DO set-point limit is 3 mg/l. Therefore the selection of controller settings becomes more delicate when the upper DO set-point is high. Section 6.5 looks more into how the energy saving is changed for a range of ammonium set-points, upper DO set-point limits and treatment volumes.

The behaviour of the best performing ammonium controllers in the first set of simulations (Table 6.2) are plotted in Figure 6.7, Figure 6.8 and Figure 6.9. The settings in the best performing controller for each scenario when selection was based on the energy saving performance measure in Eq. (6.1) are given in Table 6.6. The controller gain ranges from -0.15 to -0.5 with higher |K| at lower ammonium set-points. |K| is lower when the upper DO set-point limit is 2 than if it is 2.5 or 3. The integral time is almost always the longest possible (2 d). The last columns in Table 6.6 are summarised in Fig-ure 6.10.

To achieve the lowest energy consumption in absolute terms, the upper DO set-point limit should be low and the ammonium set-point should be high, which is expected. However, the energy saving for ammonium PI con-trol compared to constant DO control is higher at higher DO set-point limits, and higher at lower ammonium set-points. Hence, the highest saving (6.9 %) was found when the DO set-point was 1 mg/l and the upper DO set-point limit was 1 to 3 mg/l (Table 6.6).


(a) (b) (c)

Figure 6.4. Controller maps showing relative air flow rate of NH4 PI control in comparison to constant DO control with the same effluent NH4 concentration. Set-tings according to Table 6.3 Min/max DO SP = 1/2 mg/l. (a) Scenario 1 (NH4 SP = 1 mg/l), (b) scenario 2 (NH4 SP = 2 mg/l), and (c) scenario 3 (NH4 SP = 3 mg/l).

(a) (b) (c)

Figure 6.5. Controller maps showing relative air flow rate of NH4 PI control in comparison to constant DO control with the same effluent NH4 concentration. Set-tings according to Table 6.3. Min/max DO SP = 1/2.5 mg/l. (a) Scenario 4 (NH4 SP = 1 mg/l), (b) scenario 5 (NH4 SP = 2 mg/l), and (c) scenario 6 (NH4 SP = 3 mg/l).

(a) (b) (c)

Figure 6.6. Controller maps showing relative air flow rate of NH4 PI control in comparison to constant DO control with the same effluent NH4 concentration. Set-tings according to Table 6.3. Min/max DO SP = 1/3 mg/l. (a) Scenario 7 (NH4 SP = 1 mg/l), (b) scenario 8 (NH4 SP = 2 mg/l), and (c) scenario 9 (NH4 SP = 3 mg/l).

0 10 20 30 40 500.95

0.96

0.97

0.98

0.99

1

1.01

Air

flow

rat

e co

mpa

red

to c

onst

ant D

O c

ontr

ol

|Ki|

0 10 20 30 40 500.95

0.96

0.97

0.98

0.99

1

1.01

|Ki|

0 10 20 30 40 500.95

0.96

0.97

0.98

0.99

1

1.01

|Ki|

K=−0.01 K=−0.025 K=−0.05 K=−0.075 K=−0.1 K=−0.15 K=−0.25 K=−0.5 K=−0.75 K=−1

0 10 20 30 40 500.94

0.95

0.96

0.97

0.98

0.99

1

1.01

Air

flow

rat

e co

mpa

red

to c

onst

ant D

O c

ontr

ol

|Ki|

0 10 20 30 40 500.94

0.95

0.96

0.97

0.98

0.99

1

1.01

|Ki|

0 10 20 30 40 500.94

0.95

0.96

0.97

0.98

0.99

1

1.01

|Ki|

K=−0.01 K=−0.025 K=−0.05 K=−0.075 K=−0.1 K=−0.15 K=−0.25 K=−0.5 K=−0.75 K=−1

0 10 20 30 40 500.93

0.94

0.95

0.96

0.97

0.98

0.99

1

1.01

1.02

Air

flow

rat

e co

mpa

red

to c

onst

ant D

O c

ontr

ol

|Ki|

0 10 20 30 40 500.93

0.94

0.95

0.96

0.97

0.98

0.99

1

1.01

1.02

|Ki|

0 10 20 30 40 500.93

0.94

0.95

0.96

0.97

0.98

0.99

1

1.01

1.02

|Ki|

K=−0.01 K=−0.025 K=−0.05 K=−0.075 K=−0.1 K=−0.15 K=−0.25 K=−0.5 K=−0.75 K=−1


(a) (b) (c)

Figure 6.7. DO concentrations for the best performing NH4 controllers when min/max DO SP = 1/2 mg/l. (a) Scenario 1 (NH4 SP = 1 mg/l), (b) scenario 2 (NH4 SP = 2 mg/l), and (c) scenario 3 (NH4 SP = 3 mg/l).

(a) (b) (c)

Figure 6.8. DO concentrations for the best performing NH4 controllers when min/max DO SP = 1/2.5 mg/l. (a) Scenario 4 (NH4 SP = 1 mg/l), (b) scenario 5 (NH4 SP = 2 mg/l), and (c) scenario 6 (NH4 SP = 3 mg/l).

(a) (b) (c)

Figure 6.9. DO concentrations for the best performing NH4 controllers when min/max DO SP = 1/3 mg/l. (a) Scenario 7 (NH4 SP = 1 mg/l), (b) scenario 8 (NH4 SP = 2 mg/l), and (c) scenario 9 (NH4 SP = 3 mg/l).

Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun0.5

1

1.5

2

2.5

3

DO

(m

g/l)


1

1.5

2

2.5

3


1

1.5

2

2.5

3


1

1.5

2

2.5

3

DO

(m

g/l)


1

1.5

2

2.5

3


1

1.5

2

2.5

3


1

1.5

2

2.5

3

DO

(m

g/l)


1

1.5

2

2.5

3


1

1.5

2

2.5

3


Table 6.6. Summary of simulation results. Qair, NH4, TN and DO are average annu-al concentrations. Each controller was compared to a constant DO controller with the same effluent ammonium concentration. SP = set-point. The lower DO set-point limit was 1 mg/l in all controllers.

Scenario NH4 SP (mg/l)

Max DO SP (mg/l)

Qair (m3/d)

NH4 (mg/l)

TN (mg/l)

DO (mg/l)

K

Ti (d)

Ref DO SP (mg/l)

ES vs Ref (%)

1 1 2 74 070 1.89 9.00 1.5 -0.25 2 1.8 4.6

2 2 2 71 206 2.07 9.00 1.3 -0.15 2 1.5 3.8

3 3 2 69 382 2.26 9.04 1.2 -0.15 1 1.4 3.1

4 1 2.5 77 473 1.73 8.94 1.7 -0.50 2 2.1 6.3

5 2 2.5 73 235 1.94 8.95 1.4 -0.25 2 1.7 4.3

6 3 2.5 70 051 2.20 9.02 1.3 -0.15 2 1.4 3.1

7 1 3 81 779 1.61 8.94 1.9 -0.50 2 2.3 6.9

8 2 3 75 137 1.85 8.92 1.5 -0.25 2 1.8 4.4

9 3 3 70 402 2.17 9.01 1.3 -0.15 2 1.5 3.1

Figure 6.10. Energy saving in percent for NH4 feedback control compared to con-stant DO control for scenario 1 to 9 in Table 6.2.

If a water quality parameter should be the selection criterion, other control-lers would have been selected compared to those in Table 6.6. The effluent ammonium concentrations for all the simulated controllers in scenario 4 to 6 are plotted in Figure 6.11. The controller maps look very much the same for Scenario 1 to 3 and 7 to 9, only with a shift in the average ammonium con-centration.

A majority of the controllers are within 0.1 mg/l from each other, mean-ing there is not a large difference between them. The total nitrogen concen-tration does not follow the same trend as the ammonium concentration. If denitrification is carbon limited a decreased ammonium concentration lead to an increased nitrate concentration and the total nitrogen concentration remains the same. Since this work does not include control loops for denitri-fication which could be used to also decrease the total nitrogen concentra-

0

1

2

3

4

5

6

7

8

NH4 SP=1 mg/l NH4 SP=2 mg/l NH4 SP=3 mg/l

Ener

gy sa

ving

(%)

DO SP 1/2 mg/l DO SP 1/2.5 mg/l DO SP 1/3 mg/l


tion, the total nitrogen concentration becomes a rough measure of the per-formance of the ammonium PI controllers.

An interesting effect in Figure 6.11 is that the very slow control can either lead to a high effluent ammonium concentration if the ammonium set-point is high (Figure 6.11(b), (c)), but the concentration is very low if the set-point is low (Figure 6.11(a)). The reason is that the low set-point forces the DO concentration to be high most of the time of the year, and since the controller is very slow the DO concentration will stay high. This leads to lower ammo-nium concentrations in the effluent. Higher ammonium set-points lead to the opposite effect with lower ammonium concentrations.

(a) (b) (c)

Figure 6.11. Controller maps showing the effluent NH4 concentrations for min/max DO SP = 1/2.5 mg/l. (a) Scenario 4 (NH4 SP = 1 mg/l), (b) scenario 5 (NH4 SP = 2 mg/l), and (c) scenario 6 (NH4 SP = 3 mg/l).

In the simulations the same controller settings was used all year around. Given the different performance of the treatment plant over the year (Figure 6.1) it could be expected that it would be beneficial to use different control-ler settings during different periods. A slower controller during the warm period would decrease the risk of unnecessarily high DO peaks during this period while a faster controller could decrease the DO concentration more during low loaded periods of the day when the effluent concentrations are higher.

Simulations in this chapter had constant DO limits all year, which is not how many treatment plants are operated. A varied upper and lower DO limit is considered in Chapter 7. The constant DO controllers used in this chapter had a constant DO set-point all year around. Given the profile of the ammo-nium concentration, a plant operator is likely to at least have different DO set-points during summer and winter periods. There is a potential to decrease the upper DO set-point limit in the ammonium controller and the DO set-point in the DO controller during constant DO control which could decrease the energy consumption. However, the energy saving ought to be similar as

0 10 20 30 40 50

1.5

1.6

1.7

1.8

1.9

2

2.1

2.2

2.3

2.4

2.5

Ave

rage

flow

pro

port

iona

l effl

uent

NH

4 co

ncen

trat

ion

(mg/

l)

|Ki|

0 10 20 30 40 50

1.5

1.6

1.7

1.8

1.9

2

2.1

2.2

2.3

2.4

2.5

|Ki|

0 10 20 30 40 50

1.5

1.6

1.7

1.8

1.9

2

2.1

2.2

2.3

2.4

2.5

|Ki|

K=−0.01 K=−0.025 K=−0.05 K=−0.075 K=−0.1 K=−0.15 K=−0.25 K=−0.5 K=−0.75 K=−1


in Table 6.6 if the reduction in upper DO set-point limit and DO set-point is performed in a similar manner.

The controller settings in the simulations range from slow to very fast. The simulations did not include noise; hence the sensitivity to measurement noise was not taken into consideration when simulating the fastest control-lers.

In the literature review in Chapter 3 there was a significantly higher ener-gy saving reported for ammonium feedback control than what this simula-tion study has demonstrated. Why is the maximum energy saving in this simulation study only just below 7 % when other work report savings of 25 % or even more? The main reason limiting the saving is that Table 6.6 only compares ammonium feedback control to constant DO control with the exact same effluent ammonium concentrations. Higher savings is achieved if the effluent ammonium concentration is allowed to be increased. Table 6.7 summarise the result from scenario 1 to 9 if the selection of controllers is performed based on the specific energy saving (Eq. (6.3)) calculated from constant DO control where the DO set-point was the same as the upper DO set-point limit in the ammonium feedback controller. The energy saving was calculated based on the specific air flow rate, see Eq. (6.3). In Table 6.7 the energy saving is higher when the DO limit is higher, which is the opposite of the results in Table 6.6.

An energy saving of over 20 % can be achieved if the effluent ammonium concentration can be compromised. The effluent ammonium concentration in the three reference constant DO controller was 1.8, 1.6 and 1.4 mg/l for a DO set-point of 2, 2.5 and 3 mg/l, respectively. Hence, a direct comparison between the controllers in Table 6.7 is problematic, but it explains why higher energy savings can be achieved from experiments with ammonium feedback control.

An increased effluent ammonium concentration compared to constant DO control is not necessarily negative. It may be that the plant was operated with very high DO concentrations to be sure to meet the effluent permit. With ammonium feedback control, safe operation is achieved since the DO set-point will reflect the effluent ammonium concentration, and a slight increase in effluent ammonium concentration can allow for a large energy saving with no risk to violate the effluent permit.

An interesting effect shown in Table 6.7 is that for high ammonium set-points the controller should be slowest possible (K = -0.01, Ti = 2 d) which achieves a lower DO concentration through relatively passive DO control, but also a higher effluent ammonium concentration.


Table 6.7. Summary of simulation results based on specific energy saving. Qair, NH4, TN and DO are average annual concentrations. Each controller was compared to constant DO control where the set-point equals the upper DO set-point limit in the NH4 controller. SP = set-point. The lower DO set-point limit was 1 mg/l in all con-trollers.

Scenario NH4 SP (mg/l)

Max DO set-point (mg/l)

Qair (m3/d)

NH4 (mg/l)

TN (mg/l)

DO (mg/l)

K

Ti (d)

Ref DO SP (mg/l)

ESsp vs Ref (%)

1 1 2 73 467 1.93 8.97 1.4 -0.75 2 2 8.5

2 2 2 71 180 2.08 8.99 1.3 -0.25 2 2 11.2

3 3 2 68 183 2.51 9.20 1.2 -0.01 2 2 12.9

4 1 2.5 77 156 1.76 8.91 1.6 -1.0 2 2.5 14.4

5 2 2.5 73 235 1.94 8.95 1.4 -0.25 2 2.5 18.2

6 3 2.5 68 183 2.51 9.20 1.2 -0.01 2 2.5 20.9

7 1 3 81779 1.61 8.94 1.9 -0.50 2 3 19.9

8 2 3 72 447 2.19 9.17 1.4 -0.01 2 3 25.6

9 3 3 68 183 2.51 9.20 1.2 -0.01 2 3 29.3

6.5 Energy saving for different treatment volumes In the second and third set of simulations (Table 6.4, Table 6.5), the ammo-nium set-point or the upper DO set-point limit was varied for three different treatment volumes. Figure 6.12 shows the controller gain in the best per-forming ammonium controllers related to the effluent ammonium concentra-tion (Figure 6.12(a)) and for each choice of ammonium set-point (Figure 6.12(b)) or upper DO set-point limit (Figure 6.12(c)) in the simulations.

The controller gains in the best performing ammonium controllers in Fig-ure 6.12 are increasing for lower ammonium concentrations. This is con-sistent with results presented in Section 5.7 and in Section 6.4. In the best performing controller the controller gain should never be slower than K = -0.2. The controller gain is most often the same for a particular choice of ammonium set-point regardless of treatment volume. The controller gain is similar for the same effluent ammonium concentration for a variation in ammonium set-point or for a variation in upper DO set-point limit (Figure 6.12(a)).

6.5 Energy saving for different treatment volumes 127

(a)

(b) (c)

Figure 6.12. The controller gain in the best performing NH4 controllers for three different treatment volumes related to (a) effluent NH4 concentration (b) NH4 set-point and (c) upper DO set-point limit.

Figure 6.13 shows the average annual air flow rate and the energy saving calculated with Eq. (6.1). The energy consumption is increased for larger treatment volumes – since larger treatment volumes remove more ammoni-um – but for a particular effluent ammonium concentration, the cost is higher for a smaller volume (Figure 6.13(a)).

There is a maximum energy saving for all treatment volumes in Figure 6.13(b) when the upper DO set-point limit is increased or when the effluent ammonium set-point is decreased. The trend towards increased energy sav-ing is interrupted at low ammonium concentrations (Figure 6.13(b)).

The maximum energy saving is not identical for a variation in ammonium set-point compared to when the upper DO set-point limit is changed. This could be a result of keeping either the upper DO limit or the ammonium set-point fixed in Table 6.4 and Table 6.5. The parameters were not changed concurrently to find the maximum points.

The energy saving and energy consumption in Figure 6.13 depend on how much ammonium is removed in the simulation. It does not depend on wheth-er it is the ammonium set-point or the upper DO set-point limit that is changed to reach a specific ammonium concentration in the effluent, apart from a slight deviation for V = 1800 m3.

0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8−1.5

−1.25

−1

−0.75

−0.5

−0.25

0

Effluent NH4 (mg/l)

Con

trol

ler

gain

, K

V = 1800 m3, var. NH4 SP

V = 2000 m3, var. NH4 SP

V = 2400 m3, var. NH4 SP

V = 2400 m3, var. max DO SP



1.5 2 2.5 3 3.5−1.5

−1.25

−1

−0.75

−0.5

−0.25

0

Max DO set−point (mg/l)

Con

trol

ler

gain

, K

0.4 0.8 1.2 1.6 2 2.4 2.8−1.5

−1.25

−1

−0.75

−0.5

−0.25

0

NH4 set−point (mg/l)

Con

trol

ler

gain

, K


(a) (b)

Figure 6.13. (a) Air flow rate and (b) energy saving related to of effluent NH4 con-centration.

Why is the maximum energy saving compared to constant DO control de-creasing when the WWTP removes more ammonium? The marginal cost for removing more ammonium at low concentrations is higher, which is demon-strated by a steeper slope in Figure 6.13(a) towards lower effluent ammoni-um concentrations. The slope of the air flow rate curve looks slightly differ-ent for constant DO control. The slope does not increase as much as for am-monium feedback control at low ammonium concentrations. This is why the energy saving compared to constant DO control is limited when very low effluent ammonium concentrations are desired, which is demonstrated in Figure 6.14 and Figure 6.15. When the ammonium concentration is decreased the air flow rate of ammonium feedback control is initially further from that of constant DO control. Around an effluent ammonium concentration of 1.6 mg/l, the difference in air flow rate starts to decrease, and at this point the energy saving is at its maximum.

Figure 6.14. The air flow rate related to effluent NH4 concentration for a variation of the NH4 set-point and of the upper DO set-point limit. V = 2000 m3.

0.5 1 1.5 2 2.5 37

7.5

8

8.5

9

9.5x 10

4

Effluent NH4 (mg/l)

Qai

r (m

3/d)

0.5 1 1.5 2 2.52

3

4

5

6

7

8

9

Effluent NH4 (mg/l)

Ene

rgy

savi

ng (

%)

V = 2000 m3, var. max DO SP V = 2400 m3, var. max DO SP V = 2400 m3, var. max DO SP

V = 1800 m3, var. NH4 SP V = 2000 m3, var. NH4 SP V = 2400 m3, var. NH4 SP

1.5 1.6 1.7 1.8 1.9 2 2.1 2.2 2.3

7

7.2

7.4

7.6

7.8

8

8.2

8.4

8.6

8.8

9x 10

4

Effluent NH4 (mg/l)

Qai

r (m

3/d)

Constant DO control

NH4 FB, var. NH4 SP

NH4 FB, var. upper DO SP limit

6.6 Conclusions 129

Figure 6.15. The difference in air flow rate is decreasing for low effluent NH4 con-centrations. Air flow rates calculated from interpolation of data points in Figure 6.14. V = 2000 m3.

The dips in energy saving in Figure 6.13(b) occur when the average effluent ammonium concentration is twice as high as the ammonium set-point – irre-spective of treatment volume. Trying to operate the plant at the limit of the plant capacity forces the DO concentration during ammonium PI control to be high most of the time (Figure 6.16). Hence, the benefit of using ammoni-um PI control is not fully explored when the plant operates close to the plant limit.

(a) (b)

Figure 6.16. The DO concentration with NH4 PI control is substantially higher when the NH4 set-point is reduced from (a) 0.8 mg/l to (b) 0.4 mg/l. The constant DO concentration to achieve the same effluent NH4 concentration is plotted for compari-son.

6.6 Conclusions Simulations were made in BSM1_LT with the purpose to investigate the effect of the PI controller settings on the energy consumption and effluent ammonium concentration. Special attention was given to the choice of con-

1.5 1.6 1.7 1.8 1.9 2

7

7.5

8

8.5

9

9.5

10x 10

4

Effluent ammonium (mg/l)

Air

flow

rat

e (m

3/d)

2000

2500

3000

3500

4000

4500

5000

5500

Diff

eren

ce in

air

flow

rat

e (m

3/d)

Constant DO controlNH4 FB controlDifference in Qair

Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun

0.5

1

1.5

2

2.5

3

DO

(m

g/l)

Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun

0.5

1

1.5

2

2.5

3

DO

(m

g/l)

NH4 SP = 0.8 mg/l, K =−0.75 Constant DO concentration NH4 SP = 0.4 mg/l, K =−1 Constant DO concentration


troller gain, integral time, ammonium set-point and upper DO set-point limit. A basis for the simulations was that over time, the plant is to achieve an av-erage nitrogen concentration in the effluent. Comparisons were made to con-stant DO control. The energy saving of ammonium PI control compared to constant DO con-trol ranged from around 3 % – for high ammonium set-points or low upper DO set-point limits – up to around 8 % – for low ammonium set-points or high upper DO set-point limits. The saving was lower than what has been reported by other authors (see Chapter 3). The energy saving can be in-creased significantly if the effluent ammonium concentration can be reduced slightly. As an example, an increase in effluent ammonium concentration of around 0.2 mg/l compared to constant DO control allowed for an energy saving of around 14 % instead of 6 % for one of the simulated scenarios. Increasing the upper DO set-point limit or the ammonium set-point reduced the effluent ammonium concentration at the expense of a larger air flow rate. However, for low ammonium concentrations the energy saving compared to constant DO control was larger, and the controller gain could be allowed to be higher. Therefore it can be concluded that if a WWTP operates with low ammonium set-points and/or is forced to use high DO concentrations to reach the treatment permit, the benefit from varying the DO concentration through ammonium feedback control is larger compared to keeping the DO concentration constant. The energy saving compared to constant DO control reached a maximum point for very low ammonium concentrations, where the energy saving start-ed to decrease when the DO concentrations were increased further. This result holds for several simulated treatment volumes, keeping the load the same. A similarity to Chapter 5 was that the best performing ammonium controller in all simulated scenarios had a long integration time. Apart from at low controller gains (|K|), energy consumption was increased when integral ac-tion was increased in the controller. The controller gain in the best perform-ing ammonium controller was larger for larger energy savings. Evaluating the performance of the ammonium feedback controller based on the specific air flow rate – the air flow rate normalised with removed amount of ammonium – could be problematic since the specific air flow rate favours slow controllers with low energy consumption and in average higher ammo-nium concentrations, particularly if the DO set-point is allowed to be high.

131

7 AMMONIUM PI CONTROLLER DESIGN – 7

PLANT SIMULATIONS

IMULATIONS WITH BSM1_LT in Chapter 6 showed that the energy saving from using ammonium feedback control depends on the control goal and controller settings. Chapter 7 extend these results by applying

the method from Chapter 6 to the models representing the three case studies in this thesis: Henriksdal, Käppala and Himmerfjärden WWTPs. These plants have different process layout, loads, air flow rate limits, sensor placements and control goals compared to each other and compared to the benchmark model in Chapter 6.

Introduction 7.1In Chapter 6, focus was to find guidelines on how an ammonium PI control-ler should be designed in order to remove ammonium in a cost-effective manner. The results showed that it is more worthwhile to implement ammo-nium feedback control when then effluent ammonium concentration should be low. The results also showed that the energy saving is up to 7 % com-pared to constant DO control, if the effluent ammonium concentrations should be the same. The BSM1_LT simulations in Chapter 6 are limited by how the BSM1_LT model is set-up. Chapter 7 try to widen the perspective by simulating three real treatment plants over a two year period to see if the design guidelines in Chapter 6 also hold for the Henriksdal, Käppala and Himmerfjärden models. Further introduction is found in Section 6.1.

S

132 7 Ammonium PI controller design – plant simulations

Model setup 7.2The Henriksdal model was calibrated based on plant data for 2011, and vali-dated for parts of 2012. The Käppala and Himmerfjärden models were cali-brated for 2012 and validated for 2011. One treatment line was simulated for each of the plants. A description of the calibration procedure for the Hen-riksdal, Käppala and Himmerfjärden models as well as a detailed account of the model-setup is given in Appendix A. A summary of the volumes and influent concentrations is presented in Table 7.1.

Table 7.1. Model setup and influent concentrations for the Henriksdal, Käpppala and Himmerfjärden models (one treatment line). TKN = Total Kjeldahl Nitrogen.


Anoxic volumes (m3) 15 600 4 291 0

Aerobic volumes (m3) 13 500 8 800 2 700

Aerobic sludge age (d) 8.7 6.8 7.5

Average flow (m3/d) 38 194 20 901 16 651

TKN (mg/l) 43.2 43.0 27.2

NH4 (mg/l) 30.1 30.9 20.5

COD (mg/l) 277 250 143

TSS (mg/l) 135 107 72

Control structure and controller settings 7.3Similar simulations were performed for the plant models as for BSM1_LT in Chapter 6 (see Section 6.3.2). Ammonium feedback control was simulated with different settings for two full years, and compared to constant DO con-trol. The settings in the ammonium PI controller are presented in Table 7.2. The effluent ammonium profiles for 2011 and 2012 can be found in Appen-dix A. Internal flows (WAS, RAS and internal recycle flow) was set to measured values used at the plants, see Appendix A.

Table 7.2. PI controller settings simulated for the three plant models.

Henriksdal WWTP

K -0.025 -0.05 -0.075 -0.1 -0.2 -0.3 -0.5 -0.75 -1 -1.5 -2 Ti (d) 0.01 0.02 0.05 0.075 0.1 0.2 0.5 1 2

Käppala WWTP

K -0.025 -0.05 -0.075 -0.1 -0.25 -0.5 -0.75 -1 -1.5 -2

Ti (d) 0.02 0.05 0.075 0.1 0.2 0.5 1 2

Himmerfjärden WWTP

K -0.025 -0.05 -0.075 -0.1 -0.15 -0.25 -0.5 -0.75 -1

Ti (d) 0.01 0.02 0.05 0.075 0.1 0.2 0.5 1 2

7.4 Aeration system limits 133

In the Henriksdal and Käppala models the ammonium sensor was placed after the secondary settler. The sensor was placed in the last aerated zone in the Himmerfjärden model. The control structure was the same as in Figure 6.2. The ammonium set-point was 1 mg/l at Henriksdal WWTP, 0.8 mg/l at Käppala WWTP and 2 mg/l at Himmerfjärden WWTP to reflect the actual settings at the plants.

The controllers in Table 7.2 were compared to constant DO control in two different ways. The first comparison was the same as that in Chapter 6 where the controllers were compared to constant DO control with the same effluent ammonium concentration. The energy consumption of constant DO control was calculated based on a polynomial fit to simulation data similar to Figure 6.3 and the best performing controller was the controller with the highest energy saving according to Eq. (6.1). Secondly, the ammonium controllers were compared to a reference simulation with constant DO controllers where the DO set-point reflected the control strategy at the plants if the plants were to operate with constant DO control. The best performing controller was the controller with the highest specific energy saving according to Eq. (6.3).

Aeration system limits 7.4Simulations were performed with and without air flow rate limits. The air flow limits rarely has an effect at Käppala WWTP during normal operation. The limits in the Henriksdal and Himmerfjärden models were the same as the limits at the plants. The effect of the air flow limits at Henriksdal and Himmerfjärden WWTPs is discussed below.

At Henriksdal WWTP the plant operators can choose to aerate two or three zones. Despite no aeration in the last aerated zone, oxygen-rich water will cross over from the adjacent zone since there are no walls between the zones. The air flow rate is increased in the adjacent zone during these peri-ods since air is used in the last aerated zone. The last two years the operators have avoided aerating the last zone since it increases the effluent nitrate con-centrations from the plant when high DO concentrations are recirculated to the anoxic zones. However, in the model the last aerobic zone was aerated apart from during summertime. The reasoning behind this choice was that it is not straight forward to model stray oxygen from an adjacent zone in the model with separate continuous stirred-tank reactors (CSTR). In periods the DO concentrations in the last zone is high in the plant (> 1.5 – 2 mg/l) de-spite no aeration. This has an impact on the nitrification and if the last zone was not aerated in the model the ammonium concentration was to become too high. Therefore, the zone was modelled as aerobic with a set-point of 1 mg/l, apart from during the summer period when a set-point of 0.4 mg/l was used to simulate no aeration in the last zone and oxygen escaping the adja-cent zone.


The air flow limits at Henriksdal WWTP create DO peaks in the last aer-ated zone when the aeration capacity cannot be reduced enough. This was captured in the Henriksdal model (Figure 7.1). More discussion about the DO peaks at Henriksdal is found in Chapter 9.

Figure 7.1. The effect of air flow rate limits in the last aerobic zone in the Henriks-dal model. DO set-point is 1 mg/l. Example from March 2012.

At Käppala WWTP there are occasional DO peaks in the last aerobic zone but to a much lesser extent than at Henriksdal WWTP, hence this was not modelled. The third aerated zone out of four is turned off during parts of the year, and this was done during summertime in the simulation model. The DO concentration was simulated at 0.4 mg/l during this period since. Similar to Henriksdal WWTP there are no walls between the aerobic zones.

Himmerfjärden WWTP was simulated with DO control in all six zones and ammonium control to determine the DO set-point. This is similar to the experimental line at the plant, described in more detail in Chapter 8. The calibration was performed on a model which did not have DO control as explained in Appendix A. The model was rebuilt for DO and ammonium control. Based on air flow and DO data from the experimental line the air flow model parameters were adjusted slightly to better describe the air flow limitations in the plant after reconstruction to zone DO control (Appendix A). It is only zone 2 that rarely suffers from air flow rate limitations. In zone 1, 3 and 4 the aeration capacity is not high enough and in zone 5 and 6 ca-pacity is too high resulting in DO peaks. This behaviour was captured in the model (Figure 7.2).

Since air flow rate limits are introduced in the simulations in this chapter, master controller windup could appear in the ammonium controller (see Sec-tion 5.3 and 8.12.1). However, for the ammonium controller to windup due to saturation in the slave DO controller, there would have to be an impact on the effluent ammonium concentration. Since ammonium dynamics is slow and there are several aerated zones receiving the DO set-point from the am-monium controller, it is likely that a possible effect of DO controller satura-tion in one or several aerated zones is disguised by the overall effluent am-monium pattern generated by daily and weekly variations.

04−Mar 11−Mar 18−Mar 25−Mar 01−Apr0.5

1

1.5

2

2.5

3

DO

con

cent

ratio

n (m

g/l)

7.4 Aeration system limits 135

(a) (b)

Figure 7.2. The effect of air flow rate limits at Himmerfjärden model. DO set-point is 2.5 mg/l. (a) Zone 1-3 and (b) zone 4-6. Example from March 2012.

It is important to limit the control signal from the ammonium controller, i.e. the DO set-point (see Chapter 5). The DO limits in the ammonium control-lers are presented in Figure 7.3 together with the DO concentrations in the constant DO controller. The constant DO reference controllers in Figure 7.3 were used to calculate the specific energy saving according to Eq. (6.3).

In the Henriksdal model the constant DO set-point was the same as meas-ured DO concentrations at the plant. The DO concentration varies over the year since the operators frequently change the DO set-point. During opera-tion with ammonium feedback control at the plant the upper and lower DO limits are altered from time to time, which was also the case in the simula-tion model (Figure 7.3(a), (b)).

In the Käppala model the maximum DO concentration was changed slightly during the year and the DO set-point in the reference simulation with constant DO control is the same as the maximum DO set-point limit in the ammonium controller. No treatment line is operated with constant DO con-trol at Käppala and Himmerfjärden WWTPs; hence no measured values could be used.

Himmerfjärden WWTP has since 2013 a new effluent permit where the plant should reach 8 mg/l in total nitrogen as an annual average. Therefore the plant has to maximise nitrification and the DO set-point limits and DO set-point in the constant DO controller was set to relatively high values in the model.

(a) (b)


1

1.5

2

2.5

3

DO

con

cent

ratio

n (m

g/l)


1

1.5

2

2.5

3

3.5

4

DO

con

cent

ratio

n (m

g/l)

Zone 1 Zone 2 Zone 3 Zone 4 Zone 5 Zone 6

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec0.5

1

1.5

2

2.5

3

3.5

4

4.5

DO

con

cent

ratio

n (m

g/l)


1

1.5

2

2.5

3

3.5

4

4.5

DO

con

cent

ratio

n (m

g/l)

HENRIKSDAL2011

HENRIKSDAL2012


(c) (d)

Figure 7.3. DO limits in ammonium controller (dashed) and DO concentration in reference controller (solid). (a) Henriksdal WWTP 2011, (b) Henriksdal WWTP 2012, (c) Käppala WWTP and (d) Himmerfjärden WWTP. The maximum DO limit for Käppala WWTP is the same as the DO set-point in the reference controller.

Lambda tuning 7.5The controllers in Table 7.2 were compared to controller settings achieved through lambda tuning (Åström and Hägglund, 1995). To perform lambda tuning on a process a step response test is performed. A step change in the control signal is made when the controller is in manual mode, and the effect on the controlled variable is captured. The time constant of the open-loop system is estimated through finding the time it took for the controlled varia-ble to reach 63 % of its final value. The integral time is selected to be the same as the open-loop time constant. The controller gain is calculated as

pT=λ (7.1)

u

yKS Δ

Δ= (7.2)

( )LK

TK

S +=

λ (7.3)

λ = Closed loop time constant p = Performance parameter, user choice KS = Process gain in open loop Δu = Change in control signal Δy = Change in controlled variable K = Controller gain in PI controller T = Open loop time constant L = Time delay


1

1.5

2

2.5

3

3.5

4

4.5

DO

con

cent

ratio

n (m

g/l)


1

1.5

2

2.5

3

3.5

4

4.5

DO

con

cent

ratio

n (m

g/l)

KÄPPALA2011, 2012

HIMMERFJÄRDEN2011, 2012

7.6 The Henriksdal model 137

The user chooses p, which is how many times the open loop time constant should be multiplied to decide the closed loop time constant (Eq. (7.1)). For moderately fast control, p is chosen from 2 to 3.

Lambda tuning was performed for the three plant models by taking a step in the DO set-point of 1 mg/l, with a constant influent. Due to the long re-sponse time of ammonium it was not possible to perform a step-response test with a dynamic influent. In the experiment p was chosen to be 2.5.

The Henriksdal model 7.6First, the best controller was selected based on the energy saving compared to constant DO control with the same effluent ammonium concentration (Eq. (6.1)). The controller maps are given in Figure 7.4 and Figure 7.6 with ex-amples of ammonium controller behaviour in Figure 7.3 and Figure 7.5.

The trends of the controller maps in Figure 7.4(a) and Figure 7.6(a) re-semble those from the BSM1_LT simulations in Chapter 6 with initially improved energy efficiency when the speed of the controller is increased until a minimum is found. Most controllers are within a range of 1 percent-age points from each other during 2011 and 2012. The fastest controllers are also close to the best performing controllers, which was not the case for the BSM1_LT simulations.

The controller map profile depends on the placement of the ammonium sensor (Figure 7.4(b), Figure 7.6(b)). The DO concentration with the sensor placed after the settler was in average higher than when the sensor was placed in situ, due to the delayed signal. The average ammonium concentra-tion was similar, but the energy saving was lower given the higher DO con-centration. According to Figure 7.4(b) and Figure 7.6(b) there was a smaller energy saving for shorter integral times when the sensor was placed after the settler. An example of the difference in DO profiles is presented in Figure 7.8.


(a) (b)

Figure 7.4. Controller map for the Henriksdal model 2011. Energy consumption for NH4 feedback controllers with settings according to Table 7.2, compared to constant DO control with the same effluent NH4 concentration. (a) NH4 sensor in aeration tank and (b) NH4 sensor after settler.

(a) (b)

Figure 7.5. Example of NH4 controller behaviour for the Henriksdal model 2011. NH4 sensor placed after the settler. The three controllers represent the three marked points in the controller map in Figure 7.4(b). (a) 2011, (b) detail from June 2011.

0 10 20 30 40 500.95

0.96

0.97

0.98

0.99

1

1.01

1.02

|Ki|

Air

flow

rat

e co

mpa

red

to c

onst

ant D

O c

ontr

ol

0 10 20 30 40 500.94

0.95

0.96

0.97

0.98

0.99

1

1.01

1.02

|Ki|

Air

flow

rat

e co

mpa

red

to c

onst

ant D

O c

ontr

ol

K=−0.025

K=−0.05

K=−0.075

K=−0.1

K=−0.2

K=−0.3

K=−0.5

K=−0.75

K=−1

K=−1.5

K=−2

2011NH

4 sensor in aeration tank

2011NH

4 sensor after settler

2

13

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec1

1.5

2

2.5

3

3.5

4

DO

(m

g/l)

M T W T F S S M T W T F S S M T W T F S S0

0.5

1

1.5

2

2.5

3

3.5

DO

, NH

4 (m

g/l)

1. K = −1, Ti = 0.02 d2. K = −0.75, Ti = 1 d3. K = −0.05, Ti = 0.5 dNH4 last aerobic zone

2011

7.6 The Henriksdal model 139

(a) (b)

Figure 7.6. Controller map for the Henriksdal model 2012. Energy consumption for NH4 feedback controllers with settings according to Table 7.2, compared to constant DO control with the same effluent NH4 concentration. (a) NH4 sensor in aeration tank and (b) NH4 sensor after settler.

(a) (b)

Figure 7.7. Example of NH4 controller behaviour for the Henriksdal model 2012. NH4 sensor placed after the settler. The three controllers represent the three marked points in the controller map in Figure 7.6(b). (a) 2011, (b) detail from June 2012.

Figure 7.8. DO concentration and ammonium concentration with K = -1, Ti = 0.02 d and different NH4 sensor placements. Example from June 2011.

0 10 20 30 40 500.96

0.97

0.98

0.99

1

1.01

1.02

|Ki|

Air

flow

rat

e co

mpa

red

to c

onst

ant D

O c

ontr

ol

0 10 20 30 40 500.96

0.97

0.98

0.99

1

1.01

1.02

|Ki|

Air

flow

rat

e co

mpa

red

to c

onst

ant D

O c

ontr

ol

K=−0.025

K=−0.05

K=−0.075

K=−0.1

K=−0.2

K=−0.3

K=−0.5

K=−0.75

K=−1

K=−1.5

K=−2

2012NH


2012NH


1

3

2


1.5

2

2.5

3

3.5

4

DO

(m

g/l)


0.5

1

1.5

2

2.5

3

DO

, NH

4 (m

g/l)

1. K = −1, Ti = 0.02 d2. K = −0.75, Ti = 0.5 d3. K = −0.05, Ti = 0.5 dNH4 last aerobic zone

2012

M T W T F S S M T W T F S S M T W T F S S

0

0.5

1

1.5

2

2.5

3

3.5

DO

, NH

4 (m

g/l)

DO concentration, NH4 sensor in situDO concentration, NH4 sensor after settlerNH4 last aerobic zone


The controller maps when the specific energy saving (Eq. (6.3)) was the basis for calculating the controller performance are given in Figure 7.9. The ammonium sensor was placed in the aeration tank. The controller maps were similar if the sensor was placed after the settler. For 2011 and 2012 there is a decreased energy saving for short integral times, due to an increased DO concentration for these controllers. The average DO concentration was in-creased due to a faster reaction to ammonium peaks. The energy saving was significantly higher for 2011 since the operators chose higher DO set-points during 2012 compared to 2011 (Figure 7.3).

The best controller for the 2012 simulation was the slowest possible con-troller. The benefit of this controller is that it never exceeds 2.5 mg/l in DO set-point. Since it is costly to operate at high DO concentrations this control-ler has an in average lower energy consumption than the faster controllers which increases the DO set-point to around the maximum DO limit. The slowest controller was also chosen in some of the scenarios in Chapter 6 when the specific energy saving was the performance measure (Table 6.7). With slow ammonium control, disturbance rejection is poor.

(a) (b)

Figure 7.9. Controller map for the Henriksdal model (a) 2011 and (b) 2012. Energy consumption for NH4 feedback controllers with settings according to Table 7.2, compared to constant DO control according to Figure 7.3. NH4 sensor in situ. Note: different scales on the y-axes.

0 10 20 30 40 500.93

0.94

0.95

0.96

0.97

0.98

0.99

1

|Ki|

m3

air/

kgN

H4

rem

oved

com

pare

d to

con

stan

t DO

con

trol

0 10 20 30 40 500.85

0.855

0.86

0.865

0.87

0.875

0.88

0.885

0.89

0.895

0.9

|Ki|

m3

air/

kgN

H4

rem

oved

com

pare

d to

con

stan

t DO

con

trol

K=−0.025

K=−0.05

K=−0.075

K=−0.1

K=−0.2

K=−0.3

K=−0.5

K=−0.75

K=−1

K=−1.5

K=−2

20122011

7.7 The Käppala model 141

The Käppala model 7.7The controller maps for the Käppala model are plotted in Figure 7.10 for 2011 and in Figure 7.12 for 2012. The controller performance was calculated with Eq. (6.1), where the constant DO reference controller had the same effluent ammonium concentration as each of the simulated controllers. Ex-amples of controller performance are given in Figure 7.11 and Figure 7.13.

(a) (b)

Figure 7.10. Controller map for the Käppala model 2011. Energy consumption for NH4 feedback controllers with settings according to Table 7.2, compared to constant DO control with the same effluent NH4 concentration. (a) NH4 sensor in aeration tank and (b) NH4 sensor after settler.

(a) (b)

Figure 7.11. Example of NH4 controller behaviour for the Käppala model 2011. NH4 sensor placed after the settler. The three controllers represent the three marked points in the controller map in Figure 7.10(b). (a) 2011, (b) detail from June 2012.

0 10 20 30 40 500.92

0.925

0.93

0.935

0.94

0.945

0.95

0.955

0.96

0.965

0.97

|Ki|

Air

flow

rat

e co

mpa

red

to c

onst

ant D

O c

ontr

ol

0 10 20 30 40 500.92

0.925

0.93

0.935

0.94

0.945

0.95

0.955

0.96

0.965

0.97

|Ki|

Air

flow

rat

e co

mpa

red

to c

onst

ant D

O c

ontr

ol

K=−0.025

K=−0.05

K=−0.075

K=−0.1

K=−0.25

K=−0.5

K=−0.75

K=−1

K=−1.5

K=−2

2011NH


2011NH


2

13


1.5

2

2.5

DO

(m

g/l)


1

2

3

4

5

DO

, NH

4 (m

g/l)

K=−1, Ti=0.02 dK=−0.5, Ti=0.5 dK=−0.05, Ti=0.5 dNH4 last aerobic zone

2011


(a) (b)

Figure 7.12. Controller map for the Käppala model 2012. Energy consumption for NH4 feedback controllers with settings according to Table 7.2, compared to constant DO control with the same effluent NH4 concentration. (a) NH4 sensor in aeration tank and (b) NH4 sensor after settler.

(a) (b)

Figure 7.13. Example of NH4 controller behaviour for the Käppala model 2012. NH4 sensor placed after the settler. The three controllers represent the three marked points in the controller map in Figure 7.12(b). (a) 2012, (b) detail from June 2012.

0 10 20 30 40 500.95

0.96

0.97

0.98

0.99

1

1.01

|Ki|

Air

flow

rat

e co

mpa

red

to c

onst

ant D

O c

ontr

ol

0 10 20 30 40 500.95

0.96

0.97

0.98

0.99

1

1.01

|Ki|

Air

flow

rat

e co

mpa

red

to c

onst

ant D

O c

ontr

ol

K=−0.025

K=−0.05

K=−0.075

K=−0.1

K=−0.25

K=−0.5

K=−0.75

K=−1

K=−1.5

2012NH


2012NH


3

1

2


1.5

2

2.5

DO

(m

g/l)


1

2

3

4

DO

, NH

4 (m

g/l)

K=−1, Ti=0.02 dK=−1, Ti=1 dK=−0.05, Ti=0.5 dNH4 last aerobic zone

2012

7.8 The Himmerfjärden model 143

The controller maps in Figure 7.10 and Figure 7.12 reach a plateau where there is little effect on the energy saving when the controller speed is in-creased. There are many controllers which are within the range of 0.1 to 0.2 percentage points of the best controllers and could therefore be considered to have similar performance. A probable reason why the energy consumption does not increase at higher integral gains as it does for the Henriksdal model (Figure 7.6) is the lack of frequent ammonium peaks in the Käppala model. As for the Henriksdal model, there is a slight decrease of energy saving at shorter integral times when the ammonium sensor is placed after the second-ary settler (Figure 7.10(b), Figure 7.12(b)) and the energy saving with the best ammonium controller was also slightly lower.

The controller maps when the specific energy saving (Eq. (6.3)) was the basis for calculating the controller performance are given in Figure 7.14. The ammonium sensor was placed in the aeration tank, with similar trends when the sensor was placed after the settler. Constant DO control had settings according to Figure 7.3. The controller map profile was not changed by a different calculation method of energy saving, but the saving was larger than in Figure 7.10 and Figure 7.12.

(a) (b)

Figure 7.14. Controller map for the Henriksdal model (a) 2011 and (b) 2012. Ener-gy consumption for NH4 feedback controllers with settings according to Table 7.2, compared to constant DO control according to Figure 7.3. NH4 sensor in situ.

The Himmerfjärden model 7.8The controller maps for the Himmerfjärden model are presented in Figure 7.15. The best performing controller was selected based on the energy saving according to Eq. (6.1). The profiles are similar to the Henriksdal profile when the ammonium sensor was placed in situ (Figure 7.4(a), Figure 7.6(a)). The best controller had the longest simulated integral time with K = -0.75 for

0 10 20 30 40 500.91

0.92

0.93

0.94

0.95

0.96

0.97

0.98

|Ki|

m3

air/

kgN

H4

rem

oved

com

pare

d to

con

stan

t DO

con

trol

0 10 20 30 40 500.91

0.92

0.93

0.94

0.95

0.96

0.97

0.98

|Ki|

m3

air/

kgN

H4

rem

oved

com

pare

d to

con

stan

t DO

con

trol

K=−0.025

K=−0.05

K=−0.075

K=−0.1

K=−0.25

K=−0.5

K=−0.75

K=−1

K=−1.5

K=−2

2011 2012


the two simulated years. An increased integral gain creates a slight decrease in energy saving.

An example of controller behaviour is plotted in Figure 7.16 for 2011 and Figure 7.17 for 2012. At a few occasions the DO concentration has a dip at maximum load where the DO concentration cannot stay at the maximum DO set-point (Figure 7.17(b)). This is caused by a limited air flow supply. In Figure 7.16 and Figure 7.17 the DO concentration in zone 2 is plotted since this is the zone with least difficulties due to air flow rate limits.

In Figure 7.18, the controller map is plotted where a direct comparison is made to the controller with the DO concentration in Figure 7.3 and the best controller was selected based on the specific energy saving (Eq. (6.3)). The energy saving was larger, but the effluent ammonium concentration was slightly higher.

(a) (b)

Figure 7.15. Controller map for the Himmerfjärden model (a) 2011, (b) 2012. Ener-gy consumption for NH4 feedback controllers with settings according to Table 7.2, compared to constant DO control with the same effluent NH4 concentration.

(a) (b)

Figure 7.16. Example of NH4 controller behaviour for the Himmerfjärden model 2011. The three controllers represent the three marked points in the controller map in Figure 7.15(a). DO concentration from zone 2. (a) 2011, (b) detail from August 2011.

0 10 20 30 40 500.91

0.92

0.93

0.94

0.95

0.96

0.97

0.98

0.99

1

1.01

|Ki|

Air

flow

rat

e co

mpa

red

to c

onst

ant D

O

0 10 20 30 40 500.91

0.92

0.93

0.94

0.95

0.96

0.97

0.98

0.99

1

1.01

|Ki|

Air

flow

rat

e co

mpa

red

to c

onst

ant D

O

K=−0.025

K=−0.05

K=−0.075

K=−0.1

K=−0.15

K=−0.25

K=−0.5

K=−0.75

K=−1

2011 2012

3

2

1

3

3

2

1


1.5

2

2.5

3

3.5

DO

(m

g/l)

W T F S S M T W T F S S M T W T F S S M T0

1

2

3

4

5

6

DO

, NH

4 (m

g/l)

1. K=−1, Ti=0.02 d2. K=−0.75, Ti=2 d3. K=−0.05, Ti=0.5 dNH4 last aerobic zone

2011

7.9 Comparison between plant models 145

Figure 7.17. Example of NH4 controller behaviour for the Himmerfjärden model 2012. The three controllers represent the three marked points in the controller map in Figure 7.15(b). DO concentration from zone 2. (a) 2012, (b) detail from October 2011.

(a) (b)

Figure 7.18. Controller map for the Himmerfjärden model (a) 2011 and (b) 2012. Energy consumption for NH4 feedback controllers with settings according to Table 7.2, compared to constant DO control according to Figure 7.3.

Comparison between plant models 7.9A summary of the results from the simulations with the three plant models is presented in Figure 7.19, Table 7.3 and Table 7.4. In Table 7.3 the controller performance was calculated based on energy saving (Eq. (6.1)) and in Table 7.4 the best controller was selected based on the specific energy saving (Eq. (6.3)). The results are similar to those in Chapter 6 with an energy saving from around 3 to 9 % if no compromise was made to the average effluent ammonium concentration (Table 7.3). For example, if the effluent ammoni-um concentration could be allowed to be around 0.2 to 0.3 mg/l higher than the reference simulation the energy saving could be above 10 % (Table 7.4).

Placing the ammonium sensor after the secondary settler had an impact on the energy consumption rather than on the treatment results for the Henriks-


1.5

2

2.5

3

3.5

DO

(m

g/l)

T W T F S S M T W T F S S M T W T F S S M0

1

2

3

4

5

6

7

DO

, NH

4 (m

g/l)

1. K=−1, Ti=0.02 d2. K=−0.75, Ti=2 d3. K=−0.05, Ti=0.5 dNH4 last aerobic zone

2012

0 10 20 30 40 500.9

0.91

0.92

0.93

0.94

0.95

0.96

0.97

0.98

Integral gain, Ki

m3

air/

kgN

H4

rem

oved

com

pare

d to

con

stan

t DO

con

trol

0 10 20 30 40 500.9

0.91

0.92

0.93

0.94

0.95

0.96

0.97

0.98

Integral gain, Ki

m3

air/

kgN

H4

rem

oved

com

pare

d to

con

stan

t DO

con

trol

K=−0.025

K=−0.05

K=−0.075

K=−0.1

K=−0.15

K=−0.25

K=−0.5

K=−0.75

K=−1

2011 2012


dal and Käppala models (Table 7.3). The effect was larger in the Henriksdal model (1 percentage unit difference) than in the Käppala model (0.3 percent-age unit difference). A probable reason to this is that the effluent ammonium variation was larger in the Henriksdal model than in the Käppala model (Figure 4.11).

The variation of the effluent ammonium concentration also had an impact on how sensitive the energy saving was to the choice of PI controller set-tings. More frequent effluent ammonium peaks (Himmerfjärden, Henriksdal) results in smaller energy saving for the fastest controllers. Less frequent effluent ammonium peaks (Käppala) meant that the results were insensitive to the controller settings, as long as the controller was not too slow.

The flow was higher during 2012 due to more precipitation, which partly can explain the difference in energy saving between 2011 and 2012 in the Käppala and Himmerfjärden models. More precipitation brings about more frequent effluent ammonium peaks and a smaller difference in DO concen-tration compared to constant DO control. The difference was around 3.5 percentage units between the years (Figure 7.19). In the Henriksdal simula-tions, the energy saving was also slightly smaller in 2011 (Table 7.3). In Table 7.4 the large difference between the years arise since the operators changed the DO set-point in the treatment lines with constant DO control less often and less aggressively in 2012 than in 2011.

Including air flow rate limits in the simulations do impact the results, par-ticularly so in the Himmerfjärden model where the air flow rate often satu-rates. The average air flow rate in the simulations with constant DO control were approximately 5 % higher than without air flow rate limits. The in-crease in air flow rate for the ammonium PI controller was just below 2 %. This explains why the energy saving was larger without air flow rate limits. In the Henriksdal model where the air flow rate saturates occasionally in the last aerobic zone only the effect on the energy saving was minor.

Compared to the results in Table 7.3, the results in Table 7.4 are more re-alistic since they include a more realistic variation of the DO set-points ac-cording to Figure 7.3. For the Käppala and Himmerfjärden models the DO set-points in the constant DO controller was varied less compared to the Henriksdal model and the difference between Table 7.3 and Table 7.4 origi-nates from the difference in effluent ammonium concentration, rather than how the operators changed the DO set-point during constant DO control.

The results from the simulations are subject to assumptions and settings made in the calibration of the models. The simulations do not include meas-urement noise or sensor failures which might have had an impact on the results.

7.9 Comparison between plant models 147

Figure 7.19. Energy saving according to Table 7.3 and specific energy saving ac-cording to Table 7.4 for (a) Henriksdal WWTP, (b) Käppala WWTP and (c) Him-merfjärden WWTP.

Table 7.3. Summary of results from simulations with the Henriksdal, Käppala and Himmerfjärden models. The best performing controller selected based on energy saving (Eq. (6.1)). Comparison to constant DO control with the same effluent NH4 concentration.


2011 2012 2011 2012 2011 2012

Qair limit

K best controller -1.5 -2 -1 -1.5 -0.75 -0.75

Ti best controller (d) 2 2 1 1 2 2

Qair (m3/d) 91 240 96 700 55 522 45 492 107 174 106 221

Effluent NH4 (mg/l) 1.1 1.0 1.6 1.4 2.2 2.4

Effluent total nitrogen (mg/l) 8.1 8.3 9.3 8.6

Energy saving (%) * -3.9 -3.3 -7.8 -4.3 -8.7 -5.0

No Qair limit

K best controller -1.5 -2 -0.75 -0.75

Ti best controller (d) 2 2 2 2

Qair (m3/d) 91 165 96 668 109 416 108 710

Effluent NH4 (mg/l) 1.1 1.0 2.1 2.3

Effluent total nitrogen (mg/l) 8.1 8.3

Energy saving (%) * -4.1 -3.5 -5.3 -5.7

Qair limit and NH4 sensor after settler

K best controller -0.75 -0.75 -0.5 -1

Ti best controller (d) 1 0.5 0.5 1

Qair (m3/d) 91 592 97 310 55 679 45 561

Effluent NH4 (mg/l) 1.1 1.0 1.6 1.4


Energy saving (%) * -2.9 -2.4 -7.5 -3.9

* Compared to constant DO control.

02468

10121416

2011 2012

Ener

gy sa

ving

(%)

Energy saving Specific energy saving

0

2

4

6

8

10

2011 2012

Ener

gy sa

ving

(%)


0

2

4

6

8

10

2011 2012

Ener

gy sa

ving

(%)



Table 7.4. Summary of results from simulations with the Henriksdal, Käppala and Himmerfjärden models. The best performing controller selected based on specific energy saving (Eq. (6.3)). Comparison to constant DO control according to Figure 7.3**. TN = total nitrogen.


2011 2012 2011 2012 2011 2012

Qair limit

K best controller -0.5 -0.025 -1.5 -1.5 -0.5 -0.75

Ti best controller (d) 2 2 2 2 0.5 2

Qair (m3/d) 90 703 96 127 55 488 45 470 107 120 106 221

Effluent NH4 (mg/l) 1.1 1.1 1.6 1.4 2.2 2.4


Energy saving (%) * -6.9 -14.9 -8.6 -8.0 -10.6 -8.5

Specific energy saving (%) * -7.0 -14.4 -8.4 -7.7 -9.2 -7.5

m3 air/kg TN removed (%) * -7.0 -14.8 -8.8 -8.2

No Qair limit

K best controller -0.5 -0.025 -0.15 -0.25

Ti best controller (d) 2 2 0.2 0.5

Qair (m3/d) 90 628 96 097 108 965 108 219

Effluent NH4 (mg/l) 1.1 1.1 2.1 2.3

Effluent total nitrogen (mg/l) 8.1 8.3

Energy saving (%) * -6.9 -14.9 -13.5 -11.1

Specific energy saving (%) * -6.9 -14.3 -11.7 -9.7

m3 air/kg TN removed (%) * -7.0 -14.7

Qair limit and NH4 sensor after settler

K best controller -0.5 -0.025 -1.5 -1.5

Ti best controller (d) 2 1 2 2

Qair (m3/d) 91 169 96 728 55 554 45 537

Effluent NH4 (mg/l) 1.1 1.1 1.6 1.4


Energy saving (%) * -6.5 -14.4 -8.5 -7.8

Specific energy saving (%) * -6.5 -13.9 -8.3 -7.5

m3 air/kg TN removed (%) * -6.5 -14.3 -8.6 -8.1

* Compared to constant DO control.

** Effluent NH4 with constant DO control:

Henriksdal: 1.1 mg/l (2011), 0.9 mg/l (2012)

Käppala: 1.5 mg/l (2011), 1.3 mg/l (2012)

Himmerfjärden: 1.9 mg/l (2011), 2.2 mg/l (2012)

7.10 Comparison to lambda tuning 149

Comparison to lambda tuning 7.10The method in this study was to test a wide range of controller settings and see the effect on the energy consumption and treatment results. How do the best controllers with the best performance in Table 7.3 compare with con-troller settings selected through lambda tuning?

The results from lambda tuning performed with a constant influent are presented in Table 7.5. All the controllers have the same integral time with lambda tuning. The integral time is 2 hours, which means the controllers end up to the right in the controller maps in this chapter. The integral time in lambda tuning is the open-loop time constant.

The controller gains (|K|) selected with lambda tuning were higher than when selection was based on the energy saving. The controller gain from lambda tuning reflects the process gain in the models. A larger process gain (|KS|) requires a smaller controller gain (|K|). The process gain is not con-stant. It depends on in what DO region the step change in DO set-point is taken and on the model equations governing the growth rate of nitrifying bacteria (see Figure 5.5). At lower DO concentrations or when the half-saturation constant of oxygen is higher, the impact of a step change in DO will be larger and the process gain (|KS|) will therefore be higher.

There are similarities between when controller settings are selected based on Eq. (6.1), Eq. (6.3) or from lambda tuning. All methods select a larger controller gain (|K|) for the Henriksdal and Käppala models compared to the Himmerfjärden model. In Chapter 6, Eq. (6.1) selected a higher controller gain for higher DO concentrations (Section 6.4), which the lambda tuning method also would since it is sensitive to the process gain. In this respect, all methods take the process gain into account and balance a higher process gain (|KS|) with a lower controller gain (|K|).

Another similarity between lambda tuning and the selection of controller settings based on energy saving is the response to an increased time delay in the open-loop system. When the sensor is placed after the settler the time delay in the system was increased, and a slower controller was selected (Ta-ble 7.3). When the time delay (L) is increased, the controller gain calculated with lambda tuning is decreased (Eq. (7.3)). There was no measureable time delay in the simulation model, which is not the case at a real plant (Table 7.5).

Table 7.5. Controller settings in the three plant models when lambda tuning was used.

Δy (mg/l) Δu (mg/l) KS T (d) L (d) Ti (d) K

Henriksdal 0.135 1 -0.135 0.08 0 0.08 -3.0 Käppala 0.141 1 -0.141 0.08 0 0.08 -2.8

Himmerfjärden 0.407 1 -0.407 0.08 0 0.08 -1.0


Conclusions 7.11In Chapter 7, simulations of ammonium PI control were performed in mod-els calibrated for the Henriksdal, Käppala and Himmerfjärden WWTPs. Compared to constant DO control, ammonium PI control reduced the energy requirement from around 3 up to 9 % if the annual average ammonium con-centration was exactly the same. If the average ammonium concentration was allowed to be 0.2 mg/l higher with ammonium PI control the saving was up to approximately 15 %. In the models simulating real plant performance the main causes of differ-ence in energy saving between evaluation years was how the reference strat-egy was operated and variations in influent load due to precipitation. A regu-lar change of the DO set-point during constant DO control reduces the mar-gins on how much ammonium control can be expected to contribute in terms of energy saving. Frequent effluent ammonium peaks caused by rain events also reduce the potential for energy saving. The energy saving was relatively insensitive to the choice of controller gain and integral time, unless the controller was very slow. For plants with fre-quent effluent ammonium peak a very fast controller decreased the energy saving. If the ammonium peaks were few, the energy consumption was sim-ilar for a large number of controller settings. For all plant models, most con-trollers were within 2 percentage points from the best performing controller, which allows for relatively simple manual tuning. Placing the ammonium sensor after the secondary sedimentation tank did only have a small influence on the effluent ammonium concentration, but it did reduce the energy saving. There was also a trend towards lower energy saving for shorter integral times. The benefits of placing the sensor after the settler are a possible improvement of the quality of the sensor readings and less sensor maintenance. If the controller gain and integral time was determined with lambda tuning, the controller became faster than the best performing controllers according the energy saving performance measure. There were several similarities be-tween how the energy saving performance measure and lambda tuning se-lected controller settings.

PART III

FULL-SCALE AMMONIUM FEEDBACK CONTROL

153

8 LONG-TERM EVALUATION OF FULL-8

SCALE AMMONIUM FEEDBACK CONTROL

ETWEEN 2010 AND 2014 Henriksdal, Käppala and Himmerfjärden WWTPs have been cooperating with the aim to implement ammoni-um feedback control in their activated sludge processes. This chapter

summarises the results from the long-term evaluations of controllers at the plants. The focus of the chapter is to present examples of possibilities and challenges with full-scale ammonium control. Apart from giving an account of the problem of energy estimations, the chapter also offers lessons learnt from the implementation process. This work continues on the path created by other authors to further increase the knowledge from working with full-scale implementation and evaluation.

Introduction 8.1Ammonium control has been shown to contribute to energy savings around 5 to 25 % in full-scale installations (see Chapter 3). A majority of full-scale evaluations of ammonium control are performed during a shorter time period than 2 months. Apart from the STAR controller (Nielsen and Önnerth, 1995), Lindberg (1997) and Suescun et al. (2001) published early results with ammonium feedback control from pilot-scale experiments. Ingildsen et al. (2002) and Ayesa et al. (2006) published experiments with full-scale ammonium control performed in Spain and Denmark. Ingildsen et al. (2002) points out the importance of an even flow distribution between parallel lines to be able to compare controllers. Ayesa et al. (2006), mentioned the diffi-culty of performing a quantitative assessment of a control strategy, due to variations in load and temperature. Rieger et al. (2012c) evaluate ammoni-

B

154 8 Long-term evaluation of full-scale ammonium feedback control

um-based control in three full-scale plants in Switzerland. The paper empha-sises the importance of data quality evaluation and present examples of where plant equipment limits the performance of process controllers Rieger et al. (2012c) also bring up the issue of nitrite formation and sludge quality problems, as well as the knowledge and effort required by the operators at the plants to maintain the quality of the measuring and control equipment

In the work presented in this chapter the goal was primarily to reduce the energy requirement, thereby reducing the operation costs. All the plants wanted to achieve an improved or maintained effluent ammonium concentra-tion. There was an initial wish to also look at the nitrous oxide emissions, but due to the long implementation and evaluation time of the ammonium controllers it was only possible to achieve nitrous oxide measurements at one of the three plants.

Control strategies and instrumentation 8.2In this project, ammonium PI controllers have been used, see Figure 8.1. All controllers at the plants were discrete time implementations of PI controllers (Eq. (2.1)). All controllers have anti-windup and the controllers at Henriks-dal and Käppala WWTPs use tracking, making e.g. bumpless transfer avail-able.

Ammonium feedback PI-control was chosen as the preferred control strategy since it has been evaluated with success by others and was preferred over more advanced optimal and predictive controllers in order to try simple things first. An ammonium PI controller has the potential to achieve close to optimal performance for periodic disturbances (see Chapter 5).

More information and data about the plants are given in Chapter 4. The processes and instrumentation level in the experimental lines are depicted in Figure 8.2.

Figure 8.1. Cascade control of ammonium. One ammonium controller determines the DO set-point of several aerated zones. For simplicity only one aerated zone is included in the figure.

8.3 Henriksdal WWTP 155

Figure 8.2. The activated sludge processes and instrumentation in the experimental lines. Top: Henriksdal WWTP, middle: Käppala WWTP and bottom: Him-merfjärden WWTP.

Henriksdal WWTP 8.3 Control structure 8.3.1

Six out of seven treatment lines at Henriksdal WWTP were used for control-ler evaluation. Ammonium feedback control was introduced in three lines, and the other three lines operated with constant DO control. The plant had ammonium sensors installed after the secondary settlers at the start of the project, and these were used for ammonium control. The DO set-point in the last aerobic zone was fixed (1 to 2 mg/l) not to risk elevated DO concentra-tions recirculating to the anoxic zones.

Controller tuning 8.3.2No improvements of the air flow and DO controller tunings were needed. In particular during cold periods of the year the effluent ammonium concentra-tion varies on a daily basis. The ammonium controller was manually tuned to be slow, following weekly and monthly variations rather than daily.


Käppala WWTP 8.4 Control structure 8.4.1

Käppala had the possibility to operate with DO control and ammonium feed-back control in several lines at the start of the project as reported by Thun-berg et al. (2009). Two treatment lines were used in the project, operated interchangeably with constant DO control and with ammonium feedback control. The ammonium sensor was placed after the secondary settler, and the DO set-point was constant in the last aerobic zone.

Controller tuning 8.4.2Some minor changes to the air flow and DO controller tunings were made during the project. At Käppala WWTP, nitrification is complete most of the time apart from during rain events. Since the ammonium concentration rare-ly reaches the ammonium set-point of around 1 mg/l the ammonium control-ler was tuned to be fast despite the fact that the sensor was placed after the settler.

Himmerfjärden WWTP 8.5 Control structure 8.5.1

Himmerfjärden WWTP does not have individual control of DO in the six aerated zones, but controls the air flow rate based on the DO in the second aerated zone and there is only one air flow valve for each treatment line. This leads to a sharp increase in DO in the last zones of the basin. One treatment line at Himmerfjärden was reconstructed to include zone control of DO and ammonium feedback control. The ammonium sensor was placed in situ in the last aerobic zone.

Controller tuning 8.5.2The PI parameters in the air flow controllers and DO controllers were tuned with lambda tuning (Åström and Hägglund, 1995). An example from lambda tuning of an air flow controller is found in Figure 8.3. The ammonium con-troller was tuned manually, and the controller is relatively fast.

8.6 Summary of experiments 157

Figure 8.3. Lambda tuning of the air flow controllers in December 2011. The graph shows the air flow rate in zone 4 (blue) and 6 (black). Step-response experiments were performed in zone 4 and the new controller settings implemented in both zones at the vertical line.

Summary of experiments 8.6Outlet ammonium from the activated sludge process based on on-line meas-urements was used to estimate treatment performance at all plants, supple-mented with composite weekly lab sampling at Himmerfjärden WWTP. When calculations involved influent load, composite weekly samples taken after primary sedimentation were used. The experimental set-up including instrumentation and sampling of nitrogen compounds is found in Figure 8.4. A summary of the evaluation and controller settings is found in Table 8.1. The DO set-point limits were changed during the experiment and the maxi-mum and minimum values on the higher and lower set-point limits are given in Table 8.2. Parts of the time, Käppala WWTP had gain scheduling control, see further Chapter 9. The zone 2 controller settings are listed in Table 8.1.


Figure 8.4. Summary of experimental set-up and instrumentation for Henriksdal WWTP (left), Käppala WWTP (middle) and Himmerfjärden WWTP (right). REF = reference line, EXP = experimental line, Qin = inflow rate, TN = total nitro-gen, NH4 = ammonium.

Table 8.1. Summary of ammonium controller settings at the three plants.

Evaluation period

NH4 SP (mg/l)

K Ti (s) Reference controller

Temperature (°C)

Henriksdal 12 months 1.0 -0.07 10 000 Constant DO all zones

16.7 (9.2 – 23.8)

Käppala 8 months 0.8 -0.2 1 000 Constant DO all zones

11.5 (7.6 – 15.6)

Himmerfjärden 12 months 2.0 -1.2 1 200 Constant DO zone 1

14.2 (9.4 – 18.2)

Table 8.2. DO limits in the ammonium controllers during the full-scale experiments.

Lower min. DO limit (mg/l)

Higher min. DO limit (mg/l)

Lower max. DO limit (mg/l)

Higher max. DO limit (mg/l)

Henriksdal 1.5 2 2.5 4

Käppala 1 1.3 2 2.2

Himmerfjärden 1 1.5 2.5 3

REF REF EXP

Qin

NH4

REF EXP EXP EXP REF REF REF EXP REF REF REF REF REF

Qin Qin Qin Qin Qin

NH4 NH4 NH4 NH4 NH4 NH4 NH4 NH4 NH4

NH4 NH4

NH4 TN

NH4 TN

NH4NH4 TN

NH4 TN

On-line measurementLab measurement

HENRIKSDAL WWTP HIMMERFJÄRDEN WWTPKÄPPALA WWTP

Qin

Qin

8.7 Evaluation methods 159

Evaluation methods 8.7 Energy estimation 8.7.1

An attempt was made to quantify the energy consumption from the measured air flow rates. More information is given in the results section.

Cost-benefit analysis 8.7.2A cost-benefit analysis was carried out to look at the feasibility of using the ammonium controllers in full-scale. Reconstruction costs, installation of actuators and sensors, changes to control system and annual maintenance costs of sensors were included in the analysis. The depreciation of installa-tions and sensors were 15 years on equipment and 30 years on constructions, with an interest rate of 4 %.

Estimated energy savings (% of power consumption) was used to quantify the benefit of the ammonium feedback controllers. At Himmerfjärden WWTP part of the air to the aeration system is created from a gas motor run on biogas. A potentially higher sale of biogas was included in the benefit analysis. The benefit of improved process supervision has not been quanti-fied. A cost-benefit ratio was calculated by dividing the annual costs with the annual benefits.

Statistical testing 8.7.3Statistical testing was used to find out whether the difference between the experimental and reference line(s) was significant. The test was performed on daily averages of the DO concentration and daily (Henriksdal and Käppa-la WWTPs) and weekly (Himmerfjärden WWTP) averages of the effluent ammonium concentration. Weekly ammonium data had to be used at Him-merfjärden WWTP since the ammonium sensor was not correct enough for comparison.

A two-sided paired Student t-test with significance level of 0.05 was used (Snedecor and Cochran, 1989), with the following hypotheses:

H0: The mean difference between the experimental line(s) and the reference line(s) is zero. H1: The mean difference between the experimental line(s) and the reference line(s) is significantly different from zero.

Measurements of N2O 8.7.4During autumn and winter 2012 to 2013 a measurement campaign was set up to measure N2O emissions, with the goal to quantify the contribution of N2O emissions from the biological treatment to the carbon footprint


(Thunberg et al., 2013). Since the N2O emissions were measured in the ven-tilation shaft from one of the experimental lines in this work, the data was used to evaluate whether any connections could be drawn between the choice of control strategy and the emissions of N2O (Åmand et al., 2013b).

Ammonium controller performance 8.8Examples from ammonium controller operation at the three plants are found in Figure 8.5 and a summary of the results are presented in Table 8.3. The p-value is the probability of achieving the observed results if the null hypothe-sis is true. Since p < 0.05 for all experiments and the null hypothesis can be rejected there is a statistically significant reduction of DO concentration in the experimental lines between 12 and 55 %. The difference in ammonium concentration was not statistically significant (p > 0.05). At Henriksdal and Käppala WWTPs the ammonium peaks due to snow melting are removed from the effluent ammonium calculations since this is not a representative period of operation and the ammonium sensors are not calibrated for the high concentration range (10 to 15 mg/l) obtained during these events. At all plants, periods of malfunctioning sensors or periods when the treatment lines had operational problems were not included in the summary in Table 8.3.

The DO set-point is the same to all aerated zones at Himmerfjärden WWTP and it is the same in all but the last zone at Henriksdal and Käppala WWTPs. The difference in DO concentration for the sensor in the first and last zone at Himmerfjärden WWTP in Table 8.3 is a result of the limitation in aeration intensity in the zones.

During the evaluation period, no negative effects have been seen on the sludge properties due to ammonium control, but it is still being investigated. No impact on denitrification has been observed at Henriksdal and Käppala WWTPs.

8.8 Ammonium controller performance 161

Figure 8.5. Example of ammonium controller operation at Henriksdal, Käppala and Himmerfjärden WWTPs.

Table 8.3. Process performance at Henriksdal, Käppala and Himmerfjärden WWTPs based on monthly averages and standard deviation. NH4 is measured at two points in the reference lines at Himmerfjärden WWTP: block A (A) and block B (B). 2Second aerobic zone, 6sixth aerobic zone, Llab sample.

Dissolved oxygen Outlet NH4

Ref.

(mg/l) Exp.

(mg/l) Diff. (%)

p-value t-test

Ref. (mg/l)

Exp. (mg/l)

p-value t-test

Henriksdal 2.9 ± 0.7 2.5 ± 0.7 -12 ± 7.7 p < 0.001 1.4 ± 0.8 1.3 ± 0.7 p = 0.51 Käppala 2.0 ± 0.2 1.4 ±0.3 -31 ± 9.5 p < 0.001 0.8 ± 0.3 0.7 ± 0.4 p = 0.06 Himmer-fjärden

2.22 ± 0.4 4.96 ± 0.5

1.72 ± 0.2 2.26 ±1.3

-222 ± 10 -556 ± 8.3

p < 0.0012 p < 0.0016

2.0 ± 1.7A 1.8 ± 1.9B

2.2L ± 1.3 p = 0.39A p = 0.73B

18−Oct 23−Oct 28−Oct 02−Nov 07−Nov 12−Nov 17−Nov 22−Nov0

2

4

6

8

10

DO

, NH

4 (m

g/l)

07−Dec 17−Dec 27−Dec 06−Jan 16−Jan 26−Jan 05−Feb0

2

4

6

8

10

DO

, NH

4 (m

g/l)

27−Oct 28−Oct 29−Oct 30−Oct 31−Oct 01−Nov 02−Nov0

2

4

6

8

10

DO

, NH

4 (m

g/l)

NH4

DO

NH4

DO

NH4

DO set−point

DO zone 1

DO zone 2

DO zone 3

DO zone 4

DO zone 5

DO zone 6

HENRIKSDAL WWTP

KÄPPALA WWTP

HIMMERFJÄRDEN WWTP


Estimation of energy saving and cost-benefit 8.9analysis

At all plants there were systematic differences in air flow consumption in the treatment lines making a direct comparison between experimental and refer-ence lines difficult. The differences between the treatment lines were over 10 % at all the plants, which is in the range of expected energy savings from ammonium control.

At Henriksdal WWTP a large decrease in DO of around 40 % could result in a zero difference in air flow rate and vice versa. Since this is not reasona-ble, an analysis was made to compare each of the lines separately to each other. After normalising the air flow rate according to several periods when the DO concentration was the same in the lines, one experimental and one reference line was chosen for comparison. The air flow saving was 6.8 % over one year, for an average difference in DO of -9.0 %. A saving of 7 % results in an annual cost-benefit ratio of 0.01 since the only cost is related to small changes in the control system.

At Käppala WWTP it became clear during the experiment that one of the two experimental lines had a non-functioning air flow meter. Therefore, a period of four weeks was chosen for evaluation in the other line. The weath-er, inflow rate and temperature were stable since temperatures were below zero degrees and the ground was frozen. During this period the air flow re-duction due to ammonium control was 13 %, with a DO reduction of 40 %. The DO reduction was larger than during most of the evaluation period, making the estimate optimistic. The cost-benefit ratio for Käppala WWTP is 0.32, assuming an energy saving of 10 %. The energy reduction in terms of Nm3air/kgNH4 removed was 11 % during the four week period.

At Himmerfjärden WWTP it is known that the flow distribution between the lines is not equalised and the variation between the eight lines is large. Therefore, all seven references lines were averaged before comparisons with the experimental lines. The air flow reduction in the experimental line during the evaluation period was 15 %. The year before reconstruction the experi-mental line had a 4.7 % higher air flow rate than the average of the other lines. If this is compensated for, the air flow reduction was 19 %, resulting in an annual cost-benefit ratio of 0.31. A large part of the energy savings at Himmerfjärden WWTP is due to the implementation of DO zone control. The use of ammonium control has been estimated during a 2 week experi-ment to contribute with up to 4 % reduction in energy compared to constant DO control.

The energy estimation problem was partly different at the three plants. For Henriksdal and Himmerfjärden WWTPs the task became to try to com-pensate for systematic differences between the treatment lines, while for Käppala WWTP the task became to look for systematic differences between evaluation periods, see Figure 8.6.

8.10 Comparison to simulation results 163

Figure 8.6. When evaluating parallel treatment lines it important to be attentive to systematic differences between the treatment lines. If one treatment line but different evaluation periods is under study, differences in influent and other time varying parameters need to be considered.

Comparison to simulation results 8.10Simulations were made in Chapter 7 with the Henriksdal, Käppala and Himmerfjärden models (Appendix A). How do these simulations compare to the controllers in the full-scale experiments? In the control systems at the plant the signals are recalculated from engineering units to percentage units based on upper and lower limits of the signals. To achieve similar perfor-mance of the simulated controllers the controller gain was re-scaled:

minmax

minmax

yy

uuKK plantBSM −

−= (8.1)

where KBSM is the controller gain in the model, Kplant is the controller gain in the plant control system, umax and umin are the upper and lower limits on the DO signal, and ymax and ymin are the upper and lower limits on the ammonium signal. The integral time was the same as in the plant control systems.

After using Eq. (8.1) the simulated ammonium controllers are comparable to the full-scale controllers at Käppala WWTP and Himmerfjärden WWTP. The simulated controller gain at Henriksdal WWTP had to be multiplied by a factor of 5 to achieve satisfactory results when comparing to the full-scale controller. Figure 8.7 marks the full-scale controllers on the controller maps for 2012 for the three plants. For Käppala WWTP both controllers in the gain scheduling controllers are plotted. See further Chapter 9.

REF EXP

REF

EXP

T1 T1 T2

Parallel experimental lines One experimental line

How similar are the experimental lines?How similar are the evaluation periods?


(a) (b) (c)

Figure 8.7. Full-scale controllers compared to all the simulated controllers in the plant BSM1_LT models. The controller settings in the full-scale controllers are (a) K = -0.05 and Ti = 0.1 d for Henriksdal WWTP (b) K = -0.05, Ti = 0.05 d (orange) and K = -0.25 and Ti = 0.02 d (blue) for Käppala WWTP and (c) K = -0.25 and Ti = 0.01 d for Himmerfjärden WWTP. GS = gain scheduling.

All the full-scale controllers have a position in the controller map where the energy saving from ammonium control is within a few percent of the best achievable results in the maps. The positions of the full-scale controllers in the controller maps also give a hint of what the priorities were behind the manual tuning of the full-scale controllers. At Henriksdal WWTP the con-troller is slow to avoid too fast daily variations (see Section 8.12.3). At Himmerfjärden WWTP ammonium removal was high priority and the am-monium sensor was placed in situ in the aeration process motivating fast control.

The results from the full-scale experiments are compared to the simulated full-scale controllers in Table 8.4. The evaluation periods are not the same since the full-scale evaluation was performed mostly during 2013. Overall the simulation results are similar to the full-scale results. As mentioned in Section 7.9, the energy saving at Henriksdal WWTP depends on the constant DO set-point in the reference simulation with constant DO control. The DO maximum limits are lower in the simulations than during full-scale evalua-tion. The appreciated energy saving at the plant is within the range of the simulation results.

At Käppala WWTP the DO concentrations in the simulations are very close to the measured concentrations. The average ammonium concentra-tions are slightly higher in the simulations since very high peaks are re-moved from the full-scale data for sensor quality reasons. The estimated energy saving of around 10 % in the full-scale experiments is slightly higher than the simulation results.

The ammonium feedback controller at Himmerfjärden WWTP was never compared to constant DO control for a long period of time. The good match between simulated data and full-scale data at Himmerfjärden WWTP in

0 10 20 30 40 500.92

0.93

0.94

0.95

0.96

0.97

0.98

0.99

1

|Ki|

0 10 20 30 40 500.92

0.93

0.94

0.95

0.96

0.97

0.98

0.99

1

|Ki|

0 10 20 30 40 50

0.87

0.89

0.91

0.93

0.95

0.97

0.99

1

|Ki|

m3

air/

kgN

H4

rem

oved

com

pare

d to

con

stan

t DO

Constant DOK=−0.025K=−0.05K=−0.075K=−0.1K=−0.2K=−0.3K=−0.5K=−0.75K=−1K=−1.5K=−2Full−scale controller

Constant DO (2 mg/l)K=−0.025K=−0.05K=−0.075K=−0.1K=−0.25K=−0.5K=−0.75K=−1K=−1.5Full−scale controller (GS zone 1)Full−scale controller (GS zone 2)

Constant DOK=−0.025K=−0.05K=−0.075K=−0.1K=−0.15K=−0.25K=−0.5K=−0.75K=−1Full−scale controller


8.10 Comparison to simulation results 165

terms of DO and ammonium concentration suggests that the model’s esti-mate of the energy saving due to ammonium control of 8 to 10 % is fairly accurate. The model captures the limitations in aeration capacity in the aer-ated zones well which is seen on the resemblance in DO concentration.

The ammonium and DO profiles from simulated full-scale controllers are presented in Figure 8.8.

Table 8.4. Energy saving compared to constant DO control, mean DO concentration and mean NH4 concentration is the full-scale experiments and in plant simulations with the full-scale controllers in Figure 8.7. *Estimate. 2Second aerobic zone, 6sixth aerobic zone, Llab sample.

Full-scale (2012-2013) Simulations (2011) Simulations (2012)

Saving Qair (%)

DO (mg/l)

NH4 (mg/l)

Saving Qair (%)

DO (mg/l)

NH4 (mg/l)

Saving Qair (%)

DO (mg/l)

NH4 (mg/l)

Henriksdal 6.8* 2.5 1.3 6.0 2.0 1.1 13 2.0 1.1 Käppala 10* 1.4 0.7 8.4 1.5 1.6 7.7 1.4 1.4 Himmerfjärden - 1.72

2.26 2.2L 10 1.72

2.16 2.2 7.9 1.82

2.26 2.4


(a)

(b)

(c)

Figure 8.8. Simulated NH4 and DO profiles for the full-scale NH4 PI controllers. (a) Henriksdal model 2011, (b) Käppala model 2012 and (c) Himmerfjärden model 2012. Note: different scales on y-axes.

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

0

1

2

3

4

5

6

7

8

DO

, NH

4 (m

g/l)

Effluent NH4DO concentration zone 4 and 5


0

2

4

6

8

10

12

DO

, NH

4 (m

g/l)

Effluent NH4DO concentration zone 4−6


0

2

4

6

8

10

12

DO

, NH

4 (m

g/l)

Effluent NH4DO concentration zone 2

8.11 N2O emissions at Käppala WWTP 167

N2O emissions at Käppala WWTP 8.11The emissions of N2O are plotted together with DO concentration, influent COD, outlet ammonium and effluent nitrate in Figure 8.9.

Figure 8.9. N2O emissions in the ventilation shaft from one of the experimental lines. NO3 data from the on-line sensor at the plant effluent and NH4 data from the on-line sensor at the outlet of the experimental biological treatment line.

From previous measurement campaigns of N2O at Käppala WWTP it is known that it is difficult to correlate process conditions to the emissions of N2O. In Figure 8.9, it cannot be ruled out that the DO concentration can af-fect the concentration of N2O. However it is reasonable to assume that the large peak of N2O in late November 2012 is produced as an effect of a dis-turbance in the denitrification rather than nitrification, since the N2O emis-sions are high when effluent nitrate is high and DO concentrations are kept constant at the time. No sharp increase of N2O occurs when ammonium feedback control is switched on and the DO concentration is decreased down to 1 mg/l, but there is a decrease in concentration when the ammonium peaks occur in late December. The total flux of N2O was calculated from the con-centrations in Figure 8.9 and the Käppala emission factor was during this period 1.7 % of removed TN which is in the range of earlier published re-sults (Global Water Research Coalition, 2011).

0

20

40

60

80

100

120

0

1

2

3

4

5

6

7

8

9

10

N2O

(ppm

)

DO (m

g/l)

DO mg/l NH4 out mg/l N2O

NH4 feedback controlConstant DO control

0

20

40

60

80

100

120

0

50

100

150

200

250

300

350

NO

3 (m

g/l*

10) N

2O (p

pm)

COD

(mg/

l)

COD in mg/l N2O NO3 out mg/l


Lessons learnt from controller implementation 8.12The three major causes of delays in the project were sludge issues, the con-trol systems and sensor problems. With respect to sludge issues, delays were created when sludge scrapers broke down or during periods of poor sludge quality caused by rising sludge in the settlers. To prevent denitrification in the settler which causes rising sludge, DO concentrations were not allowed to be decreased too much, hence ammonium feedback control was avoided.

The controllers and the control systems 8.12.1The project has been working with PI controllers, the most widely used con-troller for process control. Despite this, considerable time was spent to get the controller configuration correct. In the default controller implementation, all control signals did not have a maximum and minimum limitation which is crucial for a good control performance. At Henriksdal and Himmerfjärden WWTPs the integration time was initially not allowed to be large enough due to a limit set in the software. Ammonium has a long response time to a DO change, which means integration times should be long.

Himmerfjärden WWTP does not have an industrial control system but use stand-alone PID controllers connected to a SCADA system. Therefore track-ing of data between controllers was not available resulting in wind-up in the master controller. If the air flow rate controller became saturated the DO controller experienced wind-up due to integration. This wind-up leads to unnecessary aeration since the DO concentration stays high for a period after the air flow controller stopped being saturated. An example of cascade wind-up for Himmerfjärden WWTP is found in Figure 8.10. The following points are marked out in the plot: 1. The DO concentration reaches its set-point, but due to wind-up in the

DO master controller the control signal to the air flow controller is kept high and the DO concentration continues to rise;

2. The valve position is finally lowered when the air flow set-point is de-creased by the DO controller and the DO concentration eventually starts to decrease;

3. The DO concentration is back at its set-point.

Another example of cascade windup is shown in Figure 8.11, where the valve is saturated most periods of the day. During the short time where the valve is not saturated there is a DO peak due to master controller wind-up. The ammonium feedback controller is active which can be seen on the vary-ing DO set-point. With this very poor control authority this is an example of when process control cannot improve process performance due to process design limitations.

8.12 Lessons learnt from controller implementation 169

Figure 8.10. Example of cascade windup in the master DO controller at Him-merfjärden WWTP in November 2013. Data from zone 3 every 5 minutes.

Figure 8.11. Example of several days of cascade windup in the master DO controller at Himmerfjärden WWTP in May 2013. Data from zone 4 every 5 minutes.

Attempts were made to limit the maximum air flow rate at a level which would prevent master controller wind-up but this did not fully solve the problem. Treatment performance was during 2013 high priority which meant conservative air flow limits was avoided. The air flow controllers were re-moved from the first three zones with highest air flows during the project. The air flow meters were suspected to throttle the air flow rate. Control of the DO concentration where the DO controller output is the valve set-point has worked satisfactorily, though control performance still suffers from limi-tations due to control signal saturation.

The ammonium sensor performance 8.12.2Despite weekly maintenance, the in situ ammonium sensor at Him-merfjärden WWTP had the poorest performance of the ammonium sensors in the project when comparing to lab samples. Most often the sensor showed too high values (Figure 8.12), which meant that nitrification was not at risk.

When the ammonium sensors at Käppala WWTP were calibrated at very low ammonium concentrations (i.e. normal operation) the concentrations at the peaks were exaggerated by the sensors. This exaggeration was more pro-nounced in one of the two experimental lines (BB11). Figure 8.13 shows

12:00 13:00 14:00 15:00 16:00 17:00 18:00 19:000

1

2

3

4

DO

(m

g/l)

200

400

600

800

1000

Qai

r (N

m3/

h), V

alve

pos

ition

(10

*%)

DO set−point DO concentration Air flow rate Valve opening

21 3

00:00 12:00 00:00 12:00 00:00 12:000

1

2

3

4

5

DO

(m

g/l)

200

400

600

800

1000

Qai

r (N

m3/

h), V

alve

pos

ition

(10

*%)

DO set−point DO concentration Air flow rate Valve opening


examples of ammonium sensor performance at Käppala WWTP in January 2014. The exaggerated ammonium peaks as well as the shift after the peaks motivates gain scheduling control, see further Chapter 9.

Figure 8.12. Monthly mean concentrations of NH4 measured by weekly composite lab samples and the in situ sensor placed in the last aerobic zone in the experimental line at Himmerfjärden WWTP.

Figure 8.13. Comparison between the two NH4 sensors in the project at Käppala WWTP in January 2014. The period includes several sensor errors: noise, drift, shift and exaggeration of peaks. The pump failure was a broken RAS pump causing a lowering of the water level in the settler leaving the NH4 sensor dry.

The ammonium sensors at Henriksdal WWTP were most often working sat-isfactorily during the evaluation period. An example from the quality control of the ammonium sensors performed on a weekly basis is shown in Figure 8.14.

0

1

2

3

4

5

6

7

8

Sep-12 Oct-12 Nov-12 Dec-12 Jan-13 Feb-13 Mar-13 Apr-13 May-13

NH4

(mg/

l)

NH4 in situ sensor Lab NH4-N

01−Jan 03−Jan 05−Jan 07−Jan 09−Jan 11−Jan 13−Jan 15−Jan 17−Jan

0

2

4

6

8

10

NH

4 (m

g/l)

NH4 BB10 NH4 BB11

Sensor drift andhigh noise level

Calibration BB11

Sensor BB11 oftenexaggerates NH4 peaks

Calibration BB10

CalibrationBB10+BB11

Shift after NH4 peak

Pump failure

8.12 Lessons learnt from controller implementation 171

Figure 8.14. Example of NH4 sensor quality control at Henriksdal WWTP 2012 and 2013. Grab samples analysed at the lab compared to NH4 sensor value of the grab sample. Red points: sensor calibration took place.

The quality of the ammonium sensors have limited the performance of the ammonium controllers, hence the results in the full-scale study are in spite of the sensor problems. Other sources report that it is possible to make the sen-sors work satisfactory in situ at the plants (Rieger and Siegrist, 2002; Kaelin et al., 2008).

The placement of the ammonium sensor 8.12.3The ammonium sensors at Henriksdal and Käppala WWTPs are placed after the secondary settler. The standard procedure for ammonium feedback con-trol is to place the sensor in situ, commonly in the last aerated zone. By do-ing so a time delay of a few hours is created, compared to the ammonium concentration in the first aerated zone. If the sensor is placed after the settler there is an additional time delay in the signal due to the hydraulic retention time (HRT) in the settler. Käppala WWTP took the decision several years ago to move the sensors from the activated sludge basin due to problems with the sensor quality and time consuming sensor maintenance. Henriksdal WWTP had ammonium sensors at the settler outlet since several years back and took the decision during the cause of the project not to move the sensors to the aeration tanks to avoid added maintenance.

During winter 2012 different PI controller settings in the ammonium con-trollers were tested at Henriksdal WWTP. Figure 8.15 shows an example from February where fast control is switched on (February 10) in treatment line 5. The DO and ammonium concentrations are similar at this point in all the treatment lines. In line 5 the ammonium peaks are most often higher than in the two other lines until around February 16 when the DO concentrations in the three treatment lines are similar.

0

1

2

3

4

5

6

7

8

9

10

NH4

(mg/

l)

Lab analysis Sensor value


Figure 8.15. Comparison between fast and slow NH4 control at Henriksdal WWTP in 2012. Treatment line 5 (fast control) compared to treatment line 3 and 4 (slow control). Minimum DO concentration: 1 mg/l.

It is expected that this increased ammonium peak is caused by the decreased DO concentration being delayed compared to the ammonium load to the aeration tank. The HRT in the secondary sedimentation tanks is around four to five hours during normal flows. The lowest inflow rate to the plant arrives around 7 am (Figure 4.11). Five hours later the inflow rate has tripled and its maximum is reached around 2 pm. The DO concentrations during fast am-monium control is at its lowest point when the load is near its peak which can provoke an increase in the peak measured a few hours later by the am-monium sensor after the sedimentation tanks. HRT in the sedimentation tanks at Käppala WWTP is five to six hours, but due to the smaller variation in load (Figure 4.11) the effect of the time delay should be smaller.

There is an inconsistency in Figure 8.15. The minimum DO concentration in line 5 is the same as the average concentration in line 3, and despite this the ammonium concentrations are in average lower in line 3 than in line 5. Given uncertainties about possible unknown differences between the three treatment lines in Figure 8.15, Figure 8.15 should not be seen as a proof of the hypothesis that fast ammonium control increase the daily ammonium peaks. The figure does however suggest that fast ammonium control at Hen-riksdal WWTP cannot reduce the ammonium peaks. Also, the average DO concentration during the period in Figure 8.15 was higher with fast ammoni-um control than with slower control, which was discussed in relation to the plant simulations in Chapter 7. The increased peaks during fast control did not occur in the simulations in Chapter 7.

10−Feb 12−Feb 14−Feb 16−Feb0

2

4

6

8

DO

, NH

4 (m

g/l)

NH4 block 3 DO block 3 NH4 block 5 DO block 5

Slow NH4control

Fast NH4 control line 5, slow line 3, 4Max DO line 5: 4 mg/l

Fast NH4 control line 5, slow line 3, 4Max DO line 5: 3 mg/l

10−Feb 12−Feb 14−Feb 16−Feb0

2

4

6

8

DO

, NH

4 (m

g/l)

NH4 block 4 DO block 4 NH4 block 5 DO block 5

8.13 Time aspects 173

Time aspects 8.13For how long should a control experiment be carried out? Here it is useful to think about cycles. There are three main repetitive cycles at a treatment plant: daily, weekly and annual variations. Covering daily and weekly varia-tions in a control experiment could be useful from a control engineering point of view – the experiment can give information about the dynamics of the controller and the response to short-time disturbances. But from an ener-gy point of view the most important cycle to cover is the annual cycle.

Due to the effect of load, flow distribution, temperature and the perfor-mance of equipment the potential to save energy from ammonium feedback control varies over the year. Therefore the results from a control experiment are more vulnerable if short time periods are considered. One example is presented in Figure 8.16 showing DO concentrations at Henriksdal WWTP during the experiments. The main factor influencing the difference in DO concentration between experimental and reference lines is variations in the operation of the reference lines. In fact, during April and May 2013, there is no statistically significant difference between the two control strategies (p = 0.61). In around 75 % of published full-scale or pilot-scale studies on ammonium-based control the time of experiment is two months or less, see further Chapter 3.

Figure 8.16. Mean daily DO concentration in reference lines and experimental lines at Henriksdal WWTP during the one year experiment.

During the experiment at Käppala WWTP, ammonium control contributes with a consistent reduction of DO concentration of between 0.7 and 0.8 mg/l during all months except during peak-flow periods such as snow melting. Under such circumstances a reasonable energy assessment can be made with a shorter evaluation period.

0

1

2

3

4

5

DO (m

g/l)

Mean DO reference lines Mean DO experimental lines


Energy estimation 8.14How should the effect of a control strategy be quantified in terms of energy? The most straight-forward way should be to look at the air flow rate or pow-er consumption, but as this chapter has shown this is often not simple in reality. Therefore, there is seldom a true number representing the energy saving from a full-scale control strategy comparison. If a precise estimate of the difference in energy demand between control strategies is needed, a well-calibrated dynamic model can be expected to give a better estimation.

If air flow rate measurements should be used as the main measure of en-ergy cost, there should be a linear relationship between the power consump-tion and the air flow rate, and the energy tariff should be flat, i.e. no daily peak charge. These two criteria have been met at the plants in this study.

When looking at energy consumption during evaluation of control strate-gies, the core task is to only capture effects on energy consumption that could be related to changes caused by implementing a different control strat-egy. It is tempting to relate the air flow rate to variations in the process. Fac-tors that could impact the energy consumption are removal rate of e.g. am-monium, inflow and suspended solid concentration in the aeration tank. By normalising with any of these factors, one could impose a correlation on the data not related to the control strategy itself.

As an example, there is a correlation between the influent load and the removed amount of ammonium (Figure 8.17). It costs relatively less to treat more nitrogen. This type of normalisations should be used with care since if two points in Figure 8.17 are compared the comparison will not only include the effect of the choice of control strategy, but also the effect of the different load situations. Due to this reason, one week was removed from the estima-tion of energy consumption at Käppala WWTP since the influent ammonium concentration was around 10 mg/l lower than during the other four weeks. If this week was included the saving from using ammonium feedback control was only 1 % based on specific air flow rate (Nm3/kgNH4 removed) despite the large reduction in DO concentration.

8.15 Conclusions 175

Figure 8.17. Specific air flow rate (Eq. (6.2)) as a function of NH4 load for two cold and two warm months. Data from Henriksdal WWTP.

Conclusions 8.15Ammonium feedback control has been operated for up to a year at Henriks-dal, Käppala and Himmerfjärden WWTPs. The estimated energy saving is in the range of 7 to 19 % and the cost-benefit analyses suggest it is beneficial to implement ammonium feedback control at all plants. The long evaluation period of the experiments was necessary to have a fair comparison between the experimental and reference lines. Statistical testing is a rarely used tool in relation to full-scale aeration control evaluations. In this study it proved a helpful method to judge if the choice of control strategy had an impact on the DO and ammonium concentrations. In the data analysis from full-scale experiments it is important to be attentive to circumstances that add systematic differences to the experimental treat-ment lines and that do not relate to the control strategies under study. Quan-tifying the energy savings potential from air flow measurements is challeng-ing in full-scale plants.

0

50

100

150

200

250

100 200 300 400 500

Spec

ific a

ir flo

w ra

te (N

m3 /

kgN

H 4)

NH4 load (kg/h)Sep. 2009 Nov. 2009Jan. 2010 Aug. 2010


There are benefits from implementing cascade PID controllers in an indus-trial control system compared to working with stand-alone PID controllers in cascade. Master controller wind-up can then be avoided. Measurements of N2O in the ventilation shaft at Käppala WWTP suggest the main cause of very high N2O emissions is a disturbance in the denitrification process. Low DO concentrations cannot be ruled out as a cause to increased N2O emissions. The performance of the ammonium sensors have been an issue at two of the plants. The ammonium concentration was often too high which means ener-gy consumption rather than treatment performance was compromised. Drift, shift, noise and an exaggeration of ammonium peaks have been the main sensor errors. Ammonium control with the ammonium sensor placed after the settler has proven to be a feasible alternative in the full-scale evaluations. Placing the ammonium sensor after the secondary sedimentation tank creates a time delay in the sensor signal. The lowest DO concentration occurred near peak load at Henriksdal WWTP when fast ammonium control was used. This was suspected to increase the daily ammonium peaks which is why slower am-monium control was chosen. During the project the plant personnel have been satisfied with the control-lers, helping them take the mind of the DO set-points in the process and re-acting quickly to sudden disturbances in the process. Henriksdal and Käppa-la WWTPs have decided to install ammonium control in all their treatment lines in the near future. Himmerfjärden WWTP will be reconstructed within the next few years and no changes to the aeration control system will be made until then.

PART IV

IMPROVING AMMONIUM FEEDBACK CONTROL

179

9 GAIN SCHEDULING CONTROL 9

AIN SCHEDULING CONTROL is a method to create a non-linear controller by combining a set of linear controllers. In this chapter, gain scheduling (GS) was used to schedule between two PI ammo-

nium controllers. The effluent ammonium concentration decided which of the controllers to use. GS was implemented in full-scale at Käppala WWTP, and was evaluated in the Käppala simulation model. The overall aim with GS was to: (1) avoid high DO concentrations when these are not necessary and (2) avoid fast ammonium control when this is not required. The tenden-cy of the ammonium sensor to exaggerate the ammonium concentration as well as the high cost at high aeration intensities motivates the first aim and the placement of the ammonium sensor motivates the second aim.

Introduction 9.1Gain scheduling refers to a set of linear controllers where a process parame-ter is chosen as scheduling variable and the controller is selected depending on the level of the this variable (Rugh and Shamma, 2000). This creates a non-linear controller. Within process control, a common example is schedul-ing of PID-controller parameters in a non-linear process. An example of a version of gain scheduling in wastewater treatment is Gerkšič et al. (2006). The paper schedules the PI controller parameters in a DO controller. A local linearisation is performed around constant values of the DO concentration and KLa. In simulations, the scheduling controller had less overshoot when the model was subject to changes in the DO set-point. In the pilot plant eval-uation of the controller the results were obscured by measurement noise.

In this study the approach has not involved linearisation of a process model. Instead gain scheduling was used to change the settings in the am-monium controller at high ammonium concentrations. The controllers were

G

180 9 Gain scheduling control

tuned manually. The PI controller parameters as well as the upper limit of the control signal from the ammonium controller, i.e. the DO set-point, were scheduled.

Control structure and control strategies 9.2The control structure was cascade control with an air flow rate controller (slave), DO controller (master/slave) and an ammonium controller (master) as depicted in Figure 8.1. The ammonium sensor was placed after the sec-ondary settler at Käppala WWTP to minimise maintenance of the ammoni-um sensor. The basic controller implementation was as follows, using two controller zones:

+

+

=

,)(

1)(

,)(1

)(

)(

02,

2

01,

1

t

i

t

i

deT

teK

deT

teK

tu

ττ

ττ

2max,min

1max,min

)(

)(

utuu

utuu

<<

<<

if

if

ZLtSV

ZLtSV

>

<

)(

)(

(9.1)

where u(t) is the control signal from the ammonium controller (i.e. the DO set-point), e(t) is the control error, K1, Ti,1, K2 and Ti,2 are the PI controller parameters in the two respective zones, umin is the fixed lower control signal limit, umax,1 and umax,2 are the upper control signal limits in the two respective zones, SV(t) is the scheduling variable and ZL is the zone limit deciding when the controller should switch to another zone.

Full-scale evaluation at Käppala WWTP 9.3

Method 9.3.1The motivation behind the experiments with gain scheduling at Käppala WWTP was to study what was achievable in the plant control system and to select the best settings in the gain scheduling controller.

This study has evaluated three versions of gain scheduling in full-scale operation (Figure 9.1). The first controller (GS F1) used the ammonium con-centration directly as the scheduling variable while the second controller (GS F2) used the DO concentration in the first aerated zone. Since the DO set-point was controlled in closed-loop by the ammonium controller, the DO concentration will eventually rise when ammonium concentration increase. These two controllers scheduled the controller gain and integral time. The third controller (GS F3) used the ammonium concentration as scheduling

9.3 Full-scale evaluation at Käppala WWTP 181

variable and scheduled the controller gain, integral time and also the upper DO set-point limit.

Figure 9.1. The three gain scheduling controllers investigated at Käppala WWTP. The zone 1 controller was slower than the zone 2 controller. SV = scheduling varia-ble.

The controller settings in the second controller zone were manually to be fast and therefore have a quick reaction to a change in the ammonium concentra-tion. The controller was tuned to be slower in the first controller zone to avoid costly aeration when the effluent ammonium concentration was rela-tively low. Also, when the ammonium sensor is placed after the settler the sensor will read a delayed signal compared to the actual concentration in the aeration basin motivating less aggressive control. In zone 1, fast daily varia-tions in DO concentration was avoided.

The controller settings in the ammonium controller at Käppala WWTP are given in Table 9.1. The upper and lower DO limits in GS F1 and GS F2 were the same as during ammonium PI control without GS. The upper DO limit in zone 1 in GS F3 was decided in communication with the plant personnel. The lower limit on DO is slightly higher in GS F3 since the evaluation was performed during a colder period of the year.

Table 9.1. Controller settings at Käppala WWTP in the three gain scheduling con-trollers. F in GS F# stands for full-scale, ZL = zone limit.

Contr. zone

NH4 set-point (mg/l)

ZLNH4 (mg/l)

ZLDO (mg/l)

K Ti (s)

umin umax

GS F1 1 0.8 3 -0.05 4 500 1.2 2.2

2 0.8 3 -0.2 1 000 1.2 2.2

GS F2 1 0.8 1.5 -0.05 4 500 1.2 2.2

2 0.8 1.5 -0.2 1 000 1.2 2.2

GS F3 1 0.8 3 -0.05 4 500 1.3 1.8

2 0.8 3 -0.2 1 000 1.3 2.3

Gain scheduling was implemented in the two experimental lines at Käppala WWTP (Section 8.6) by configuring and switching on the gain scheduling extension C6 in the PIDCONA PC element (ABB Industrial Systems, 1998) in the Käppala WWTP control system (ABB 800xA). Extension C5 was switched on to allow for an externally supplied limit on the control signal which was needed to schedule umax in GS F3.


Results and discussion 9.3.2Gain scheduling with different settings has been operated for over a year at Käppala WWTP. An example from gain scheduling based on the effluent ammonium concentration is found in Figure 9.2. The slow controller is switched to a faster controller on October 18 when the ammonium passes ZLNH4. The negative aspect of this controller was a slow decrease of the DO concentration after an ammonium peak. The ammonium concentration passed ZLNH4 and the controller was again slow, but the DO concentration was high despite no ammonium in the effluent.

Figure 9.2. Gain scheduling with GS F1 at Käppala WWTP (15 min data). NH4 concentration as scheduling variable.

To avoid the slow decrease in DO after an ammonium peak, the controller was re-programmed, making the DO concentration in the first aerobic zone the scheduling variable. This way, the controller was fast as long as the DO concentration was high, as can be seen in Figure 9.3. The key experience from the DO scheduling controller was that the DO zone limit should be updated relatively often (a couple of times per month). The operator would have to develop a feeling for how to change the zone limit based on the pro-cess state unless an automatic procedure could be developed for this pur-pose. However, the zone limit cannot be changed in the face plate (operator window) of the controller, but only in the PC element, which requires the involvement of a programmer.

Figure 9.3. Gain scheduling with GS F2 at Käppala WWTP (15 min data). DO con-centration as scheduling variable.

09−Oct 11−Oct 13−Oct 15−Oct 17−Oct 19−Oct 21−Oct

0

2

4

6

DO

, NH

4 (m

g/l)

DO aerobic zone 1NH4ZL

NH4

20−May 22−May 24−May 26−May 28−May 30−May 01−Jun 03−Jun

0

2

4

6

DO

, NH

4 (m

g/l)

DO aerobic zone 1

NH4

ZLDO

9.3 Full-scale evaluation at Käppala WWTP 183

One negative aspect of the DO scheduling controller was that at times the controller was made fast without need. Since the ammonium controller is integrating, DO increases slowly over time which can trigger a zone shift even at low ammonium concentrations. An example is given in Figure 9.4 where the DO concentration was increased to its upper limit when the con-centration passed ZLDO even though the ammonium just barely reaches above its set-point of 0.8 mg/l.

Figure 9.4. Example of when the DO-scheduling GS controller increases the DO concentration without need.

The final controller evaluated in full-scale was the controller which sched-uled the PI controller parameters and the upper limit of the DO set-point. An example of the behaviour of GS F3 compared to GS F2 is found in Figure 9.5. The high ammonium concentrations were due to snow melting in early January 2014. The ammonium concentration was expected to be exaggerated in the treatment line where GS F3 operates (see Section 8.12.2). Despite this the average DO concentration was in average lower in the treatment line with GS F3.

When the ammonium concentration passed the zone limit the ammonium controller became fast, and the upper DO limit was increased. During Janu-ary 12 and January 15 the ammonium concentration for GS F3 was for a period above its set-point of 0.8 mg/l but below the zone limit of 3 mg/l and the upper DO limit was decreased.

After the first ammonium peak the ammonium sensors experienced a shift of the signal which was corrected when the sensors were calibrated on Janu-ary 15. The shift was higher in the sensor used in the controller with GS F3. This is common behaviour of the ammonium sensors after high ammonium peaks. The sensor often settles at a value above the ammonium set-point but below 3 mg/l which is the ZLNH4. By scheduling the upper DO set-point, unnecessary aeration can be avoided during periods of sensor shifts by limit-ing the DO concentration to a lower value.

To evaluate the effect of gain scheduling on treatment results and energy consumption is difficult in full-scale operation due to measurement errors and differences between parallel treatment lines. This motivates quantitative comparisons using a process model.

02−Mar 07−Mar 12−Mar 17−Mar

0

1

2

3

DO

, NH

4 (m

g/l)

DO aerobic zone 1NH4ZL

DO


Figure 9.5. Gain scheduling with GS F2 and GS F3 at Käppala WWTP (15 min data). NH4 concentration as scheduling variable.

Simulations with gain scheduling control 9.4 Method 9.4.1

The motivation behind the simulations with the plant model was to further improve the results from the full-scale experiments and to quantify the effect of gain scheduling on the energy consumption and on the treatment results.

Three types of gain scheduling ammonium controllers were investigated by simulations in the Käppala simulation model (see Appendix A). The first controller was a combination of GS F1 and GS F2 investigated in full-scale (GS S1, Figure 9.6). The second controller (GS S2) was equal to GS F3. The third controller was a combination of GS S1 and GS S2, i.e. the scheduling was based on the ammonium concentration and the DO concentration. In GS S2 and S3 the upper limit on the control signal, controller gain and integral time were scheduled.

Figure 9.6. The three gain scheduling controllers investigated in the Käppala simula-tion model. GS S1 is a combination of GS F1 and GS F2. GS S2 equals GS F3 and GS S3 is a combination of GS S1 and GS S2.

The controller settings in the simulations are presented in Table 9.2. The ammonium controller settings in the simulator were recalculated from the settings in Table 9.1 according to Eq. (8.1). The DO limits were similar to

09−Jan 11−Jan 13−Jan 15−Jan 17−Jan

0

0.5

1

1.5

2

2.5

3

DO

(m

g/l)

0

2

4

6

8

10

NH

4 (m

g/l)

DO GS F2 DO GS F3 NH4 set−point

GS zone limit NH4 GS F2 NH4 GS F3

9.4 Simulations with gain scheduling control 185

the limits in the full-scale controllers. The zone limits are the same as in the full-scale experiments. Bumpless transfer was implemented in the model in order to achieve a smooth switch between the two controller zones. The im-plementation was similar to back-calculation anti-windup (Figure 5.3).

Simulations were performed for year 2012. The full-scale experiments were performed after the simulation period, but the simulation results were assumed to be similar enough for a comparison.

Table 9.2. Controller settings in the Käppala simulation model in the three GS con-trollers. S in GS S# stands for simulation.

Contr. zone

NH4 set-point (mg/l)

ZLNH4 (mg/l)

ZLDO (mg/l)

K Ti (d)

umin umax

GS S1 1 0.8 3 1.5 -0.05 0.05 1.2 2.2

2 0.8 3 1.5 -0.2 0.015 1.2 2.2

GS S2 1 0.8 3 -0.05 0.05 1.2 1.7

2 0.8 3 -0.2 0.015 1.2 2.2

GS S3 1 0.8 3 1.5 -0.05 0.05 1.2 1.7

2 0.8 3 1.5 -0.2 0.015 1.2 2.2

Results and discussion 9.4.2The gain scheduling controllers were compared with two reference control-lers: a simulation with constant DO concentrations and a zone 1 ammonium controller without gain scheduling and with an upper DO limit of 2.2 mg/l. The DO concentrations from these controllers are found in Figure 9.7. The ammonium profile with constant DO control is presented in Figure 9.8.

A close-up from October, November and December of the DO concentra-tions in the three modelled gain scheduling ammonium controllers are found in Figure 9.9 to Figure 9.11. The full year is plotted for GS S3 in Figure 9.11.

When scheduling on both DO and ammonium concentration (GS S1 and GS S3), switching from zone 1 to zone 2 was performed when the ammoni-um concentration crossed ZLNH4 in the simulations. To make sure that this is always how it is done, a rule could be included where the change from zone 1 to zone 2 always is done based on ammonium concentration and not based on the DO concentration, in order to avoid the situation occurring in Figure 9.4.


Figure 9.7. DO concentrations in the two reference controllers. The NH4 FB zone 1 refers to the controller in zone 1 (K = -0.05 ant Ti = 0.05 d).

Figure 9.8 Effluent NH4 profile during 2012 with NH4 set-point and ZLNH4 marked. Data from a simulation with DO set-point of 2 mg/l.

Figure 9.9. Simulated DO concentration with GS S1 from October to December. Scheduling on NH4 and DO concentration.


1

1.5

2

2.5D

O (

mg/

l)

NH4 FB zone 1

Constant DO control


0

2

4

6

8

10

12

14

NH

4 (m

g/l)

NH4 concentraiton

ZLNH4

NH4 set−point

01−Oct 15−Oct 01−Nov 15−Nov 01−Dec 15−Dec

1

1.5

2

2.5

DO

(m

g/l)

DO concentration ZLDO

9.4 Simulations with gain scheduling control 187

Figure 9.10. Simulated DO concentration with GS S2 from October to December. Scheduling on NH4, different upper DO limits in zone 1 and 2.

Figure 9.11. Simulated DO concentration with GS S3 from October to December. Scheduling on NH4 and DO, different upper DO limits in zone 1 and 2. 3 month zoom (upper plot) and the whole of 2012 (lower plot).


1

1.5

2

2.5

DO

(m

g/l)


1

1.5

2

2.5

DO

(m

g/l)



1

1.5

2

2.5

DO

(m

g/l)



The three gain scheduling controllers achieve the goal of increased aeration intensity during high ammonium concentrations. Through scheduling on the DO concentration (GS S1 and GS S3), the slow decrease in DO after the peak as seen in Figure 9.2 was avoided thereby minimising energy loss. Through scheduling the upper limit of the control signal and not only the gain and integral time (GS S2 and GS S3); the DO concentration is only high when motivated by higher ammonium concentrations.

The three gain scheduling controllers are in some respects similar to hav-ing a look-up table to decide the DO set-point based on fixed levels of the ammonium concentration. The gain scheduling controllers are however smoother than control based on a look-up table and also offers less wear and tear on the control valves and blowers.

A summary of ammonium concentrations, DO concentrations and energy consumption (measured as air flow rate) is given in Table 9.3. The constant DO controller had lower average and maximum ammonium concentrations since the DO concentration was high at all times but at the cost of high air flow rates.

At the ten peak ammonium events during 2012 when the Zone 1 control-ler reached its upper limit in Figure 9.7, the best performance was found with the GS S3 controller, with a 6.7 % energy saving compared to the con-stant DO controller and a 4.2 % saving compared to the Zone 1 ammonium controller (Table 9.3). The average reduction over the whole year was 11.4 % compared to constant DO control and 1.3 % compared to the Zone 1 ammonium controller for the GS3 controller. The effluent ammonium con-centration was slightly higher when using ammonium feedback control.

Table 9.3. Summary of simulation results. Constant DO control and slow NH4 FB are reference controllers. FB = feedback, Qair = air flow rate.

Mean NH4 Max NH4 Mean DO Peak NH4 Qair

reduction Controller (mg/l) (mg/l) (mg/l) (%)

Constant DO control 1.23 12.35 2.20

NH4 FB zone 1 1.34 12.45 1.49 -2.5

GS S1 1.23 12.45 1.41 -4.6

GS S2 1.26 12.50 1.38 -6.2

GS S3 1.27 12.50 1.36 -6.7

9.5 Conclusions 189

Conclusions 9.5Gain scheduling was investigated at Käppala WWTP in full-scale operation and in a simulation study. The full-scale evaluation showed that switching from a slow to fast controller zone was best managed by scheduling based on ammonium concentration, but switching from a fast to slow zone was better to do with DO scheduling.

Simulations showed that gain scheduling offered the best possibilities in terms of energy saving if not only the PI controller gain and integral time were scheduled, but also the upper limit of the DO set-point. This controller was proven to operate satisfactorily at Käppala WWTP if an additional ex-tension in the PID controller was switched on. With gain scheduling, the aim to increase the aeration intensity at peak load but avoid high DO concentrations during low load could be achieved through a relatively simple change to the already implemented ammonium PI controller. Presently it is an obstacle that the zone limit in the scheduling controller cannot be changed by the operators but only by a programmer. During periods when gain scheduling was active, the controller could de-crease the energy consumption compared to regular ammonium feedback control with around 4 %. The annual reduction in aeration energy will de-pend on how often ammonium control is active and how often the controller shift between the zones. In the Käppala simulation model the annual saving was around 1 % compared to ammonium feedback control without gain scheduling.

191

10 DO DEVIATION CONTROL 10

O DEVIATION CONTROL is a concept originally used at Käppala WWTP before the start of this research project. The method was modified to fit the process at Henriksdal WWTP during spring

2012. The basis of the DO deviation controller is straight-forward: When the DO concentration deviates from its set-point in the last aerobic zone – change the DO set-point in the earlier zones. Since the controller uses the DO concentration in the aeration tank as an indication of aeration require-ment rather than the ammonium concentration measured after the secondary settler, the delayed response caused by the ammonium sensor placement discussed in Chapter 8 is avoided.

Introduction 10.1The DO deviation controller was developed at Käppala WWTP in collabora-tion with Uppsala University in 2007 (Thunberg et al., 2009). The purpose was to reduce aeration intensity during low loaded periods when DO peaks in the last aerated zone occurred. DO peaks occur when more oxygen is add-ed to the process than what is consumed by the microorganisms. The con-troller is no longer in operation at Käppala WWTP and the plant personnel experienced that it did not interact well with the ammonium feedback con-troller. The controller was implemented at Henriksdal WWTP in 2012 and changes were made to the settings compared to the settings at Käppala WWTP to improve its performance and adapt it for the specific requirements at Henriksdal WWTP.

A related method was developed by Ekman et al. (2006). As mentioned in Section 3.5.2, Ekman et al. (2006) developed a method for aerobic volume control that only required measurements of the DO concentration. The DO set-points in two aerated zones were decided based on the deviation of the

D

192 10 DO deviation control

DO concentration from the fixed DO set-point in a third zone. The controller was tested in a pilot-plant but appeared difficult to tune.

Control structure and control strategies 10.2The control structure was similar to ammonium feedback control but with an extra control loop for the DO deviation controller. The control structure is presented in Figure 10.1. There are three aerated zones at Henriksdal WWTP, zone 4 to 6 (see Section 4.4.3).

Figure 10.1. Block diagram with NH4 feedback control and DO deviation control according to the implementation at Henriksdal WWTP. DO4, DO5 and DO6 are the DO concentrations in zone 4, 5 and 6 in the activated sludge tanks at Henriksdal WWTP, respectively. NH4 was measured after the secondary settler.

The two controllers in Figure 10.2 change the DO set-point in the first two aerated zones according to the effluent ammonium concentration (NH4 feed-back controller) and according to the deviation of the DO concentration from its set-point in the last aerobic zone (DO deviation controller). When the DO concentration is higher than the set-point in zone 6 (DO6SP - DO6 < 0), the DO6 control error is negative and the DO set-point in zone 4 and 5 is de-creased.

The set-point of the DO6 control error (DO6,DEVSP) is fixed. The set-point can be changed to allow for a less aggressive control. If the DO6,DEVSP is lower than zero the DO deviation controller will not act until the zone 6 DO concentration is this distance higher than its set-point, and therefore the con-troller can avoid reacting to noise in the DO sensor. An option would be to use a low pass filter on the DO6 signal, but this would also add a delay to the response of the deviation controller.

The DO deviation controller has the effect that the incoming ammonium load is pushed further towards the end of the process at DO peaks, since the DO set-point in the first zones is reduced. In this way, the DO peak in the last zone and the overall aeration intensity can be reduced. Reduced DO

10.3 Full-scale evaluation at Henriksdal WWTP 193

peaks in the last aerated zone could imply less oxygen is recirculated back to the anoxic zones which could be beneficial for denitrification.

At Henriksdal WWTP the ammonium controller is relatively slow not to risk low DO concentrations during peak load since the ammonium sensor signal is delayed approximately 4 to 5 hours due to its position after the set-tler (see Section 8.12.3). At the plant, DO peaks are common in the last aer-obic zone when this zone is aerated.

The controller was not allowed to increase the DO concentration at peri-ods when the DO concentration was lower in zone 6 than the set-point; hence the maximum output from the controller was zero. If the DO concen-tration in the last zone is lower than its set-point it is likely that it is due to a lack of aeration capacity. Increasing the DO set-point for zone 4 and 5 would not be helpful in such a situation.

The two PI controllers in Figure 10.4 have anti-windup, but there is in theory no protection against windup of the ammonium controller as a result of changes made to the DO set-point by the DO deviation controller. Fortu-nately, it is unlikely that the ammonium controller will experience windup. For windup to occur, the contribution of the DO deviation controller would have to have an effect on the ammonium concentration. For instance, if the DO deviation controller contributed to an increase in ammonium concentra-tion, the ammonium PI controller would start increasing the DO set-point, which would counteract the effect of the DO deviation controller. However, the DO deviation controller is only active during periods with complete nitri-fication, since this is what triggers the DO peaks. The effect of the deviation controller is merely to reallocate where in the treatment line nitrification occurs. The controller does not impact the effluent ammonium concentration itself. Also, as soon as more ammonium is left for the last zone to nitrify, the DO peak would not occur, meaning the DO deviation controller would be passive. Hence, the implementation of the DO deviation controller in Figure 10.1 is protected against windup in the ammonium PI controller due to changes made by the DO deviation controller.

Full-scale evaluation at Henriksdal WWTP 10.3 Method 10.3.1

The DO deviation controller in Figure 10.3 was implemented at Henriksdal WWTP during spring 2012. The controller settings are given in Table 10.1. The combined DO set-point from the two controllers had to stay within the limits of the ammonium controller. The maximum contribution from the deviation controller is a reduction of the DO set-point by 0.5 mg/l. The DO deviation controller starts to act as soon as the DO concentration in the last


aerated zone is 0.3 mg/l above its set-point. The DO deviation controller was tuned manually.

Table 10.1. Controller settings in the NH4 controller and the DO deviation controller at Henriksdal WWTP.

K Ti (s) Set-point

(mg/l) umax

(mg/l)umin

(mg/l) DO deviation controller 0.1 800 -0.3 0 -0.5

NH4 controller -0.01 10 000 1 3.5 or 4 1.5 or 1.8

After the implementation of the DO deviation controller the controller oper-ated for a couple of weeks at Henriksdal WWTP. Since then, the plant opera-tors have only occasionally switched on aeration in zone 6 since this has a negative effect on denitrification when oxygen-rich water is recirculated back to the anoxic zones. Therefore there are no long-term results with DO deviation control from Henriksdal WWTP.

Results and discussion 10.3.2An example of DO deviation control in full-scale operation at Henriksdal WWTP is found in Figure 10.4. The example comes from a period when there is a shift from cold to warmer water temperatures and ammonium re-moval becomes easier in periods. The uncertainty in removal meant the op-erators kept the DO set-point at a high level in the three treatment lines oper-ating with constant DO control. Over a weekend the effluent ammonium concentration was decreased, which resulted in a decreased DO set-point in the treatment line with ammonium feedback control. The DO peaks in the last aerated zone were reduced due to ammonium control. The DO deviation controller decreased the DO set-point in the first two aerated zones during periods of DO peaks. The set-point was not reduced below the lower control signal limit in the ammonium controller.

10.3 Full-scale evaluation at Henriksdal WWTP 195

Figure 10.4. Example of DO deviation control at Henriksdal WWTP in 2012. Top: Ammonium feedback control with DO deviation control (treatment line 4). Bottom: Constant DO control (treatment line 1). DO set-point in zone 6 is 2 mg/l.

A second example is presented in Figure 10.5. An unintended emptying of one of the digesters into the treatment plant created a large peak load when the sludge passed through the biological treatment process. Prior to this inci-dent all treatment lines achieved complete nitrification and the last aerated zones were not aerated. The DO in zone 6 prior to August 30 in Figure 10.5 is stray oxygen from the adjacent zone.

In treatment line 3 (Figure 10.5, top), aeration was switched on April 29 before the ammonium load from the digester reached the activated sludge process. In treatment line 4 and 5, aeration in zone 6 was switched on April 30 around noon. The effect of this can be seen on the magnitude of the am-monium peak which was nearly twice as high in treatment line 4 as in line 3. During the peak load period the oxygen demand exceeds the aeration capaci-ty which can be seen when comparing the actual DO concentration with the reference DO concentration. The DO concentration was lower than the set-point in both treatment lines.

The DO deviation controller was active in treatment line 3, and since the DO concentration in zone 6 was higher than its set-point most part of the day the DO set-point in zone 4 and 5 was lowered. A combination of switching on aeration in the last aerated zone before the peak load and the DO devia-tion controller minimised the effluent ammonium concentration and also the energy requirement.

27−Apr 28−Apr 29−Apr 30−Apr 01−May 02−May 03−May

0

1

2

3

4

5

DO

, NH

4 (m

g/l)

DO zone 4 DO zone 6 NH4 after settler

27−Apr 28−Apr 29−Apr 30−Apr 01−May 02−May 03−May

0

1

2

3

4

5

DO

, NH

4 (m

g/l)


Constant DO control

NH4 feedback + DO deviation control


Figure 10.5. Example of DO deviation control at Henriksdal WWTP in 2013. Top: NH4 feedback control with DO deviation control (treatment line 3). Bottom: NH4 feedback control (treatment line 4). DO set-point in zone 6 when aeration is switched on is 1 mg/l. Note: different scales on the y-axes.

The DO deviation controller has been switched on from time to time when the last zone was aerated at Henriksdal WWTP. Often the second aerated zone requires more air than what can be provided by the aeration system. When the DO deviation controller decreased the DO set-point in the two first aerobic zones, it was easier for the DO controller to keep the set-point in the second aerated zone. This created a more smooth operation.

The examples in Figure 10.4 and Figure 10.5 are useful to understand the behaviour of the DO deviation controller and the full-scale experiments has shown the feasibility of using this method in a real plant. However the data is not useful to quantify the effect of DO deviation control on the energy consumption. For this reason, simulations with DO deviation are performed in Section 10.4.

In Chapter 2 it was stated that the most common control problem is that of disturbance rejection rather than set-point tracking. The DO deviation controller is an exception to this as Figure 10.5 clearly shows.

28−Aug 28−Aug 29−Aug 29−Aug 30−Aug 30−Aug 31−Aug 31−Aug 01−Sep

0

1

2

3

4

5

6

DO

, NH

4 (m

g/l)

DO set−point DO zone 4 DO zone 5 DO zone 6 NH4 after settler

28−Aug 28−Aug 29−Aug 29−Aug 30−Aug 30−Aug 31−Aug 31−Aug 01−Sep

0

2

4

6

8

10

DO

, NH

4 (m

g/l)

DO set−point DO zone 4 DO zone 5 DO zone 6 NH4 after settler

NH4 feedback

NH4 feedback + DO deviation control

10.4 Simulations with DO deviation control 197

Simulations with DO deviation control 10.4 Method 10.4.1

The control structure was the same as in Figure 10.1 and the settings are found in Table 10.2. Simulations were made for periods during spring 2011 and spring 2012. The Henriksdal simulation model was used (see Appendix A). The calculation of controller settings was the same as in Section 8.10.

Table 10.2. Controller settings in the NH4 controller and the DO deviation controller at Henriksdal WWTP.

K Ti (d) Set-point (mg/l)

umax (mg/l)

umin (mg/l)

DO deviation controller 0.1 0.01 -0.3 0 -0.5

NH4 controller -0.05 0.1 1 3.5 or 4 1.5 or 1.8

Results and discussion 10.4.2Simulation results from the Henriksdal simulation model with DO deviation control and ammonium feedback control from periods in spring 2011 and spring 2012 are presented in Figure 10.6. A summary of the mean DO con-centrations and energy consumption during the two periods are found in Table 10.3. There was nearly no impact on the ammonium and nitrate con-centrations in the simulations.

Figure 10.6. Simulation results from 2011 (top) and 2012 (bottom) with DO devia-tion control and NH4 feedback control.

18−Mar 20−Mar 22−Mar 24−Mar 26−Mar

0

1

2

3

4

5

DO

, NH

4 (m

g/l)


01−Apr 03−Apr 05−Apr 07−Apr 09−Apr 11−Apr

0

1

2

3

4

5

DO

, NH

4 (m

g/l)


Table 10.3. Mean DO concentrations and effluent NH4 concentration in simulations with NH4 PI control with and without DO deviation control.

NH4 PI control NH4 PI + DO deviation control

DO zone 4 (mg/l)

DO zone 6 (mg/l)

NH4 (mg/l)

DO zone 4 (mg/l)

DO zone 6 (mg/l)

NH4 (mg/l)

Saving Qair (%)

2011 3.16 1.23 1.30 3.06 1.16 1.30 2.2 2012 3.17 1.21 1.32 3.08 1.15 1.32 2.0

The simulations confirm that it is possible to decrease the average DO con-centration in the first two aerated zones by using DO deviation control. The simulations also confirm that the DO peaks are reduced in the last aerobic zone.

The energy saving of around 2 % is modest. One reason to this is that the DO peaks are narrower in the simulations than at the plant (cf. Figure 10.4 and Figure 10.6). The reason could be that nitrification is close to complete for a longer period of the day in the plant than in the plant model. During model calibration it was an issue that the ammonium concentration rarely became zero or close to zero in the model for the Henriksdal and the Käppa-la plant models. The results are dependent on how the model can capture the air flow limits and how often the air flow rate reaches the limits. The rela-tively simple air flow model gives a limited possibility to model this correct-ly, as discussed in Section A.7.

The DO deviation controller was evaluated together with ammonium feedback control at Henriksdal WWTP, but the controller could also be used without ammonium control. An option to using a PI deviation controller is to use an on-off controller. The on-off controller will give a less smooth behav-iour than the PI deviation controller.

Conclusions 10.5A DO deviation controller was implemented and tested in full-scale evalua-tions at Henriksdal WWTP. The DO set-point was decreased in the first two aerated zones when a DO peak appeared in the last aerated zone, since a DO peak is a sign that ammonium is completely nitrified and the aeration inten-sity can be decreased. At Henriksdal WWTP, the DO deviation controller could decrease the DO concentration in periods of DO peaks, and the controller functioned well together with the ammonium PI controller. Since the plant seldom aerates the last aerated zone, long-term evaluation was not possible. The same con-troller was implemented in the Henriksdal simulation model. The energy reduction due to DO deviation control was around 2 %, with no impact on the effluent nitrogen concentration.

199

11 REPETITIVE CONTROL 11

HE TWO CONTROL strategies investigated in Chapter 9 and 10 were focused on decreasing the DO set-point at times when high DO concentrations could be avoided. This chapter has a different pur-

pose: improve disturbance rejection and decrease the ammonium concentra-tion in the effluent. The method used to achieve this is Repetitive Control (RC). RC is a technique used for repetitive processes. By updating the con-trol signal based on previous control errors the RC controller learns the be-haviour of a repetitive disturbance. RC can be used to react to a disturbance before it is noticed by a feedback. Simulations are performed in the BSM1_LT model.

Introduction 11.1A challenge for process control in wastewater treatment plants is that the incoming wastewater normally cannot be controlled but is to be considered a disturbance to the system. The load variations are often large with an inher-ently diurnal pattern. RC is a method used for continuous repetitive process-es. The idea behind RC is to update the control signal based on the control error in previous periods to achieve rejection of a periodic disturbance or tracking of a periodic reference. RC was originally developed for control of power supply (Inoue et al., 1981a, 1981b). The RC design is based on the internal model principle (Francis and Wonham, 1976).

RC in wastewater treatment can be motivated if additional disturbance re-jection is required to either meet an average effluent criteria or a never-to-exceed limit. RC can provide a feedforward-like action without requiring additional sensors to measure the disturbance. At a treatment plant, influent ammonium is known to be difficult to measure given the hostile environment for the sensor.

T

200 11 Repetitive control

A close relative to RC is Iterative Learning Control (ILC). ILC was de-veloped in the 1980’s within robotics research and is traditionally applied to finite time dimensional systems where the system returns to the same initial condition before each repetition. Initial condition reset is a critical assump-tion for the theoretical analyses of ILC but it has been shown that under the same conditions, ILC and RC are mathematically the same (Ahn et al., 2007). With the states never returning to the same conditions for each peri-od, the stability analysis for RC and ILC is different. Unlike for ILC, stabil-ity for RC will depend on the system dynamics (Longman, 2000). However, for engineering practice there is no real difference in terms of design and tuning since satisfying the true stability criteria for RC and ILC is not essen-tial for practical implementation (Longman, 2000).

ILC has earlier been applied to a Sequencing Batch Reactor (SBR) for wastewater treatment in Kim et al. (2009) and more recently in Nygren et al. (2014). Kim et al. (2009) use state-space ILC to control the DO concentra-tion and the external carbon dosage. Compared to a trial-and-error tuned DO PID controller the ILC controller was better at tracking the set-point. The state-space model was a reduced-order version of an ASM model. Nygren et al. (2014) used various types of approximate models for designing ILC con-trollers and found that an adjoint-based ILC gave the best performance. The results are also supported by a theoretical analysis.

Traditionally, RC was developed in the frequency domain and ILC in the time domain. This chapter will show the notation in the time domain to be consistent with the other controllers presented in the thesis.

Control structure and control strategies 11.2When applying repetitive control, the controller is supposed to track a refer-ence or reject a periodic disturbance. A simple RC algorithm has few design parameters which makes the control design especially easy. A simple exam-ple of an RC algorithm is

( ))()()( δγ +−+−= pteptuQtuRC (11.1)

where uRC(t) is the control signal from the repetitive controller, Q is a filter, p is the period length, γ is the learning gain, e(t) the control error and δ is the chosen number of time steps that the previous control error should be shift-ed. If δ > 1, a linear phase lead RC is created.

The Q-filter is more common in an ILC context and can be made to adapt the updating law in a variety of ways. The filter can be used as a forgetting factor (Q = [0 1]) or as a weighted-average window (Wang et al., 2009). The bandwidth of the Q-filter can improve the robustness of the learning algo-rithm and also provide a less aggressive controller. In this study the Q-filter

11.2 Control structure and control strategies 201

was chosen as a first-order low pass filter. Often there is a learning filter, an L-filter, in a learning algorithm such as Eq. (11.1). In this study the L-filter is the simplest possible, i.e. a constant γ. The choice of γ is a trade-off be-tween (monotonic) stability and robustness to noise on one hand, and con-vergence speed on the other.

The RC updates a signal from one period to the other, and works as a feedback controller with respect to the periods since the control error for the previous period is added to the output signal. At the same time, the algorithm can be “non-causal” since it can use knowledge from later time steps in pre-vious periods, as with δ > 1 in (11.1).

Higher-order algorithms, where information from more than one period back can be incorporated in the learning, has been used to handle sensitivity to non-periodic inputs and uncertainty in the period-time (Chang et al., 1988; Steinbuch, 2002).

To improve the control with respect to non-periodic disturbances, a learn-ing law is in many applications combined in series or parallel to an existing feedback controller as an add-on controller (Li et al., 2004). When a feed-back controller is already working on the system, RC manipulating the command given to the feedback controller is preferred (Longman, 2000). An overview of such a system with a feedback PI controller and an add-on RC controller is found in Figure 11.1. The alternative would be for the RC to update the manipulated variable directly. The control structure in Figure 11.1 is implemented for cascade control of ammonium in the BSM1_LT model according to Figure 11.2.

Figure 11.1. RC in series with a feedback PI controller.

Figure 11.2. RC in series with a PI ammonium controller. The RC changes the am-monium control error based on the control error of previous periods.


A picture of the RC implementation in MATLAB/Simulink® is given in Figure 11.3. There is a saturation of the RC signal in Figure 11.3, where uRC > 0. This was to prevent the RC from decreasing the DO set-point during the period of low load.

Figure 11.3. Overview of the implementation of RC in MATLAB/Simulink. Start/stop refers to criteria defining when learning should be switch on and off. τ is the low pass filter time constant.

Several aspects of the simple RC algorithm are investigated by simulations with the BSM1_LT model, previously described in Section 6.2. RC was switched on September 10th and learning was active during 20 days. After the 20 days of learning the output signal from the RC (uRC) was the same as on day 20. The controller operated with the same uRC for 40 days more.

Table 11.1 summarises the simulations performed in BSM1_LT. The simulations looked at different choices of (1) shift of control error, δ, (2) learning gain, γ, and (3) time constant in the low-pass filter Q, τ, in Eq. (11.1). p was 24 hours. The ammonium PI controller was the same in all simulations. The ammonium set-point was 1 mg/l and the upper and lower DO set-point limit was 1 and 2.5 mg/l, respectively.

The purpose of the simulations was to provide improved disturbance re-jection through reacting early to the load peak and thereby reducing the max-imum ammonium concentration and lowering the effluent ammonium and total nitrogen concentrations. To have perfect tracking of the set-point is rarely possible in a WWTP; hence the learning trajectory was of less im-portance.

Table 11.1. Settings performed in the simulations with RC in BSM1_LT. The change compared to the other simulations in bold.

Simulation δ (h) γ τ (d) K Ti (d)

RC 1 0.25 0.8 0.3 -0.3 0.2RC 2 4 0.8 0.3 -0.3 0.2RC 3 6 0.8 0.3 -0.3 0.2

RC 4 6 0.5 0.3 -0.3 0.2 RC 5 6 1 0.3 -0.3 0.2RC 6 6 0.8 0.2 -0.3 0.2RC 7 6 0.8 0.4 -0.3 0.2

11.3 Simulations with repetitive control 203

Simulations with repetitive control 11.3 The choice of δ 11.3.1

The optimal DO profile in Chapter 5 demonstrated benefits with increasing the DO concentration prior to the load peak arrives. This is one of the possi-ble benefits with using RC in a repetitive process; by learning when the peak arrives the DO concentration can be increased before the feedback controller reacts. This is demonstrated in Figure 11.4. By increasing δ the PI + RC controller increases the DO concentration earlier.

Figure 11.4. DO concentration after 19 days of learning with RC 1 (δ = 0.25 h), RC 2 (δ = 4 h) and RC 3 (δ = 6 h). Learning is switched off on September 30. Settings according to Table 11.1.

The choice of γ 11.3.2The learning gain, γ, is commonly chosen between 0.5 and 1 and is used to tune how fast learning will occur. Decreasing the learning gain has a smoothing effect on the DO set-point (Figure 11.5). If the learning period was extended, the stationary response of the three settings would be the same.

Figure 11.5. DO concentration after 19 days of learning with RC 1 (γ = 1), RC 3 (γ = 0.8) and RC 5 (γ = 0.5). Learning is switched off on September 30. Settings accord-ing to Table 11.1.

29−Sep 30−Sep 01−Oct 02−Oct 03−Oct

0.5

1

1.5

2

2.5

3

DO

(m

g/l)

δ = 0.25 h δ = 4 h δ = 6 h


0.5

1

1.5

2

2.5

3

DO

(m

g/l)

γ = 1 γ = 0.8 γ = 0.5


The low pass filter 11.3.3The low pass filter in Eq. (11.1) can – similar to the learning gain – be used to have a smoother output from the RC controller. Without a low pass filter the DO concentration will be decreased more sharply which increases the effluent ammonium concentration.

Figure 11.6. DO concentration after 19 days of learning with RC 6 (τ = 0.2), RC 3 (τ = 0.3) and RC 7 (τ = 0.4). Learning is switched off on September 30. Settings ac-cording to Table 11.1.

Summary of simulations 11.3.4The PI + RC 2 ammonium controller is compared to a PI ammonium con-troller in Figure 11.7. Learning is switched off on September 30. The control error and output from the RC controller are plotted in Figure 11.8.

Figure 11.7. Comparison between PI + RC control and pure PI control after 20 days of learning and for the rest of the evaluation period.


0.5

1

1.5

2

2.5

3

DO

(m

g/l)

τ = 0.2 τ = 0.3 τ = 0.4

30−Sep 07−Oct 14−Oct 21−Oct 28−Oct 04−Nov

0

2

4

6

8

10

NH

4 (m

g/l)

Ammonium PI control Ammonium PI+RC control

30−Sep 07−Oct 14−Oct 21−Oct 28−Oct 04−Nov

0.5

1

1.5

2

2.5

3

DO

(m

g/l)

11.3 Simulations with repetitive control 205

Figure 11.8. Output from RC controller (uRC), control error from NH4 PI controller (e) and DO set-point. The NH4 control error entering the NH4 PI controller is the sum of e and uRC. The RC controller only acts on the first NH4 peak while the NH4 PI controller takes care of the second and smaller NH4 peak.

There is a visible decrease in the ammonium peaks with the PI + RC control-ler compared to only using PI ammonium control (Figure 11.7). This reduc-tion in ammonium peak is achieved through increasing the DO set-point prior to when the maximum daily load arrives, which is also visible in Figure 11.7. The effluent ammonium concentration does not have a completely regular pattern. When needed the feedback controller compensates for this, for instance in early November. There are examples of when the RC part of the PI + RC controller contributes to unnecessary aeration during days when the ammonium peak is very low (e.g. October 14), but the RC controller – being taught to do so – still increases the DO set-point prior to the daily peak load event.

A summary of the treatment results in the simulated RC + PI controllers is given in Table 11.2. The energy consumption is increased with RC control, which is expected since its purpose is to reduce the effluent ammonium con-centration. There is a significant reduction in the ammonium peaks with RC + PI control, and there is also an effect on the average effluent ammoni-um concentration. The average total nitrogen concentration is only slightly lower, since a reduction in ammonium concentration adds more nitrate to the effluent.

30−Sep 03−Oct 06−Oct 09−Oct

−7

−6

−5

−4

−3

−2

−1

0

1

2

3

mg/

l

u

RC Control error, e DO set−point


Table 11.2. Summary of simulation results with RC + PI control.

Simulation TN (mg/l) NH4 (mg/l) max NH4 (mg/l) Qair compared to PI control (%)

PI no RC 8.64 1.29 7.64 RC 1 8.55 1.02 5.07 3.9 RC 2 8.48 0.97 4.74 2.9 RC 3 8.58 0.92 4.65 6.7

RC 4 8.58 0.92 4.64 6.4 RC 5 8.56 0.94 4.66 5.2 RC 6 8.58 0.92 4.64 4.9 RC 7 8.56 0.94 4.67 6.5

If an RC controller should be applied in a full-scale process there are several aspects which need to be addressed. First of all, learning must be switched on during a representative and repetitive period of operation. If the dynamics of the system varies much over the year, which is the case for the BSM1_LT model as well as the full-scale plants studied in this thesis, the same RC out-put signal should not be used all year around but would need to be updated regularly. Non-periodic ammonium peaks, for instance caused by rains, will be handled by the PI controller. Learning should be performed with the am-monium sensor placed in situ, rather than after the secondary settler (see Section 8.8).

Noise is not handled in this study since the experience from the full-scale trials in Chapter 8 showed that the ammonium sensors were most of the time not noisy. If noise is an issue for the ammonium sensor, appropriate filtering of the control error is required to achieve smooth learning.

There are few controller parameters to tune in order to design a RC con-troller on the simple form in Eq. (11.1). However, given the uncertainty of the expected plant performance in full-scale, it is useful to simulate how the controller parameters affect the process.

Conclusions 11.4A simple repetitive controller was combined with ammonium PI control and investigated with simulations in the BSM1_LT model. The controller showed a potential to significantly decrease the effluent ammonium peaks and the average effluent ammonium concentration through reacting proac-tively to the daily peak load, thereby creating a feedforward-like effect with-out needing a sensor to measure the disturbance. The cost is increased due to the RC controller.

PART V

CONCLUSIONS

209

12 CONCLUDING REMARKS 12

Answering the questions – a summary of the results 12.1What is aeration control and how does it improve the operation of a WWTP? The general introduction to wastewater treatment and aeration control in Part I emphasised the importance of aeration to the treatment results, but also the cost of aeration which motivates aeration control. Control of aeration at a WWTP includes for example control of the blower system or system pres-sure in the air mains, but the focus of this thesis has been to use ammonium measurements to calculate an external DO set-point to the DO controllers.

The main benefit of aeration control is the potential to save energy through avoiding unnecessary aeration, but it can also improve the overall nitrogen removal of a plant through creating a balance between the nitrifica-tion and denitrification if volume control is considered (Chapter 3). Looking at ammonium feedback control, the energy saving of published research is in the range of 5 to 25 %, partly depending on what control strategy was used for comparison. Given the last ten years of published research, ammonium feedback control can today be considered state-of-the-art in full-scale plants with many examples of successful installations. In a majority of the pub-lished full-scale studies the evaluation time was shorter than 2 months.

Optimal or model-based controllers, in this thesis categorised as advanced control, have yielded much attention the last 15 years in connection with aeration control. So far, a large majority of the studies are simulation studies (Chapter 3). There are a few examples of full-scale DO control with ad-vanced control. Ammonium-based advanced controllers in continuous aera-tion systems are to this date not verified in published studies evaluating long-term operation in full-scale. The complexity of the instrumentation and the control system always has to be considered in relation to the choice of control structure and control strategy.

210 12 Concluding remarks

How should the settings in an ammonium PI controller be chosen? The three chapters in Part II were all devoted to cost-effective design of ammonium PI controllers. A basis for all the chapters was that compliance to effluent criteria is assessed based on long-term averages. Ammonium PI control was compared to constant DO control.

The energy saving with an optimal DO profile in a simplified model of an activated sludge process was 3 to 7 % compared to constant DO control (Chapter 5). The shape of the optimal DO profile tells us that it is beneficial to react early to a load disturbance, and to have an in average lower DO concentration in the first aerated zone compared to the last zone. The best performing ammonium PI controllers could reach close to optimal performance, with an energy saving of approximately 1 to 3.5 % compared to constant DO control. It is important to limit the maximum DO concentration in the ammonium controller.

In Chapter 6, special attention was given to the choice of controller gain, integral time, ammonium set-point and upper DO set-point limit. Simula-tions were performed for a full year in the BSM1_LT model. If ammonium control was compared to constant DO control with matching effluent ammo-nium concentrations, energy saving was in the range of 3 to 7 %. The saving was approximately the same even if the load in relation to the treatment vol-ume was changed. When it is desirable to reach low effluent ammonium concentrations, there is more energy to save with ammonium PI control compared to constant DO control.

In the models simulating real plant performance in Chapter 7, the main difference in energy saving between the years depended on how the refer-ence strategy was operated and on variations in influent load due to precipi-tation. The selection of ammonium PI controller settings becomes more im-portant when there are frequent effluent ammonium peaks. Placing the am-monium sensor after the secondary settler only has a minor influence on the effluent ammonium concentration, but it does reduce the energy saving by up to 1 % in the simulations in this thesis.

Chapters 5, 6 and 7 all suggest that the best performing ammonium PI controller should have a long integral time. In Chapter 5 and 6, the energy saving was larger if the controller could be allowed to generate a large variation in DO concentration with the help of a larger controller gain (|K|). This could be allowed at higher DO concentrations where the absolute value of the process gain was lower.

The simulations in Chapter 6 and 7 suggest the energy consumption is rather insensitive to the choice of controller gain or integral time, as long as the controller is not too slow or too fast. Many controllers falls within a distance of 1 to 2 % of the best performing controller if looking at the energy saving. Special care should be taken when the upper DO set-point limit is high to avoid a too short integral time which could become costly.

12.1 Answering the questions – a summary of the results 211

Implementation and evaluation of ammonium PI control in a full-scale WWTP – what are the challenges and benefits? The full-scale evaluations of ammonium PI control at Henriksdal, Käppala and Himmerfjärden WWTPs confirmed the conclusions from Part I and Part II: ammonium PI control can achieve an energy saving without compromis-ing the effluent average ammonium concentration.

The estimated energy saving was in the range of 7 to 19 % and the cost-benefit analyses suggested it was beneficial to implement ammonium feed-back control at all plants. If comparisons were made to constant DO control the energy saving was up to approximately 10 %. The ammonium PI control-lers earned the trust and acceptance of the plant personnel.

Systematic differences between treatment lines or between subsequent evaluation periods are a threat to a fair comparison between full-scale pro-cess controllers. This made quantifications of the energy consumption and energy saving particularly difficult, and it can be concluded that these calcu-lations will always remain an estimate.

PI controllers are the far most common controller in process industries. One unexpected delay during the implementation phase of the ammonium PI controllers was controller implementation in the control systems. Industrial control systems offer several advantages compared to stand-alone controllers in a SCADA system. One crucial advantage with an industrial control sys-tem is avoiding master controller wind-up in cascade control.

Drift, shift and an exaggeration of ammonium peaks were the main sensor errors experienced during the evaluation period. The problems were more pronounced at the plant where the sensor was placed in the aeration tank. Placing the sensor after the settler allows for less maintenance and improved sensor data quality. Ammonium control with the ammonium sensor placed after the settler proved to be a feasible alternative at two of the plants. Plac-ing the sensor after the settler generates a time delay, and if the influent vari-ations are large the lowest DO concentration could occur near peak load when the ammonium controller is fast. Slower control adapting to weekly and monthly variations in the effluent ammonium concentration is preferred in these cases.

Can ammonium PI control be improved through simple modifications? Through small adjustments to the standard ammonium PI controller or am-monium cascade control structure, three controllers with the task to improve the performance of ammonium PI control were evaluated in Part IV.

Gain scheduling control was investigated at Käppala WWTP in full-scale operation and in a simulation study. Implementation in the plant control sys-tem was relatively straight-forward but its applicability was hampered by restrictions in the access rights of the operators in the control system. With


gain scheduling, the aim to increase the aeration intensity at rain events but avoid high DO concentrations during normal operation could be met. The most important parameter to schedule was the upper DO set-point limit. Ac-cording to simulations, gain scheduling ammonium PI control could de-crease energy consumption during ammonium peaks with up to 4 %. Given the few ammonium peaks at Käppala WWTP, the annual saving was around 1 %.

A DO deviation controller was implemented and tested in full-scale eval-uations at Henriksdal WWTP. The DO set-point was decreased in the first two aerated zones when a DO peak appeared in the last aerated zone. Im-plementation in the control system was possible with an additional PI con-troller, tuning was straight-forward and the deviation controller has been operated from time to time at the plant over the last two years and is appreci-ated by the plant personnel. Simulations showed an energy saving potential of around 2 % during the periods when the deviation controller was active.

The final attempt to improve the ammonium feedback control had the purpose to improve disturbance rejection and decrease the ammonium con-centration rather than to save energy. A simple repetitive controller was combined with feedback PI ammonium control and investigated with simula-tions in the BSM1_LT model. The repetitive controller could decrease the overall total nitrogen concentration and decrease the ammonium peak in the effluent. The main benefit of the repetitive controller was that it increased the DO concentration prior to the morning ammonium peak.

Overall conclusions 12.2This thesis has confirmed the usefulness of ammonium PI control, especially to achieve energy savings at WWTPs. The thesis has also shown the versatil-ity of ammonium PI control. The strategy can be used with a large range of PI controller settings, the ammonium sensor can be placed in situ or after the secondary settler and the ammonium controller can be improved through simple modifications. A major contribution of the thesis is the results and discussions resulting from the long-term studies at three large WWTPs.

The design of an ammonium PI controller does not have to be complicat-ed. It is possible to achieve satisfactory performance with manual tuning together with knowledge about the process. Process knowledge includes process delays, load and load variations, nitrification rate at different DO concentrations and knowledge about the equipment (sensors, actuators). If an operator normally changes the DO set-point in a plant with DO control, the task with ammonium feedback control instead becomes to occasionally change the ammonium set-point or the DO set-point limits. There are many controllers which perform close to the maximum achievable energy saving compared to constant DO control, as long as too slow and too fast controllers

12.2 Overall conclusions 213

are avoided. To achieve high energy efficiency the integral time should be longer than a typical tuning rule like lambda tuning suggests.

If the above advice is observed, there is a large group of ammonium PI controller settings which offer similar performance both in terms of energy consumption and effluent ammonium concentration. Lower ammonium set-points or higher DO limits will decrease the effluent ammonium concentra-tion and increase the energy requirement, but also increase the potential en-ergy saving compared to constant DO control, see Figure 12.1. It is more meaningful to implement ammonium control if the plant should operate with a low effluent ammonium concentration relative to the plant capacity.

Figure 12.1. Increased DO concentrations caused by a decreased NH4 set-point or increased DO set-point limits will increase energy consumption but also increase the energy saving compared to constant DO control designed to give the same effluent ammonium concentration as the ammonium controller.

Implementation of ammonium PI control is straight forward in most control systems, but implementation was delayed in this project. Limits to all output signals and master controller windup are important aspects to consider. Re-member that ammonium removal is slow and the integral time might have to be chosen larger than what is first allowed by the control system.

It is generally assumed that the ammonium sensor should be placed in situ to provide a fast response to changes in the process. This thesis has shown that if an in situ sensor requires much maintenance or has difficulties to measure correctly in the aeration tank, it is possible to achieve good perfor-mance with the sensor placed after the secondary settler.

The ammonium sensors have been performing sufficiently well to be able to achieve good control with an ammonium PI controller over time. At two of the plants the sensors showed higher concentrations compared to lab sam-ples and the peaks were often exaggerated. This can be costly but it does not risk the treatment results. A priority within this research project was to focus much attention on the maintenance and quality of the ammonium sensors. In retrospect, more attention should have been devoted to the air flow rate sen-sors since they proved to be faulty.


The evaluation of control strategies in a WWTP should be performed over long time periods to be able to achieve representative results. Normalisation of energy consumption with plant performance data such as effluent concen-trations is risky since it can impose a correlation on data which conceal the results. The risk of normalisations arises in simulation studies as well as in pilot plant or full-scale studies.

It is common that parallel treatment lines are dissimilar due to age of equipment, flow equalisation or different microbial cultures. Likewise, two subsequent periods of evaluation often differ due to changes in the influent or equipment. Ammonium PI controllers can be improved through simple modifications of the cascade control structure or the controller. The additional improvements of the methods tested in this thesis (gain scheduling control, DO deviation control and repetitive control) are small compared to the improvements cre-ated by the ammonium controller itself. The ammonium feedback control structure is the key to the larger share of the energy saving or disturbance rejection. Figure 12.2 provides a summary of energy saving potential with ammonium PI control. The pyramid is constructed to show the contributions of energy saving from different phenomenon. The blue areas represent comparisons to constant DO control where the effluent ammonium concentration is the same. The black area represents a possible energy saving from when the effluent ammonium concentration is allowed to be higher when ammonium control is used, or when ammonium PI control is compared to a control strategy which is more energy demanding than constant DO control.

Figure 12.2. Energy saving from ammonium PI control can reach up to 10 % com-pared to constant DO control. If more energy is to save, the effluent ammonium concentrations needs to be compromised, or comparison has to be made to another control strategy than constant DO control.

1 to 2 %

Up to 3.5 %

Up to 5 %

Up to 15 %

Improvements toammonium PI control(Chapter 9, 10)

Daily variations in DO concentration(Chapter 5, 8)

Annual and weekly variations in DO concentration(Chapter 6, 7, 8)

Non-identical effluent NH4 conc. or saving compared to other controlstrategy than constant DO control(Chapter 6, 7, 8)

12.3 Future work 215

Future work 12.3If this thesis was a play, the PI controller would have had the principal part. This thesis has explored the possibilities with ammonium PI control, but it would be ignorant to conclude that a PI feedback controller is the preferred solution to every aeration control problem. The literature review in Chapter 3 gave several examples of where feedforward control and volume control could improve the performance of the activated sludge process. They are two methods which have a potential to improve the standard PI ammonium con-troller and would have fitted nicely into Part IV of the thesis. None of the ammonium sensors situated at the influents at Henriksdal, Käppala and Himmerfjärden WWTPs were considered trustworthy enough for feedfor-ward control. Combined with the definition of the effluent criteria in Sweden (long averages), feedforward control was not considered for the plants – hence the attraction towards repetitive control. Volume control is also an attractive method but since neither of the plants had walls between the aero-bic zones, volume controller was not evaluated in this thesis.

The repetitive controller in Chapter 11 is only a scratch on the surface given the very active research field of learning control. Chapter 11 provides a hint of what can be achieved, but there is a wide range of more advanced RC algorithms that might be interesting to consider in the future.

Advanced controllers are rather brutally dismissed in the literature review in Chapter 3 since ammonium PI control have so far outperformed advanced controllers in pilot plant studies and also in some contexts in simulations. However, in general there is a lack of full-scale studies where advanced ammonium-based controllers are evaluated in full-scale operation. That does not mean that there is no need for model-based control at a WWTP. It does merely mean that the advanced controllers should be put to work only where a cost-benefit analysis suggest it is worthwhile to implement them. Ammo-nium control can provide an external DO set-point to a DO controller – but how should the ammonium set-point be determined, especially in combina-tion with other set-points in parallel control loops such as nitrate control? How can the ammonium set-point and the DO set-point limits be used to guarantee a certain removal rate of ammonium? Many simulations were performed in this thesis where the settings in the controller were kept con-stant over long periods of time. Could for example adaptive controllers im-prove the performance of ammonium PI control through regularly adapting the controller settings to the actual situation?

In this work, the ammonium controller has generated one single DO set-point used in several aerated zones. The optimisation in Chapter 5 suggested a lower DO set-point should be used in the first aerated zone. An increasing DO profile required less energy in Sahlmann et al. (2004). An interesting topic for future research would be to look at the DO profile to optimise oxy-


gen transfer and ammonium removal and combine this with ammonium feedback control.

The results from the simulations are subject to assumptions and settings made in the calibration and model set-up. The simulations do not include measurement noise or sensor failures which might have impacted the results. Sensor performance is a key to well-functioning process controllers. Meth-ods to work with sensor fault detection could be a substantial contribution to further improving the performance of controllers in the activated sludge pro-cess, irrespective of control strategy.

In the future, more work could be devoted to look at the impact of the model parameters on the benefits from using ammonium feedback control. A central parameter is the oxygen half-saturation coefficient of autotrophs which does have a large impact since it alters the response to a change in DO concentration. With the overall growth rate at the same level, a change in the half-saturation constant would impact the process gain of the ammonium response, which in turn has an impact on the choice of controller settings as discussed in Chapter 7.

All control loops in an aeration control system have not gained attention in this thesis. In many parts of this work, the control loops managing the blowers and system pressure in the air mains have been taken for granted with very little comments on how the performance of these control loops affect the operation of the ammonium control loop. In the full-scale plants in this study it is probable that energy savings could be achieved by looking into the performance of for instance the MOV controllers, see Vrečko et al. (2013).

A topic for future research is to develop more realistic models of the aera-tion systems, sensor noise and sensor dynamics since it would create more challenges to solve with process control. Both in full-scale and in simula-tions the issue of fall-back strategies – required when an actuator or sensor is not working as it should – is vital and could be a topic for future research.

217

BIBLIOGRAPHY

ABB Industrial Systems, 1998. User’s Guide PIDCONA. 3BSE 002 433R0101.

Adam, C., Peplinski, B., Michaelis, M., Kley, G., and Simon, F.-G., 2009. Thermochemical treatment of sewage sludge ashes for phosphorus recovery. Waste management 29(3), 1122–8.

Ahn, H.-S., Chen, Y., and Moore, K.L., 2007. Iterative Learning Control: Brief Survey and Categorization. IEEE Transactions on Systems, Man and Cybernetics, Part C (Applications and Reviews) 37(6), 1099–1121.

Alex, J., To Binh, T., and Hartwig, P., 2002. Improved design and optimization of aeration control for WWTPs by dynamic simulation. Water Science and Technology 45(4-5), 365–72.

Andersson, L.G., 1979. Energy savings at wastewater treatment plants. Final report contract No. 314-77-EE-DK, Water Quality Institute (DHI Water and Environment), Hørsholm, DK.

ATV, 2000. Standard ATV-DVWK-A 131E: Dimensioning of Single-Stage Activated Sludge Plants. Publishing Company of ATV-DVWK, Water, Wastewater, Waset, Hennef 2000.

Ayesa, E., De La Sota, A., Grau, P., Sagarna, J.M., Salterain, A., and Suescun, J., 2006. Supervisory control strategies for the new WWTP of Galindo-Bilbao: the long run from the conceptual design to the full-scale experimental validation. Water Science and Technology 53(4-5), 193–201.

Baeza, J.A., Gabriel, D., and Lafuente, J., 2002. In-line fast OUR (oxygen uptake rate) measurements for monitoring and control of WWTP. Water Science and Technology 45(4-5), 19–28.

Balku, S., and Berber, R., 2006. Dynamics of an activated sludge process with nitrification and denitrification: Start-up simulation and optimization using evolutionary algorithm. Computers & Chemical Engineering 30(3), 490–499.

Baroni, P., Bertanza, G., Collivignarelli, C., and Zambarda, V., 2006. Process improvement and energy saving in a full scale wastewater treatment plant: air supply regulation by a fuzzy logic system. Environmental technology 27(7), 733–46.

Beltrán, S., Irizar, I., De la Sota, A., Villanueva, J.M., and Ayesa, E., 2013. Model-based optimization of aeration systems in WWTPs, in: 11th IWA Conference on Instrumentation, Control and Automation. Narbonne, France, 18-20 September, 2013.

218 Bibliography

Beraud, B., Lemoine, C., and Steyer, J-P., 2009. Multiobjective genetic algorithms for the optimisation of wastewater treatment processes, in: do CarmoNicoletti, M., & Jain, L.C. (Eds.), Computational Intelligence Techniques for Bioprocess Modelling, Supervision and Control. Springer Verlag Berlin Heidelberg, pp. 163–195.

Brdys, M.A., and Diaz-Maiquez, J., 2002. Application of fuzzy model predictive control to the dissolved oxygen concentration tracking in an activated sludge process, in: 15th IFAC World Congress. Barcelona, Spain, 21-26 July 2002.

Brdys, M.A., Grochowski, M., Gminski, T., Konarczak, K., and Drewa, M., 2008. Hierarchical predictive control of integrated wastewater treatment systems. Control Engineering Practice 16(6), 751–767.

Brdys, M.A., and Konarzcak, K., 2001. Dissolved oxygen control for activated sludge processes, in: Proc. of the 9th IFAC/IFORS/IMACS/IFIP Symposium on Large Scale Systems: Theory & Applications. Bucharest, Romania, 18-20 July 2001.

Brouwer, H., Bloemen, M., Klapwijk, B., and Spanjers, Henri, 1998. Feedforward control of nitrification by manipulating the aerobic volume in activated sludge plants. Water Science and Technology 38(3), 245–254.

Chang, W., Such, I., and Kim, T., 1988. Analysis and design of two types of digital repetitive control systems. Automatica 31(5), 741–746.

Chotkowski, W., Brdys, M.A., and Konarczak, K., 2005. Dissolved oxygen control for activated sludge processes. International Journal of Systems Science 36(12), 727–736.

Choubert, J.-M., Stricker, A.-E., Marquot, A., Racault, Y., Gillot, Sylvie, and Héduit, A., 2009. Updated Activated Sludge Model n°1 Parameter Values for Improved Prediction of Nitrogen Removal in Activated Sludge Processes: Validation at 13 Full-scale Plants. Water Environment Research 81(9), 858–865.

Coker, E.G., 1983. The use of sewage sludge in agriculture. Water Science and Technology 15, 195–208.

Copp, J.B., 2002. The COST Simulation Benchmark: Description and Simulator Manual, Simulation Benchmark: Description. ISBN 92-894-1658-0, Office for Official Publications of the European Community, Luxembourg, 2002.

Cordell, D., Drangert, J.-O., and White, S., 2009. The story of phosphorus: Global food security and food for thought. Global Environmental Change 19(2), 292–305.

Corominas, L.L., Rieger, L., Takács, I., Ekama, G., Hauduc, H., Vanrolleghem, P.A., Oehmen, A., Gernaey, K.V., van Loosdrecht, M.C.M., and Comeau, Y., 2010. New framework for standardized notation in wastewater treatment modelling. Water Science and Technology 61(4), 841–57.

Bibliography 219

De Araujo, A.C.B., Gallani, S., Mulas, M., and Olsson, G., 2011. A systematic approach to the design of operation and control policies in activated sludge systems. Industrial & Engineering Chemistry Research 50, 8542–8557.

Dold, P., and Fairlamb, M., 2001. Estimating oxygen transfer KLa, SOTE and air flow requirements in fine bubble diffused air systems, in: WEFTEC.01. Atlanta, USA, 13–17 October, 2001, pp. 780–791.

Doyle, J.D., and Parsons, S. a, 2002. Struvite formation, control and recovery. Water research 36(16), 3925–40.

Ekman, M., 2008. Bilinear black-box identification and MPC of the activated sludge process. Journal of Process Control 18(7-8), 643–653.

Ekman, M., Björlenius, B., and Andersson, M., 2006. Control of the aeration volume in an activated sludge process using supervisory control strategies. Water research 40(8), 1668–76.

Fikar, M., Chachuat, B., and Latifi, M., 2005. Optimal operation of alternating activated sludge processes. Control engineering practice 13(7), 853-861.

Fiter, M., Güell, D., Comas, J., Colprim, J., Poch, M., and Rodríguez-Roda, I., 2005. Energy saving in a wastewater treatment process: an application of fuzzy logic control. Environmental technology 26(11), 1263–70.

Francis, B., and Wonham, W., 1976. The internal model principle of control theory. Automatica 12(5), 457–465.

Fujie, K., Sekizawa, T., and Kubota, H., 1983. Liquid mixing in activated sludge aeration tank. Journal of Fermentation Technology 61(3), 295–304.

Galarza, A., Ayesa, E., Linaza, M.T., Rivas, A., and Salterain, A., 2001. Application of mathematical tools to improve the design and operation of activated sludge plants. Case study: the new WWTP of Galindo-Bilbao. Part II: Operational strategies and automatic controllers. Water Science and Technology 43(7), 167–74.

Gaya, M.S., Wahab, N. a., Sam, Y.M., and Samsuddin, S.I., 2013. ANFIS inverse control of dissolved oxygen in an activated sludge process. IEEE 9th International Colloquium on Signal Processing and its Applications, 146–150.

Gerkšič, S., Vrečko, D., and Hvala, N., 2006. Improving oxygen concentration control in activated sludge process with estimation of respiration and scheduling control. Water Science and Technology 53(4-5), 283–291.

Gernaey, K.V., Flores-Alsina, X., Rosen, C., Benedetti, L., and Jeppsson, U., 2011. Dynamic influent pollutant disturbance scenario generation using a phenomenological modelling approach. Environmental Modelling & Software 26(11), 1255–1267.

Gernaey, K.V., Jeppsson, U., Vanrolleghem, P.A., and Copp, J.B., 2014. Benchmarking of control strategies for wastewater treatment plants.

220 Bibliography

IWA Scientific and Technical Report, ISBN 9781843391463, IWA Publishing, London, UK.

Global Water Research Coalition, 2011. N2O and CH4 emission from wastewater collection and treatment systems. London, UK, GWRC 2011-30, ISBN 978.90.77622.24.7.

Groves, K., Daigger, G., Simpkin, T., Redmond, D., and Ewing, L., 1992. Evaluation of oxygen transfer efficiency and alpha-factor on a variety of diffused aeration systems. Water environment research 64(5), 691–698.

Guerrero, J., Guisasola, A., Vilanova, R., and Baeza, J.A., 2011. Improving the performance of a WWTP control system by model-based setpoint optimisation. Environmental Modelling & Software 26(4), 492–497.

Halvarsson, B., 2010. Interaction analysis in multivariable control systems - Applications to bioreactors for nitrogen removal. PhD Thesis, Department of Information Technology, Uppsala University, Uppsala, Sweden

Halvarsson, B., Samuelsson, P., and Carlsson, B., 2005. Applications of couplings analysis on bioreactor models, in: Proceedings of the 16th IFAC World Congress. Prague, Czech Republic, 4-8 July, 2005.

Han, H.-G., Qiao, J.-F., and Chen, Q.-L., 2012. Model predictive control of dissolved oxygen concentration based on a self-organizing RBF neural network. Control Engineering Practice 20(4), 465–476.

Han, Y., Brdys, M.A., and Piotrowski, R., 2008. Nonlinear PI control for dissolved oxygen tracking at wastewater treatment plant, in: Proceedings of the 17th World Congress: The International Federation of Automatic Control. Seoul, Korea, July 6-11 2007, pp. 13587–13592.

Harri, A., and Bosander, J., 2004. Six years of operation of a fluidised bed reactor for denitrification. Water Science and Technology 50(7), 97–103.

Henze, M., Grady, C.P.L. Jr., Gujer, W., Marais, G.v.R., and Matsuo, T., 1987. Activated Sludge Model No. 1. IAWPRC Scientific and Technical Report No. 1. London: IAWQ.

Henze, M., Gujer, W., Mino, T., and Van Loosdrecht, M. C. M., 2000. Activated sludge models ASM1, ASM2, ASM2d and ASM3. IWA Publishing, London, UK.

Holenda, B., 2007. Development of modelling, control and optimization tools for the activated sludge process. PhD Thesis, Doctorate School of Chemical Engineering, University of Pannonia, Veszprém, Hungary.

Holenda, B., Domokos, E., Rédey, À., and Fazakas, J., 2008. Dissolved oxygen control of the activated sludge wastewater treatment process using model predictive control. Computers & Chemical Engineering 32(6), 1270–1278.

Husmann, M., Orth, H., Schlegel, S., and Teichgräber, B., 1998. Application of process control for improved nitrogen removal. Water Science and Technology 38(3), 263–269.

Bibliography 221

Ingildsen, P., 2002. Realising full-scale control in wastewater treatment systems using in situ nutrient sensors. PhD Thesis, Department of Industrial Electrical Engineering and Automation, Lund University, Lund, Sweden.

Ingildsen, P., Jeppsson, U., and Olsson, G., 2002. Dissolved oxygen controller based on on-line measurements of ammonium combining feed-forward and feedback. Water Science and Technology 45(4-5), 453–60.

Inoue, T., Iwai, S., and Nakano, M., 1981a. High accuracy control of a proton synchrotron magnet power supply., in: 8th IFAC World Congress. Kyoto, Japan, 24-28 August 1981.

Inoue, T., Nakano, M., and Iwai, S., 1981b. High accuracy control of servomechanism for repeated contouring, in: 10th Annual Symposium on Incremental Motion Control Systems and Devices.

Jeppsson, U., 1996. Modelling aspects of wastewater treatment processes. PhD Thesis, Department of Industrial Electrical Engineering and Automation, Lund University, Lund, Sweden.

Jeppsson, U., Alex, J., Batstone, D., Benedetti, L., Comas, J., Copp, J.B., Corominas, L., Flores-Alsina, X., Gernaey, K.V., Nopens, Ingmar, Pons, M.N., Rodriguez-Roda, I., Rosen, C., Steyer, J-P., Vanrolleghem, P.A., and Vrečko, D., 2011. Quo vadis benchmark simulation models?, in: 8th IWA Symposium on Systems Analysis and Integrated Assessment. San Sebastian, Spain, 20-22 June 2011, pp. 493–506.

Jeppsson, U., Alex, J., Pons, M.N., Spanjers, H., and Vanrolleghem, P.A., 2002. Status and future trends of ICA in wastewater treatment--a European perspective. Water Science and Technology 45(4-5), 485–94.

Jeppsson, U., Rosen, C., Alex, J., Copp, J.B., Gernaey, K.V., Pons, M.N., and Vanrolleghem, P.A., 2006. Towards a benchmark simulation model for plant-wide control strategy performance evaluation of WWTPs. Water Science and Technology 53(1), 287–295.

Jeppsson, U., and Vanrolleghem, P.A., 2014. Publications associated with the Benchmark Simulation Models, in: Gernaey, K., Jeppsson, U., Vanrolleghem, P.A., Copp, J., & Steyer, J.-P. (Eds.), Benchmarking of Control Strategies for Wastewater Treatment Plants. IWA Scientific and Technical Report, ISBN 9781843391463. IWA Publishing, London.

Kaelin, D., Rieger, L., Eugster, J., Rottermann, K., Bänninger, C., and Siegrist, H., 2008. Potential of in-situ sensors with ion-selective electrodes for aeration control at wastewater treatment plants. Water Science and Technology 58(3), 629–37.

Kampschreur, M.J., Temmink, H., Kleerebezem, R., Jetten, M.S.M., and Van Loosdrecht, M. C. M., 2009. Nitrous oxide emission during wastewater treatment. Water research 43(17), 4093–103.

222 Bibliography

Kandare, G., and Nevado Reviriego, A., 2011. Adaptive predictive expert control of dissolved oxygen concentration in a wastewater treatment plant. Water Science and Technology 64(5), 1130–1136.

Keskar, P., 2006. Control and Optimization of Unit Operations - Aeration and DO control, in: Lipták, B.G. (Ed.), Instrument Engineers’ Handbook - Process Control and Optimization. CRC Press, Boka Raton, FL, US.

Kim, YH, Yoo, C., Kim, YS, and Lee, I.-B., 2009. Simulation and Activated Sludge Model-Based Iterative Learning Control for a Sequencing Batch Reactor. Environmental Engineering Science 26(3), 661–671.

Krause, K., Böcker, K., and Londong, J., 2002. Simulation of a nitrification control concept considering influent ammonium load. Water Science and Technology 45(4-5), 413–20.

Käppalaförbundet, 2013. Miljörapport 2012. In Swedish. Langergraber, G., Alex, J., Weissenbacher, N., Woerner, D., Ahnert, M.,

Frehmann, T., Halft, N., Hobus, I., Plattes, M., Spering, V., and Winkler, S., 2008. Generation of diurnal variation for influent data for dynamic simulation. Water Science and Technology 57(9), 1483–6.

Lazić, A., Larsson, V., and Nordenborg, Å., 2012. Energy savings potential of new aeration system: Full scale trials. Water Practice & Technology 7(4), 1–9.

Levine, A.D., and Asano, T., 2004. Recovering sustainable water from wastewater. Environmental science & technology 38(11), 201A–208A.

Li, C., Zhang, D., and Xianyi, Z., 2004. A survey of repetitive control, in: EEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, Sendai, Japan, 28 September - 2 October, 2004, pp. 1160–1166.

Lindberg, C.F., 1997. Control and estimation strategies applied to the activated sludge process. PhD Thesis, Department of Material Science, Uppsala University, Uppsala, Sweden.

Lindblom, E., 2011. Uppställande och kalibrering av simuleringsmodell för test av driftstrategier på Bromma reningsverk. Sweco Environment AB. In Swedish.

Liu, H., Ramnarayanan, R., and Logan, B.E., 2004. Production of electricity during wastewater treatment using a single chamber microbial fuel cell. Environmental science & technology 38(7), 2281–5.

Liu, W., Lee, G., Schloth, P., and Serra, M., 2005. Side by side comparison demonstrated a 36% increase of nitrogen removal and 19% reduction of aeration requirements using a feed forward online optimization system, in: WEFTEC.05. Washington, DC, US, October 29 - November 2, pp. 3447–3455.

Longman, R., 2000. Iterative learning control and repetitive control for engineering practice. International Journal of Control 73(10), 930–954.

Bibliography 223

Lotito, A.M., Wunderlin, P., Joss, A., Kipf, M., and Siegrist, H., 2012. Nitrous oxide emissions from the oxidation tank of a pilot activated sludge plant. Water research 46(11), 3563–73.

Lumley, D., 2002. On-line instrument confirmation: how can we check that our instruments are working? Water Science and Technology 45(4-5), 469–476.

Luo, J., and Biegler, L.T., 2011. Dynamic optimization of aeration operations for a benchmark wastewater treatment plant, in: Preprints of the 18th IFAC World Congress. Milano, Italy, August 28 - September 2, 2011, pp. 14189–14194.

Machado, V.C., Gabriel, D., Lafuente, J., and Baeza, J.A., 2009. Cost and effluent quality controllers design based on the relative gain array for a nutrient removal WWTP. Water research 43(20), 5129–41.

Martins, M.P., Pagilla, K., Heijnen, J.J., and van Loosdrecht, M.C.M., 2004. Filamentous bulking sludge — a critical review. Water research 38, 793–817.

Metcalf & Eddy, Tchobanoglous, G., Burton, F.L., and Stensel, H.D., 2003. Wastewater Engineering - Treatment and Reuse, 4th ed. McGraw-Hill, New York.

Meyer, U., and Pöpel, H.J., 2003. Fuzzy-control for improved nitrogen removal and energy saving in WWT-plants with pre-denitrification. Water Science and Technology 47(11), 69–76.

Monod, J., 1942. Recherches sur la croissance des cultures bactériennes. Hermann: Paris.

Murphy, R.B., Young, B.R., and Kecman, V., 2009. Optimising operation of a biological wastewater treatment application. ISA transactions 48(1), 93–7.

Nielsen, M.K., and Önnerth, T., 1995. Improvement of a recirculating plant by introducing STAR control. Water Science and Technology 31(2), 171–180.

Nopens, I., Benedetti, L., Jeppsson, U., Pons, M.N., Alex, J., Copp, J.B., Gernaey, K.V., Rosen, C., Steyer, J.-P., and Vanrolleghem, P.A., 2010. Benchmark Simulation Model No 2: finalisation of plant layout and default control strategy. Water Science and Technology 62(9), 1967–1974.

Nygren, J., Pelckmans, K., and Carlsson, B., 2014. Approximate adjoint-based iterative learning control. International Journal of Control 87(5), 1028–1046.

O’Brien, M., Mack, J., Lennox, B., Lovett, D., and Wall, A., 2011. Model predictive control of an activated sludge process: A case study. Control Engineering Practice 19(1), 54–61.

Olsson, G., Carlsson, B., Comas, J., Copp, J.B., Gernaey, K.V., Ingildsen, P., Jeppsson, U., Kim, C., Rieger, L., Rodriguez-Roda, I., Steyer, J.-P., Takács, I., Vanrolleghem, P.A., Vargas, A., Yuan, Z., and Åmand, L., 2014. Instrumentation, Control and Automation in wastewater- from

224 Bibliography

London 1973 to Narbonne 2013. Water Science and Technology. doi: 10.2166/wst.2014.057

Olsson, G., and Newell, B., 1999. Wastewater Treatment Systems: Modelling, Diagnosis and Control. IWA Publishing, London, UK.

Olsson, G., Nielsen, M.K., Yuan, Z., Lynggaard-Jensen, A., and Steyer, J-P., 2005. Instrumentation, control and automation in wastewater systems. Scientific and Technical Report No. 11, IWA Publishing, London, UK.

Olsson, G., Rundqwist, L., Eriksson, L., and Hall, L., 1985. Self-tuning control of the dissolved oxygen concentration in activated sludge systems, in: Drake, R.A.R. (Ed.), Instrumentation and Control of Water and Wastewater Treatment and Transport Systems,. Pergamon Press, Oxford, pp. 473–480.

Ostace, G.S., Cristea, V.M., and Agachi, P.Ş., 2011. Cost reduction of the wastewater treatment plant operation by MPC based on modified ASM1 with two-step nitrification/denitrification model. Computers & Chemical Engineering 35(11), 2469–2479.

Piotrowski, R., Brdys, M.A., Konarczak, K., Duzinkiewicz, K., and Chotkowski, W., 2008. Hierarchical dissolved oxygen control for activated sludge processes. Control Engineering Practice 16(1), 114–131.

Rieger, L., Alex, J., Gujer, W., and Siegrist, H., 2006. Modelling of aeration systems at wastewater treatment plants. Water Science and Technology 53(4-5), 439–447.

Rieger, L., Alex, J., Winkler, S., Boehler, M., Thomann, M., and Siegrist, H., 2003. Progress in sensor technology--progress in process control? Part 1: sensor property investigation and classification. Water Science and Technology 47(2), 103–12.

Rieger, L., Bott, Charles B, Balzer, W.J., Jones, Richard M, and Beach, V., 2012a. Model-based aeration systems design - case study Nansemond WWTP, in: WEFTEC.12. New Orleans, Louisiana, USA, September 29 – October 3, 2012.

Rieger, L., Gillot, S, Langergraber, G., Ohtsuki, T., Shaw, A., Takács, I., and Winkler, S., 2013. Guidelines for using activated sludge models. Scientific and Technical Report No. 22. IWA Task Group on Good Modelling Practice. IWA Publishing, London, UK.

Rieger, L., Jones, R.M., Dold, P.L., and Bott, C.B., 2012b. Myths about ammonia feedforward aeration control, in: WEFTEC.12. New Orleans, Louisiana, USA, September 29 – October 3, 2012.

Rieger, L., and Siegrist, H., 2002. In-situ measurement of ammonium and nitrate in the activated sludge process. Water Science and Technology 45(4-5), 93–100.

Rieger, L., Takács, I., and Siegrist, H., 2012c. Improving nutrient removal while reducing energy use at three Swiss WWTPs using advanced control. Water environment research 84(2), 170–188.

Bibliography 225

Rieger, L., Thomann, M., Gujer, W., and Siegrist, H., 2005. Quantifying the uncertainty of on-line sensors at WWTPs during field operation. Water research 39(20), 5162–74.

Rosen, C., 2001. A chemometric approach to process monitoring and control-with applications to wastewater treatment operation. PhD Thesis, Department of Industrial Electrical Engineering and Automation, Lund University, Lund.

Rosen, C., Jeppsson, U., Rieger, L., and Vanrolleghem, P.A., 2008. Adding realism to simulated sensors and actuators. Water Science and Technology 57(3), 337–44.

Rosso, D., Stenstrom, M.K., and Larson, L.E., 2008. Aeration of large-scale municipal wastewater treatment plants: state of the art. Water Science and Technology 57(7), 973–8.

Rugh, W.J., and Shamma, J.S., 2000. Research on gain scheduling. Automatica 36, 1401–1425.

Sahlmann, C., Libra, J.A., Schuchardt, A., Wiesmann, U., and Gnirss, R., 2004. A control strategy for reducing aeration costs during low loading periods. Water Science and Technology 50(7), 61–8.

Samuelsson, P., and Carlsson, B., 2002. Control of the aeration volume in an activated sludge process with nitrogen removal. Water Science and Technology 45(4-5), 45–52.

Sanchez, A., and Katebi, M.R., 2003. Predictive control of dissolved oxygen in an activated sludge wastewater treatment plant, in: In Proceedings of the European Control Conference ECC’2003. Cambridge, UK, 1-4 September 2003.

Satoh, H., Mino, T., and Matsuo, T., 1999. PHA production by activated sludge. International journal of biological macromolecules 25(1-3), 105–9.

Seborg, D.E., Mellichamp, D.A., Edgar, T.F., and Francis III, J.D., 2010. Process Dynamics and Control, 3rd ed. John Wiley & Sons, Inc., New York.

Serralta, J., Ribes, J., Seco, A., and Ferrer, J., 2002. A supervisory control system for optimising nitrogen removal and aeration energy consumption in wastewater treatment plants. Water Science and Technology 45(4-5), 309–16.

Shen, T., Qiu, Y., and Shi, H., 2010. Mathematical simulation of feedforward control for nutrient removal in anaerobic–anoxic–oxic processes. Environmental Engineering Science 27(8), 633–641.

Shen, W., Chen, X., and Corriou, J.P., 2008. Application of model predictive control to the BSM1 benchmark of wastewater treatment process. Computers & Chemical Engineering 32(12), 2849–2856.

Shen, W., Chen, X., Pons, M.N., and Corriou, J.P., 2009. Model predictive control for wastewater treatment process with feedforward compensation. Chemical Engineering Journal 155(1-2), 161–174.

226 Bibliography

Snedecor, G.W., and Cochran, W.G., 1989. Statistical methods. 8th edition, Iowa State University Press, Ames, Iowa, U.S.

Stare, A., Hvala, N., and Vrečko, D., 2006. Modeling, identification, and validation of models for predictive ammonia control in a wastewater treatment plant--a case study. ISA transactions 45(2), 159–174.

Stare, A., Vrečko, D., Hvala, N., and Strmcnik, S., 2007. Comparison of control strategies for nitrogen removal in an activated sludge process in terms of operating costs: a simulation study. Water research 41(9), 2004–14.

Steffens, M.A., and Lant, P.A., 1999. Multivariable control of nutrient-removing activated sludge systems. Water Research 33(12), 2864–2878.

Steinbuch, M., 2002. Repetitive control for systems with uncertain period-time. Automatica 38(12), 2103–2109.

Stenstrom, M.K., and Poduska, R.A., 1980. The effect of dissolved oxygen concentration on nitrification. Water Research 14(6), 643–649.

Stockholm Vatten, 2013. Miljörapport 2012. In Swedish. Suescun, J., Ostolaza, X., Garcia-Sanz, M., and Ayesa, E., 2001. Real-time

control strategies for predenitrification-nitrification activated sludge plants biodegradation control. Water Science and Technology 43(1), 209–16.

Svardal, K., Lindtner, S., and Winkler, S., 2003. Optimum aerobic volume control based on continuous in-line oxygen uptake monitoring. Water Science and Technology 47(11), 305–12.

Sweeney, M.W., and Kabouris, J.C., 2011. Modeling, Instrumentation, Automation, and Optimization of Wastewater Treatment Facilities. Water environment research (Literature Review), 1347–1363.

Svenska Vatten- och Avloppsverksföreningen (VAV), 1988. Mätning, mätvärdesbearbetning och preglering i reningsverk. VAV P62. In Swedish.

Svenskt Vatten Utveckling, 2013. VA-verkens energianvändning 2011. SVU Rapport 2013-17. In Swedish.

Syvab, 2013. Miljörapport 2012. In Swedish. Takács, I., Patry, G.G., and Nolasco, D., 1991. A dynamic model of the

clarification-thickening process. Water Research 25(10), 1263–1271. Thomann, M., Rieger, L., Frommhold, S., Siegrist, H., and Gujer, W., 2002.

An efficient monitoring concept with control charts for on-line sensors. Water Science and Technology 46(4-5), 107–16.

Thornberg, D.E., Nielsen, M.K., and Eriksson, J., 1998. Upgrading of Borås Wastewater treatment ålant based on intelligent process and operation control. Water Science and Technology 37(9), 57–63.

Thornberg, D.E., Nielsen, M.K., and Lynggaard-Jensen, A., 1992. A new approach to control of wastewater treatment plants. The STAR concept. IWSA Yearbook.

Bibliography 227

Thornton, A., Sunner, N., and Haeck, M., 2010. Real time control for reduced aeration and chemical consumption: a full scale study. Water Science and Technology 61(9), 2169–75.

Thunberg, A., Erikstam, S., and Åmand, L., 2013. Modelling the effect of more stringent discharge limits on the carbon footprint at Käppala WWTP, in: 13th Nordic Wastewater Conference 2013. Malmö, Sweden, 8-10 October 2013.

Thunberg, A., Sundin, A.-M., and Carlsson, B., 2009. Energy optimization of the aeration process at Käppala wastewater treatment plant, in: 10th IWA Conference on Instrumentation, Control and Automation. Cairns, Australia, 14-17 June 2009.

Traoré, A., Grieu, S., Puig, S., Corominas, L., Thiery, F., Polit, M., and Colprim, J., 2005. Fuzzy control of dissolved oxygen in a sequencing batch reactor pilot plant. Chemical Engineering Journal 111(1), 13–19.

Tzoneva, R., 2007. Optimal PID control of the dissolved oxygen concentration in the wastewater treatment plant, in: AFRICON 2007. Windhoek, Namibia, 26-28 September 2007, pp. 1–7.

Walz, T., Coughenour II, J.R., Williams, K., Jacobs, J., Shone, L., Stahl, T., Kestel, S., and Palmer, D., 2009. Energy savings at Phoenix 23rd Avenue wastewater treatment plant using feed-forward process control, in: WEFTEC.09. Orlando, Florida, US, 12-14 October, pp. 3722–3729.

Wang, Y., Gao, F., and Doyle, F.J., 2009. Survey on iterative learning control, repetitive control, and run-to-run control. Journal of Process Control 19(10), 1589–1600.

Vanrolleghem, P.A., 2001. Model simulation for improved operation and control of wastewater treatment plants, in: Proceedings Seminaire La Modélisation En Assainissement: Remise En Cause Des Pratiques? Strasbourg, France, January 24 2001, p. 23.

Vanrolleghem, P.A., and Lee, D.S., 2003. On-line monitoring equipment for wastewater treatment processes: state of the art. Water Science and Technology 47(2), 1–34.

Weijers, S., 2000. Modelling, identification and control of activated sludge plants for nitrogen removal. PhD Thesis, Technische Universiteit Eindhoven, Eindhoven, The Netherlands.

Wiesmann, U., Choi, I.S., and Dombrowski, E.-M., 2007. Fundamentals of Biological Wastewater Treatment. WILEY-VCH GmbH & Co, KGaA: Weinheim.

Vilanova, R., Katebi, R., and Alfaro, V., 2009. Multi-loop PI-based control strategies for the activated sludge process, in: 2009 IEEE Conference on Emerging Technologies & Factory Automation. IEEE, Mallorca, Spain, 22-25 September 2009, pp. 1–8.

WPCF & ASCE, 1988. Aeration. American Society of Civil Engineers, WPCF Manual of Practice FD-13 (WPCF), Manuals and reports on engineering practice No.63 (ASCE).

228 Bibliography

Vrečko, D., Hvala, N., and Carlsson, B., 2003. Feedforward-feedback control of an activated sludge process: a simulation study. Water Science and Technology 47(12), 19–26.

Vrečko, D., Hvala, N., and Kocijan, J., 2002. Wastewater treatment benchmark: what can be achieved with simple control? Water Science and Technology 45(4-5), 127–34.

Vrečko, D., Hvala, N., Stare, A., Burica, O., Stražar, M., Levstek, M., Cerar, P., and Podbevsek, S., 2006. Improvement of ammonia removal in activated sludge process with feedforward-feedback aeration controllers. Water Science and Technology 53(4-5), 125–132.

Vrečko, D., Hvala, N., and Stražar, M., 2011. The application of model predictive control of ammonia nitrogen in an activated sludge process. Water Science and Technology 64(5), 1115–1121.

Vrečko, D., Zupančič, U., and Babič, R., 2013. Improving aeration control at the Ljubljana wastewater treatment plant. Water Science and Technology. doi: 10.2166/wst.2013.815.

Yamanaka, O., Obara, T., and Yamamoto, K., 2006. Total cost minimization control scheme for biological wastewater treatment process and its evaluation based on the COST benchmark process. Water Science and Technology 53(4-5), 203–214.

Yang, T., Zhang, L., and Chadli, M., 2013. Fuzzy modeling of activated sludge wastewater treatment processes and predictive control of dissolved oxygen. 2013 IEEE International Symposium on Industrial Electronics 1–6.

Yong, M., Yongzhen, P., and Shuying, W., 2005. Feedforward-feedback control of dissolved oxygen concentration in a predenitrification system. Bioprocess and biosystems engineering 27(4), 223–8.

Yong, M., Yong-zhen, P., Xiao-lian, W., and Shu-ying, W., 2006. Intelligent control aeration and external carbon addition for improving nitrogen removal. Environmental Modelling & Software 21(6), 821–828.

Yoo, C.K., and Kim, M.H., 2009. Industrial experience of process identification and set-point decision algorithm in a full-scale treatment plant. Journal of environmental management 90(8), 2823–30.

Zarrad, W., Harmand, J., Devisscher, M., and Steyer, J-P., 2004. Comparison of advanced control strategies for improving the monitoring of activated sludge processes. Control Engineering Practice 12(3), 323–333.

Åmand, L., and Carlsson, B., 2011. Learning from your mistakes: Repetitive aeration control in continuous wastewater treatment plants, in: 12th Nordic Wastewater Conference. Helsinki, Finland, 14-16 November 2011.

Åmand, L., and Carlsson, B., 2012a. Optimal aeration control in a nitrifying activated sludge process. Water research 46, 2101–10.

Bibliography 229

Åmand, L., and Carlsson, B., 2012b. Energy efficient ammonium feedback control, in: New Developments in IT & Water. Amsterdam, The Netherlands, 4-6 November 2012.

Åmand, L., and Carlsson, B., 2013a. The optimal dissolved oxygen profile in a nitrifying activated sludge process - comparisons to ammonium feedback control. Water Science and Technology 68, 641–649.

Åmand, L., and Carlsson, B., 2013b. Parameter scheduling in ammonium feedback control, in: 11th IWA Conference on Instrumentation, Control and Automation. Narbonne, France, 18-20 September 2013.

Åmand, L., and Carlsson, B., 2014. Aeration control with gain scheduling in a full-scale wastewater treatment plant, in: 19th IFAC World Congress. Cape Town, South Africa, 24-29 August 2014.

Åmand, L., Laurell, C., Stark-Fujii, K., Thunberg, A., and Carlsson, B., 2013a. Long-term evaluation of full-scale ammonium control in three large WWTPs, in: 11th IWA Conference on Instrumentation, Control and Automation. Narbonne, France, 18-20 September 2013.

Åmand, L., Laurell, C., Stark-Fujii, K., Thunberg, A., and Carlsson, B., 2013b. Full-scale ammonium control in three wastewater treatment plants in Stockholm, in: 13th Nordic Wastewater Conference 2013. Malmö, Sweden, 8-10 October 2013.

Åmand, L., Laurell, C., Stark-Fujii, K., Thunberg, A., and Carlsson, B., 2014. Lessons learnt from evaluating full-scale ammonium feedback control in three large wastewater treatment plants. Water Science and Technology 69(7), doi:10.2166/wst.2014.061.

Åmand, L., Nygren, J., and Carlsson, B., 2011. Applications of repetitive control in activated sludge processes, in: 8th IWA Symposium on Systems Analysis and Integrated Assessment. San Sebastian, Spain, 20-22 June 2011.

Åmand, L., Olsson, G., and Carlsson, B., 2013c. Aeration control – a review. Water Science and Technology 67(11), 2374–2398.

Åström, K.J., 1999. Automatic control - the hidden technology, in: Frank, P. (Ed.), Advances in Control–Highlights of ECC’99. London: European Control Council, Springer-Verlag, pp. 1–29.

Åström, K.J., and Hägglund, T., 1995. PID controllers: theory, design and tuning, 2nd ed. Instrument Society of America, Research Triangle Park, NC, U.S.

Åström, K.J., and Hägglund, T., 2006. Advanced PID control. ISA- The Instrumentation, Systems and Automation Society, USA.

Åström, K.J., and Kumar, P.R., 2014. Control: A perspective. Automatica 50, 3–43.

Åström, K.J., and Rundqwist, L., 1989. Integrator windup and how to avoid it, in: American Control Conference. Pittsburgh, P.A., 21-23 June, 1989, pp. 1693–1698.

230 Bibliography

Önnerth, T., Nielsen, M.K., and Stamer, C., 1996. Advanced computer control based on real and software sensors. Water Science and Technology 33(1), 237–245.

APPENDIX

233

A A CALIBRATION OF ACTIVATED SLUDGE

MODELS

HE CALIBRATION OF activated sludge models representing Hen-riksdal, Käppala and Himmerfjärden WWTPs have provided the pos-sibility to evaluate process controllers and compare those to full-scale

operation. It has also made it possible to simulate control strategies not eval-uated in full-scale operation. This appendix describes the set-up, calibration and validation of the activated sludge models in MATLAB/Simulink®.

Objective A.1The objective of the model calibration was to create dynamic process models describing the nitrogen removal at Käppala, Henriksdal and Himmerfjärden WWTPs, with the goal to be able to simulate aeration control strategies for a one year period.

Overall method A.2The models were built in MATLAB/Simulink®, based on the BSM1_LT (Gernaey et al., 2013) model. BSM1_LT is an extension of the BSM1 plat-form (Copp, 2002) offering the possibility to simulate temperature depend-ence of the model parameters as well as sensor failure and inhibition of mi-croorganisms. The BSM1_LT standard implementation was changed to fit the respective treatment plant. The inhibition option in the BSM1_LT was turned off since inhibition had a too large negative impact on nitrification. Sensor failure was not modelled In general, the modelling procedure has consisted of the following steps (Rieger et al., 2013):

T

234 Appendix A: Calibration of activated sludge models

1. Problem formulation The modelling covers the activated sludge process with the inlet being pretreated wastewater and the effluent is water leaving the secondary settler. The emphasis in calibration was on effluent ammonium and ni-trate concentrations and on air flow rate. Along the modelling process, regular meetings discussing the progress were held with personnel from the treatment plants.

2. Data collection and reconciliation Process data was collected from each of the plants and calibration was performed for data from a one year period. On-line sensor measuring air flow rate, DO, sludge concentrations, flows and temperatures were com-bined with composite weekly samples of carbon and nitrogen com-pounds. The time frame for the modelling did not allow for a dedicated measurement campaign at the plants, which is recommended especially to be able to have a proper fractionation of incoming COD to the plant since the ASM1 model is COD-based (Rieger et al., 2013).

3. Plant-model set-up For each plant, one separate treatment line was modelled to be able to compare with plant data for that specific treatment line. RAS, WAS and internal recycle flows were modelled. Side streams such as wash water from sand filters and digestate supernatant were indirectly included in the inflow. Overflow was not modelled.

4. Calibration and validation The stopping criteria for model calibration were to reach within 0.5 mg/l of ammonium compared to plant data, and 10 % in energy consumption. The sludge balance was calibrated first, followed by nitrification, denitri-fication and finally the air flow model. Simulations with a constant in-fluent gave a preliminary calibration of the sludge balance before resolv-ing to dynamic simulations. The ASM1 model does not include inert suspended solids. Instead of rebuilding the model to contain inert mate-rial the sludge balance was calibrated based on volatile suspended solids. The influent fractionation was manipulated as the first choice during cal-ibration before taking to adjusting the model parameters in ASM1. The recent suggested update of the anoxic yield of heterotrophs (YOHO,Ax = 0.54) was used (Choubert et al., 2009) since this provided the best fit to nitrate data. As a guideline, the soluble inert material was matched to the COD concentration after the secondary settler.

Daily variations in concentrations in the influent were calculated using the same method for all the plants. The method is a development of the method presented in Lindblom (2011). An assumption in the method is that the con-centration variation depends on the household wastewater flow, and addi-tional dilution occurs during rain events. By calculating a normalised dry-weather inflow vector a concentration profile is created.

A.3 The Henriksdal model 235

The BSM1_LT model contains the Takács sedimentation model (Takács et al., 1991), which was used in the modelling.

The ASM1 model does not model air flow rate. Instead the model ex-presses an indirect energy requirement in terms of the oxygen transfer rate coefficient, KLa. Air flow rate and power consumption depend on the plant equipment and cannot be generalised into one universal model. To be able to communicate energy consumption in terms of air flow rate and compare model data to plant data, the Biowin aeration model (Dold and Fairlamb, 2001) was implemented in the BSM1_LT model in MATLAB. The Biowin aeration model is an empirical model relating air flow rate to KLa:

Y

SGL CUaK = (A.1)

2

25.0

1 kDDkC += (A.2)

KLa = oxygen transfer rate (1/d) USG = specific air flow rate (m3/m2,d) C = parameter reflecting the diffuser density Y = exponential factor DD = diffuser density (%) k1, k2 = model parameters (1/d)

The Henriksdal model A.3 Data collection A.3.1

2011 was chosen as the calibration year for the Henriksdal model. During 2012 the plant had problems with floating sludge, making it difficult to mod-el the sludge balance. Instead, a part of 2012 was chosen for model valida-tion.

Composite weekly samples of pretreated wastewater were used as a basis to calculate the influent to the plant. For a few occasions weekly samples were missing. Data was interpolated for those periods. COD was not meas-ured but calculated from measured TOC (COD = 3.3*TOC, relationship from historical lab data). This relationship is not exact, creating an uncertain-ty in the influent data. The incoming COD and incoming total nitrogen con-centration are given in Figure A.1.


(a) (b)

Figure A.1. (a) Calculated COD and (b) total nitrogen concentration in pretreated wastewater at Henriksdal WWTP 2011. Weekly composite samples.

Operational data from treatment line 4 was used when possible. Treatment line 4 had good data quality during 2011 and the line was one of the experi-mental lines in the full-scale experiments (see Section 8.6). The inflow to line 4, sludge concentrations, RAS, WAS, and the settler mass balance is given in Figure A.2. The sludge balance is not perfect around the settler. It is suspected that the reason for this has to do with sensor problems. For in-stance, towards the end of the year the sludge balance is lower than one which could be due to a too high measurement of SS in RAS. The VSS:TSS ratio was 0.69 in secondary settled sludge and 0.75 in primary settled sludge.

(a) (b) (c)

(d) (e) (f)

Figure A.2. (a) Influent flow, (b) MLSS, (c) SS in RAS, (d) RAS flow, (e) WAS flow and (f) sludge balance over secondary settler. Data from treatment line 4 at Henriksdal WWTP 2011.

Data was collected from the on-line ammonium and nitrate sensors situated after each of the secondary settlers and based on composite weekly samples on effluent wastewater leaving the plant. The ammonium concentration


200

250

300

350

400

450

500

CO

D (

mg/

l)


25

30

35

40

45

50

55

Tot

al n

itrog

en (

mg/

l)

Jan FebMar Apr May Jun Jul Aug Sep Oct NovDec0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Qin

(m

3/s)


1500

2000

2500

3000

3500

MLS

S (

mg/

l)


6000

7000

8000

9000

10000

11000

12000

13000

SS

RA

S (

mg/

l)

Jan FebMar Apr May Jun Jul Aug Sep Oct NovDec0.5

1

1.5

2

2.5

3

3.5

4x 10

4

RA

S (

m3/

d)


200

400

600

800

1000

1200

1400

WA

S (

m3/

d)


0.5

0.6

0.7

0.8

0.9

1

1.1

1.2

1.3

Slu

dge

bala

nce

settl

er (

slud

ge in

/slu

dge

out)


should be comparable between the lab measurements and on-line measure-ments. During 2011 this was not the case with ammonium, where the am-monium sensors during part of the year showed considerably higher concen-trations (Figure A.3). The decision was made to base the calibration on the composite samples, since during the validation period in 2012, the sensors and weekly samples matched well (Figure A.3).

The on-line nitrate sensor in treatment line 4 was unreliable during parts of 2011. The sensor in line 5 was used for calibration instead. Measured oxygen concentrations (daily averages) were used during calibration.

(a) (b)

Figure A.3. Effluent NH4 from on-line and lab measurements. (a) 2011 and (b) 2012.

Model set-up A.3.2Modelled flows Bypass of pretreated wastewater around the activated sludge process occurs during high flows. There is a flow meter at the inlet to each treatment line, and data from this sensor was used in the calibration. Sludge supernatant is added before the sampling point after the primary settler. Water from sand filter cleaning was assumed to be small enough to have a marginal effect on the results.

Hydraulics In 2005 a tracer experiment was performed at Henriksdal WWTP to calcu-late the number of tanks-in-series to be modelled with an ASM1 model. Ac-cording to the experiments the best results were found when modelling 6 zones for denitrification, 2 zones for nitrification and 2 non-aerated zones after nitrification. In this study, 8 zones was chosen (1 mixing zone, 3 anoxic zones, 3 aerobic zones and 1 deox zone). This is the actual zone set-up in the plant (Figure 4.3), and it allowed for control of DO in the same way as in the plant. When modelling 6 anoxic zones instead of 3 to comply with the re-sults in the tracer experiment, the effect on the overall simulation results as well as the concentration profile in the model was barely noticeable. Hence, the zone division was the same as in Figure 4.3.


2

4

6

8

10

12

Am

mon

ium

con

cent

ratio

n (m

g/l)


2

4

6

8

10

12

Am

mon

ium

con

cent

ratio

n (m

g/l)

2011 2012


Influent model Nitrogen is divided into ammonium nitrogen and organic nitrogen based on measurements. Fractioning of COD was a part of the calibration task. Vola-tile particulate material to the biological reactors was calculated based on measured TSS in pretreated wastewater and the VSS:TSS ratio was assumed to be the same as that of primary settled sludge. Particulate COD:VSS frac-tion as well as the fraction of total and soluble inert material was calibrated. Heterotrophic biomass was assumed to be 10 % of incoming COD. Active biomass has been reported to constitute up to 15 % of total incoming COD (Rieger et al., 2013), and 10 % is a commonly assumed number. The calibra-tion of COD fractions is uncertain since too few measurements were availa-ble.

Air flow model The Biowin model was used, see Eq. (A.1) and (A.2).

Model calibration 2011 A.3.3The sludge concentration did not fit well to the measured data during periods in 2011. This could be due to the loss of sludge at the plant caused by leak-ing basins. The sludge production was difficult to fit to measured data, de-spite manipulation of the settling parameters in the Takács model.

Calibration of nitrification and denitrification was well described with the BSM1 standard parameter values. The results of the model calibration are found in Figure A.4 and Figure A.5. The modelled ammonium concentration fits better to the lab data than to the on-line sensor (Figure A.5) as discussed above. The dip in nitrate concentration in October to November was due to a reduced flow to line 4 (Figure A.2).

Model validation 2012 A.3.4Since treatment line 4 was turned down during summer 2012 validation was performed based on data from treatment line 5. A period from February to June was not included in the validation due to sludge property problems. The validation results for 2012 are found in Figure A.6 and Figure A.7. The sus-pended solids concentration is too low in the model. The reason could be that the calculated COD concentration in the influent is nearly 50 mg/l lower in 2012 than in 2011and that the COD fractionation is not representative for 2012. The lower suspended solids concentration leads to slightly higher am-monium concentrations.


(a) (b) (c)

(d) (e)

Figure A.4. Calibration results for Henriksdal WWTP 2011 (a) MLVSS, (b) aerobic sludge age, (c) air flow rate, (d) NH4 and (e) NO3. Periods of bypassing are removed from the plant NH4 lab data. data. Note that the NO3 sensor in treatment line 5 is used instead of the NO3 sensor in line 5.

(a) (b)

Figure A.5. Calibration results for Henriksdal WWTP 2011. (a) Lab NH4 and (b) lab NO3 compared to weekly averages of model data.


1000

1500

2000

2500

MLV

SS

(m

g/l)


5

10

15

Aer

obic

slu

dge

age

(d)


0.4

0.6

0.8

1

1.2

1.4

1.6

1.8x 10

5

Air

flow

rat

e (N

m3 /d

)

Model data (daily) Plant data (daily)


5

10

15

Am

mon

ium

con

cent

ratio

n (m

g/l)


2

4

6

8

10

Nitr

ate

conc

entr

atio

n (m

g/l)

Model data (15 min data) Plant data (hourly data) Plant data (weekly data)


0.5

1

1.5

2

2.5

3

3.5

Am

mon

ium

con

cent

ratio

n (m

g/l)


2

4

6

8

10

Nitr

ate

conc

entr

atio

n (m

g/l)

Model data (weekly) Plant data (weekly)


(a) (b) (c)

(d) (e)

Figure A.6. Validation results for Henriksdal WWTP 2012: (a) MLVSS, (b) aerobic sludge age, (c) air flow rate, (d) NH4 and (e) NO3. Periods of bypassing are removed from the plant NH4 lab data.

(a) (b)

Figure A.7. Validation results for Henriksdal WWTP 2012. (a) Lab NH4 and (b) lab NO3 compared to weekly averages of model data.

Jan FebMar AprMay Jun Jul Aug Sep Oct NovDec600

800

1000

1200

1400

1600

1800

2000

2200

2400

MLV

SS

(m

g/l)


5

10

15

Aer

obic

slu

dge

age

(d)


0.5

1

1.5

2

2.5x 10

5

Air

flow

rat

e (N

m3 /d

)



5

10

15

Am

mon

ium

con

cent

ratio

n (m

g/l)


2

4

6

8

10

Nitr

ate

conc

entr

atio

n (m

g/l)

Model data (15 min data) Plant data (hourly data) Plant data (weekly data)


1

2

3

4

5

Am

mon

ium

con

cent

ratio

n (m

g/l)


2

4

6

8

10

Nitr

ate

conc

entr

atio

n (m

g/l)


A.4 The Käppala model 241

Summary of results A.3.5A summary of the calibration and validation is found in Table A.1 and Table A.2. The ammonium concentration and air flow rate are within the stopping criteria.

Table A.1. Summary of calibration results for Henriksdal WWTP 2011.

Plant data Model data

MLVSS (mg/l) 1 553 1 633Aerobic sludge age (d) 7.8 7.7VSS in WAS (mg/l) 5 451 4 890Nitrate (mg/l) 5.7 5.5Ammonium (mg/l) 0.8 0.9Air flow rate (Nm3/d) 91 705 86 500

Table A.2. Summary of validation results for Henriksdal WWTP 2012 (January and July to December).


MLVSS (mg/l) 1 658 1 491Aerobic sludge age (d) 8.1 6.9VSS in WAS (mg/l) 6 965 4 366Nitrate (mg/l) 5.0 5.6Ammonium (mg/l) 0.7 1.2Air flow rate (Nm3/d) 97 602 96 727

The Käppala model A.4 Data collection A.4.1

2011 was initially chosen as the calibration year. After validating the results for the year of 2012, it was clear that the calibration for 2011 did not fit well for 2012. From the end of 2011 and 2012 the ammonium concentrations were too high using the calibrated settings for 2011. Since 2012 is closer in time, 2012 was chosen as the calibration year using parts of 2011 as the vali-dation period.

Composite weekly samples of pretreated wastewater were used as a basis to calculate the influent to the plant. The incoming COD and incoming total nitrogen concentration are given in Figure A.8. Arlanda airport is connected to Käppala WWTP. The airport uses glycol for their runways during winter time. By using storage tanks, the glycol is emitted to the sewage system at intervals during the year. Käppala WWTP knows how much glycol is emit-ted from Arlanda and when, but it is unknown how much of the carbon that reaches the plant. The plant personnel can tell that the glycol has an impact on the denitrification rate at the plant.


(a) (b)

Figure A.8. Influent COD and total nitrogen concentration to Käppala WWTP 2012. Weekly composite samples.

Operational data from the new part of the plant was used, using data from treatment line 11 when possible. The new part of the plant has one single sludge system. The inflow to one line in the new part (calculated from the total flow to the new part), sludge concentrations, RAS, WAS, and the settler mass balance are presented in Figure A.9. The sludge balance is around 1. The oscillations are likely due to the assumption that the suspended solids concentration in the effluent was constant (5 mg/l). This assumption was made by the plant personnel, since there is no SS sensor in place. The VSS:TSS ratio was 0.70 in secondary settled sludge and 0.82 in primary settled sludge.

(a) (b) (c)

(d) (e) (f)

Figure A.9. (a) Influent flow, (b) MLSS, (c) SS in RAS, (d) RAS flow, (e) WAS flow and (f) sludge balance over secondary settler. Data from treatment line 4 at Käppala WWTP 2012.


200

250

300

350

400

CO

D (

mg/

l)


30

35

40

45

50

55

Tot

al n

itrog

en (

mg/

l)


0.2

0.3

0.4

0.5

0.6

0.7

0.8

Qin

(m

3/s)


1500

2000

2500

3000

3500

4000

4500

MLS

S (

mg/

l)


1

1.5

2x 10

4

SS

RA

S (

mg/

l)


0.5

1

1.5

2

2.5x 10

4

RA

S (

m3/

d)


200

250

300

350

400

450

500

550

WA

S (

m3/

d)


0.6

0.8

1

1.2

1.4

1.6

1.8

2

Slu

dge

bala

nce

settl

er (

slud

ge in

/slu

dge

out)


The treatment results were monitored based on composite weekly samples on effluent wastewater leaving the plant. Measured daily oxygen concentra-tions were used during calibration. Since line 11 had ammonium control during 2012, DO concentrations from other treatment lines were used.

Model set-up A.4.2Modelled flows Bypass of pretreated wastewater around the activated sludge process occurs during high flow, but this was very rare during 2011 and 2012 and was not taken into account during modelling. Sludge supernatant is added before the sampling point after the primary settler. Water from sand filter cleaning was assumed to be so small it has marginal effects on the results.

Hydraulics The empirical formulae in Fujie et al. (1983) was used to calculate the num-ber of completely stirred tanks in the aeration zones. The formulae make use of the dispersion number calculated based on aeration data and plant geome-try. The calculation suggested one tank per zone should be used in the aero-bic zones. The dispersion number in the anoxic zones was unknown, but calculations showed that the dispersion number had to be larger than that in the aeration tanks for the anoxic zones to be divided into more than one zone, which was not considered likely.

The same number of zones that is present in the plant (1 mixing zone, 2 anoxic zones, 4 aerobic zones and 1 deox zone) was modelled. This allowed for control of DO in the same way as in the plant. Hence, the zone division was the same as in Figure 4.6.

Influent model Nitrogen was divided into ammonium nitrogen and organic nitrogen based on weekly lab measurements. Fractioning of COD was a part of the calibra-tion task. Volatile particulate material was calculated based on estimated TSS in pretreated wastewater and the VSS:TSS ratio of primary settled sludge. Particulate COD:VSS fraction as well as the fraction of total and soluble inert material was calibrated. Heterotrophic biomass was assumed to be 10 % of incoming COD. The calibration of COD fractions is uncertain since not enough measurements were available.



Model calibration 2012 A.4.3When running simulations with constant influent, it became clear that it was difficult to get the suspended solids concentrations up to plant levels and at the same time meet the denitrification demands. This was confirmed in simulations with a dynamic influent. To reach a denitrification rate which matched the plant data only by adjusting the influent fractionation was not possible. This would require that the easily biodegradable influent concen-tration was higher than measured BOD7 for pretreated wastewater at the plant. Instead, a compromise was made where the readily biodegradable COD part was decreased, allowing a larger part of the influent COD to be inert which increased the MLVSS in the model. The nitrate concentration was calibrated to match measurement data by means of adjusting ASM1 denitrification parameters.

The air flow rate model was intentionally calibrated to model lower air flow rates than measured, since the MLSS concentration was lower in the model than in the plant. The final calibration results are given in Figure A.10. Note that the detection limit of ammonium in the lab analysis was 1 mg/l.

Model validation 2011 A.4.4The results from the model validation are given in Figure A.11. The MLVSS concentration has a very good match to plant data all year, as well as the sludge age. Nitrate and ammonium has a good fit from August and onwards. Before that, the ammonium removal is too good in the model, leading to too high nitrate concentrations. One reason for this anomaly could be that Ar-landa airport used 37 % more glycol in 2011 than in 2012, adding more car-bon source to denitrification. However, the increase in glycol was not even over the whole year. The air flow rate is too high in 2011. If the air flow model would be calibrated to fit 2011, the air flow rate reaches minimum levels during 2012 in the model, which was not the case at the plant.


(a) (b) (c)

(d) (e)

Figure A.10. Calibration results for Käppala WWTP 2012. (a) MLVSS, (b) aerobic sludge age, (c) air flow rate, (d) NH4 and (e) NO3. The detection limit of NH4 in the lab analysis was 1 mg/l.

(a) (b) (c)

(d) (e)

Figure A.11. Validation results for Käppala WWTP 2011. (a) MLVSS, (b) aerobic sludge age, (c) air flow rate, (d) NH4 and (e) NO3. The detection limit of NH4 in the lab analysis was 1 mg/l.


800

1000

1200

1400

1600

1800

2000

2200

MLV

SS

(m

g/l)


4

6

8

10

12

14

16

Aer

obic

slu

dge

age

(d)


3

4

5

6

7

8

9

10x 10

4

Air

flow

rat

e (N

m3 /d

)



1

2

3

4

5

6

Am

mon

ium

con

cent

ratio

n (m

g/l)


5

6

7

8

9

10

11

12

Nitr

ate

conc

entr

atio

n (m

g/l)



800

1000

1200

1400

1600

1800

2000

2200

MLV

SS

(m

g/l)


4

6

8

10

12

14

Aer

obic

slu

dge

age

(d)


3

4

5

6

7

8

9

10x 10

4

Air

flow

rat

e (N

m3 /d

)



1

2

3

4

5

6

7

Am

mon

ium

con

cent

ratio

n (m

g/l)


4

5

6

7

8

9

10

11

Nitr

ate

conc

entr

atio

n (m

g/l)



Summary of results A.4.5A summary of the calibration and validation is found in Table A.3 and Table A.4. The ammonium concentration and air flow rate are within the stopping criteria. The air flow rate is calibrated to be lower since the MLSS is lower.

Table A.3. Summary of calibration results for Käppala WWTP 2012. *The detection limit of NH4 in the lab analysis was 1 mg/l.


MLVSS (mg/l) 1 439 1 283Aerobic sludge age (d) 6.8 6.9VSS in WAS (mg/l) 6 173 4 751Nitrate (mg/l) 6.8 6.7Ammonium (mg/l) 1.2* 0.9Air flow rate (Nm3/d) 54 780 48 135

Table A.4. Summary of validation results for Käppala WWTP 2011 (August –December). *The detection limit of NH4 in the lab analysis was 1 mg/l.


MLVSS (mg/l) 1 466 1 494Aerobic sludge age (d) 7.0 7.8VSS in WAS (mg/l) 6 435 5 428Nitrate (mg/l) 7.1 6.8Ammonium (mg/l) 1.2* 0.7Air flow rate (Nm3/d) 51 478 59 029

The Himmerfjärden model A.5 Data collection A.5.1

2012 was chosen as the year for calibration and 2011 for validation since during parts of 2012 there were more measurements available due to the full-scale evaluation campaign of control strategies. There were no lab samples from water entering the activated sludge process except during August to December 2012. There was a measurement at the inlet to the plant. To calcu-late an influent during the rest of 2012 and for 2011, the influent COD was assumed to be constant at 150 mg/l and influent ammonium after primary clarification was assumed to be the same as influent ammonium. Total nitro-gen in pretreated wastewater was calculated assuming that incoming nitrate was 0 mg/l and that 50 % of the organic nitrogen was removed in the me-chanical treatment steps since this gave a good fit to data from August to December 2012.

A.5 The Himmerfjärden model 247

Suspended solids of primary clarified wastewater was in average 75 mg /l and the VSS:TSS ratio was 0.75. The biological treatment lines are divided into block A and block B, each of the blocks having its own sludge return flow. Data from block A was used. The total flow through the activated sludge system is measured before the post denitrification in the fluidised beds. The flow to block A was assumed to be 10 % higher than the flow to block B, since the air flow rate is in average higher in the four treatment lines in block B.

Flows, suspended solids concentrations and a mass balance over the sec-ondary settler are presented in Figure A.12. More sludge enters the settler than what is measured to leave the settler. This means that there is sludge loss in the system which is not accounted for if only looking at the waste sludge. Possible reasons for this deviation are that the plant manually re-moved large volumes of bad quality sludge during summer 2012 and that the plant has problems with floating sludge which also is removed manually from the sedimentation tanks.

Ammonium is measured after the tertiary sedimentation after block A and block B with on-line sensors. The sensor placed after block A was used in the model calibration. The ammonium concentration from block A and the influent total nitrogen to the activated sludge process are plotted in Figure A.13.

Each aerated zone does not have its own control valve and there are two DO sensors in the six zones in one treatment line (zone 2 and zone 6). See further Section 4.4.3. The DO concentrations in the two zones for one of the treatment lines in block A are presented in Figure A.14.


(a) (b) (c)

(d) (e) (f)

Figure A.12. (a) Influent flow, (b) MLSS, (c) SS in RAS, (d) RAS flow, (e) WAS flow and (f) sludge balance over secondary settler. Data for one treatment line for block A at Himmerfjärden WWTP 2012.

(a) (b)

Figure A.13. Calculated total nitrogen concentration to the activated sludge process and NH4 after the sedimentation.

(a) (b)

Figure A.14. DO concentrations in zone 2 and zone 6 in one treatment line in block A.


0.1

0.15

0.2

0.25

0.3

0.35

0.4

Qin

(m

3/s)


1500

2000

2500

3000

3500

MLS

S (

mg/

l)


4000

5000

6000

7000

8000

9000

10000

11000

12000

SS

RA

S (

mg/

l)


0.8

1

1.2

1.4

1.6

1.8

2

2.2

2.4x 10

4

RA

S (

m3/

d)


20

40

60

80

100

120

140

160

180W

AS

(m

3/d)


1

1.5

2

2.5

3

3.5

4

4.5

Slu

dge

bala

nce

settl

er (

slud

ge in

/slu

dge

out)


18

20

22

24

26

28

30

32

34

Influ

ent t

otal

nitr

ogen

(m

g/l)


2

4

6

8

10

Effl

uent

am

mon

ium

(m

g/l)


2

4

6

8

DO

zon

e 2

(mg/

l)


2

4

6

8

DO

zon

e 6

(mg/

l)


Model set-up A.5.2Modelled flows Bypass of pretreated wastewater around the activated sludge process and the fluidised bed occurs during high flows. Bypass was not included in the simu-lation model since bypassed water is not sampled after the biological treat-ment process. Sludge supernatant is added before the sampling point after the primary settler.

Hydraulics One zone in the plant was modelled as one CSTR in order to be able to con-trol the DO concentration and limit the air flow rate in the same way as in the plant. Also, the method presented in Fujie et al. (1983) suggests that one CSTR per zone should be modelled. The zone division was the same as in Figure 4.9.

Influent model Nitrogen was divided into ammonium nitrogen and organic nitrogen based on measurements and calculations. Fractioning of COD was a part of the calibration task. Volatile particulate material was calculated based on meas-ured TSS in pretreated wastewater and the VSS:TSS ratio of primary settled sludge (assumed as 0.82). Particulate COD:VSS fraction as well as the frac-tion of total and soluble inert material was calibrated. Heterotrophic biomass was assumed to be 10 % of incoming COD. The calibration of COD frac-tions was uncertain since there were not enough measurements.


Calibration 2012 A.5.3Compared to Henriksdal and Käppala WWTPs, there were not enough DO sensors in the plant to be able to simulate measured concentrations in the model. The first step in the calibration process was to simulate the DO con-centrations in the process. The DO concentration profile is increasing in all the treatment lines, but there is a large variation which is likely due to varia-tions in load and performance of equipment. A DO concentration profile was created by controlling the DO concentration to the measured value in zone 1 and dividing the total air flow rate by six. The resulting DO concentration for zone 6 is presented in Figure A.15. This method could have been im-proved if a better model of the aeration system was available.


Figure A.15. Simulated and measured daily DO concentrations in the last aerobic zone.

The sludge concentrations were calibrated next. The WAS flow was manual-ly manipulated for periods of the year to have the MLVSS concentration match measurements. When the MLVSS concentration matched, there was a good match of the VSS in the RAS flow.

Using default ASM1 parameters the nitrification rate became too high. In order to match the nitrification rate at the plant for the whole of 2012 the maximum growth rate of nitrifiers and the half-saturation constant for oxy-gen were changed. If the half-saturation constant was not changed nitrifica-tion was too fast the first part of the year and too slow from July and on-wards. No calibration of denitrification was made since the activated sludge process only nitrifies. The calibration results for 2012 are presented in Fig-ure A.16.

Validation 2011 A.5.4Also for simulations of 2011 the WAS flow was manipulated for the sludge concentration to match measured data. The validation results are given in Figure A.17. The DO concentrations during 2011 were from June and on-wards lower than usual due to another control strategy of the air flow rate at the plant. The DO concentrations for 2011 were chosen to be the same as 2012. This matches well with the DO concentrations for 2011 until June. The lower ammonium concentrations towards the end of 2011 can be ex-plained by the model having higher DO concentrations than the plant due to the change in air flow control at the plant.


0

1

2

3

4

5

6

7

8

9

10

DO

con

cent

ratio

n zo

ne 6

(m

g/l)

Model

Line 1

Line 5

Line 6


(a) (b)

(c) (d)

Figure A.16. Calibration results for Himmerfjärden WWTP 2012. (a) MLVSS, (b) VSS in RAS, (c) NH4 concentration in block A and (d) air flow rate.

(a) (b)

(c) (d)

Figure A.17. Validation results for Himmerfjärden WWTP 2011. (a) MLVSS, (b) VSS in RAS, (c) NH4 concentration in block A and (d) air flow rate.


1000

1500

2000

2500

3000

MLV

SS

(m

g/l)


3000

4000

5000

6000

7000

8000

9000

10000

VS

S R

AS

(m

g/l)



2

4

6

8

10

12

Am

mon

ium

con

cent

ratio

n (m

g/l)


0.75

1

1.25

1.5x 10

5

Air

flow

rat

e (N

m3 /d

)



1500

2000

2500

3000

MLV

SS

(m

g/l)


2000

3000

4000

5000

6000

7000

8000

9000

10000

VS

S R

AS

(m

g/l)



2

4

6

8

10

12

Am

mon

ium

con

cent

ratio

n (m

g/l)


0.6

0.8

1

1.2

1.4

1.6

1.8x 10

5

Air

flow

rat

e (N

m3 /d

)



Summary of results A.5.5A summary of the calibration and validation is found in Table A.5 and Table A.6. The ammonium concentration and air flow rate are within the stopping criteria. The discrepancy between model and plant air flow rate in 2012 is expected to be mainly due to the change in air flow control at the plant.

Table A.5. Summary of calibration results for Himmerfjärden WWTP 2012


MLVSS (mg/l) 1 976 2 058VSS in WAS (mg/l) 4 699 4 722Ammonium (mg/l) 2.4 2.0Air flow rate (Nm3/d) 104 913 106 868

Table A.6. Summary of validation results for Himmerfjärden WWTP 2011.


MLVSS (mg/l) 1 929 2 123VSS in WAS (mg/l) 4 553 4 491Ammonium (mg/l) 2.2 1.9Air flow rate (Nm3/d) 91 705 110 672

Model settings A.6This section summarises the model settings for the three treatment plants. The ASM1 model parameters for the calibrated models are given in Table A.7 and the settling parameters and air flow model settings are presented in Table A.8 and Table A.9, respectively. Settings for the model fractionation and the result of the COD fractionation are summarised in Table A.10 and Table A.11.

Tab

le A

.7. A

SM1

mod

el p

aram

eter

s fo

r H

enri

ksda

l, K

äppa

la a

nd H

imm

erfj

ärde

n W

WT

Ps.

Cal

ibra

ted

para

met

er v

alue

s in

bol

d. T

he a

bbre

via-

tions

are

thos

e pr

esen

ted

by C

orom

inas

et a

l. (2

010)

. Pre

viou

s no

tatio

n re

fers

to th

e no

tatio

n as

pre

sent

ed b

y H

enze

et a

l. (2

000)

.

Des

crip

tion

U

nit

Not

atio

n P

revi

ous

nota

tion

D

efau

lt B

SM

1 (1

5 °C

)

Hen

riks

dal

(15

°C)

Käp

pala

(1

5 °C

)H

imm

erfj

ärde

n (1

5 °C

)M

ax. h

eter

otro

phic

gro

wth

rat

e d-1

µO

HO

,Max

µH

4.0

4.0

4.0

4.0

Hal

f-sa

tura

tion

coe

ffic

ient

het

erot

roph

s gC

OD

/m3

KS

B,O

HO

KS

10

10

10

10

Oxy

gen

half

-sat

urat

ion

cons

tant

aut

otro

phs

gO2/

m3

KO

2,O

HO

KO

H

0.2

0.2

0.2

0.2

Nit

rate

hal

f-sa

tura

tion

con

stan

t den

itri

fyin

g he

tero

trop

hs

gNO

3-N

/m3

KN

Ox,

OH

O

KN

O

0.5

0.5

0.5

0.5

Het

erot

roph

ic d

ecay

rat

e d-1

b O

HO

b H

0.3

0.3

0.3

0.3

Max

. aut

otro

phic

gro

wth

rat

e d-1

µA

NO

,Max

µA

0.5

0.5

0.6

0.35

A

mm

oniu

m h

alf-

satu

ratio

n co

effi

cien

t au

totr

ophs

gN

H3-

N/m

3 K

NH

x,A

NO

KN

H

1.0

1.0

1.0

1.0

Oxy

gen

half

-sat

urat

ion

coef

fici

ent a

utot

roph

s gO

2/m

3 K

O2,

AN

OK

OA

0.4

0.4

0.4

0.7

Aut

otro

phic

dec

ay r

ate

d-1

b AN

Ob A

0.

05

0.05

0.

05

0.05

C

orr.

fac

tor

for

anox

ic g

row

th o

f

hete

rotr

ophs

-

η µO

HO

, Ax

η g

0.8

0.8

0.8

0.8

Am

mon

ific

atio

n ra

te

m3 /g

CO

D,d

q a

k a

0.05

0.

05

0.05

0.

05

Max

imum

spe

cifi

c hy

drol

ysis

rat

e gC

OD

/gce

ll,d

q XC

BS

B,h

ydk h

3.

0 3.

0 4.

0 3.

0 H

alf-

satu

rati

on c

onst

ant f

or h

ydro

lysi

s of

slo

wly

bi

odeg

rada

ble

subs

trat

e gC

OD

/gce

ll

KX

CB

, hyd

K

X

0.03

0.

03

0.1

0.03

Cor

rect

ion

fact

or f

or a

noxi

c hy

drol

ysis

-

η µA

NO

,Ax

η h

0.8

0.8

0.8

0.8

Het

erot

roph

ic y

ield

gc

ell/

gCO

D

YO

HO

YH

0.67

-

- 0.

67

Het

erot

roph

ic y

ield

(ae

robi

c co

ndit

ions

) gc

ell/

gCO

DY

OH

O, O

xY

H. O

x

- 0.

67

0.7

- H

eter

otro

phic

yie

ld (

anox

ic c

ondi

tion

s)

gcel

l/gC

OD

YO

HO

, Ax

YH

. Ax

-

0.54

0.

54

- A

utot

roph

ic y

ield

gc

ell/

gN

YA

NO

YA

0.24

0.

24

0.24

0.

24

Fra

ctio

n of

bio

mas

s yi

eldi

ng p

art.

prod

ucts

-

f XB

iof p

0.

08

0.08

0.

08

0.08

N

con

tent

in b

iom

ass

gN/g

CO

D

i NX

Bio

i XB

0.08

0.

08

0.08

0.

08

N c

onte

nt in

pro

duct

s fr

om b

iom

ass

gN/g

CO

D

i N, X

UE

Bio

i XP

0.06

0.

06

0.06

0.

06


Table A.8. Calibrated settling parameters. Calibrated parameter values in bold.

Description Unit Notation Default Henriksdal Käppala Himmerfjärden

Practical max. sed. velocity m/d v0 250 250 250 250

Theoretical max. sed. velocity

m/d v0’ 474 474 474 474

Sed. parameter, hindered zone m3/g rh 5.76*10-4 8.5*10-4 8.5*10-4 5.76*10-4

Sed. parameter, flocculent zone m3/g rp 2.86*10-3 4.5*10-3 4.5*10-3 2.86*10-3

Non settleable fraction

g/m3 fns 2.28*10-3 2.28*10-3 2.28*10-3 2.28*10-3

Table A.9. Calibrated air flow model parameters.

Parameter Unit Default value Henriksdal Käppala Himmerfjärden

k1 d-1 2.5656 [2.7 2.2 1.1] [4.7 2.0 1.0 0.9] 1.5k2 d-1 0.0432 0.0432 [0.6 0.6 0.6 0.4] 0.4Y 0.82 [0.85 0.5 0.97] [0.75 0.85 1.0 1.0] 0.864

Table A.10. Data used for the influent calculation and COD fractionation. M = measured, A = assumption, C = calibrated, XCB,N = particulate biodegradable organic N, SB,N = soluble biodegradable organic N. * From Gernaey et al. (2011).


TSS (mg/l) M: 133 A: 100 M: 75VSS:TSS M + A: 0.75 M + A: 0.82 A: 0.82Particulate COD:VSS C: 1.92 C: 1.80 C: 1.60SU + XU:total COD C: 0.23 C: 0.2 C: 0.25SU:SU + XU C: 0.35 C: 0.4 C: 0.8Organic N:total N M: 0.23 M: 0.23 A: 0.21XCB,N fraction* A: 0.247 A: 0.247 A: 0.247 SB,N fraction* A: 0.753 A: 0.753 A: 0.753

Table A.11. COD fractionation for the calibration years.

COD Notation Previous notation

Henriksdal (2011)

Käppala (2012)

Himmerfjärden (2012)

Total 293 250 150Undegradable XU+ SU XI+SI 67 50 38Particulate undegradable XU XI 44 30 8Soluble undegradable SU SI 24 20 30Particulate biodegradable XCB XS 118 93 76Soluble biodegradable SB SS 78 82 22Heterotrophic organisms XOHO XBH 29 25 15Fraction of particulate COD 0.65 0.59 0.66

A.7 Discussion 255

Discussion A.7The plant models have been used to compare aeration control strategies in Chapter 7, 9 and 10. Are the models correct enough for this purpose?

Chapter 8 mentions the importance to consider cycles when comparing control strategies to each other at a treatment plant, and that the most im-portant cycle is the annual cycle. In order to be able to compare two annual cycles, both 2011 and 2012 were used when creating the controller maps in Chapter 7. For Himmerfjärden WWTP, the plants fit equally well to both years (Table A.5, Table A.6). At Henriksdal WWTP the model does not simulate the sludge problems during spring and summer 2012 (Figure A.1). The simulated treatment results are similar to a normal period at the plant even though it does not fit the actual plant data since the effluent ammonium was lower in the model than at the plant. At Käppala WWTP the model fits plant data well from august 2011 and onwards (Figure A.10, Figure A.11). Prior to that the nitrate concentrations have a poor fit and the ammonium concentration is not high enough during peaks. Since it is the relative com-parison between different control strategies this is still a valid period to use. Also, during the ammonium peaks the peaks are still high enough in the model for the ammonium controller to push the DO concentration to its up-per limit. This means there ought to be similar DO concentrations in the model as in the plant despite a lower maximum peak, which is partly con-firmed by the comparison between simulated full-scale controllers and simu-lated full-scale controllers in Table 8.4.

During the calibration work it became clear that the sludge concentrations were important to get correct, which is why this is the first task in a calibra-tion project (Rieger et al., 2013). The sludge concentrations were sensitive to the fractionation of COD, especially inert particulate material. The MLSS concentration was difficult to model correctly particularly in the Käppala model but also in the Henriksdal model. The particulate COD fraction was within a reasonable range. One option which was not investigated was if the sludge balance would have been improved if the particulate fraction was increased and the loss of easily biodegradable carbon was compensated for by modelling a reactive settler with hydrolysis.

The suspended solids concentration after the secondary settler was too high with the default settling model parameters in the Henriksdal and Käppa-la models. The settling parameters had to be changed substantially to de-crease the concentrations to more reasonable levels.

The maximum specific growth rate was lowered in the Himmerfjärden model. This is in agreement with the expectations from the plant personnel at Himmerfjärden WWTP, who are aware that they have a low nitrification rate at the plant. The same goes with the increase in nitrifier growth rate at Käp-pala WWTP, where the plant personnel know from nitrification rate experi-ments that they have higher than average nitrification rates.


The plant models have a limited capability to simulate the performance of the aeration system. More sophisticated models of aeration systems are being developed (Alex et al., 2002; Beltrán et al., 2013). However, it has not been within the scope of this research project to develop or implement those mod-els in the benchmark models. Important limitations of the plant models in-clude pressure control and blower control. How problematic this limitation is depends on which plant is considered. At Henriksdal WWTP there are from time to time problems with keeping the pressure in the air mains. This causes the DO concentration to drop since there is not enough air to keep the set-point despite fully open air flow valves. This is problematic to model with the present model set-up, since it does not necessarily occur at a specific air flow rate. This effect has not been modelled in the Henriksdal model, limit-ing the accuracy of the results in terms of actual DO concentrations in the plant. At Himmerfjärden WWTP there is a similar problem where there are frequent limitations of the air flow capacity and also high DO peaks towards the end of the treatment line. At Käppala WWTP there are periods when the blower capacity is not enough to maintain the pressure set-point, and occa-sionally there are problems with the aeration when blowers are switched on and off. Neither of this is modelled in the present model.

Another phenomenon which is not modelled properly is the stray oxygen in the last aerobic zone at Henriksdal and Käppala WWTPs. During periods when one of the aerobic zones is not aerated there is a considerable amount of oxygen in the zone which is transferred from adjacent zones. The stray oxygen is not modelled correctly since the interaction between zones is not captured when modelling CSTRs in series. Another issue is that the plant personnel at Henriksdal WWTP expect that the zones are not completely mixed, with higher DO concentrations at the surface than at the bottom. Since a CSTR has the same concentration throughout the volume this behav-iour is not captured in the plant model. See further Section 7.4.

The settings of the modelled parameters in the aeration model were ad-justed to fit the data for each aerated zone. This is problematic since the set-ting should be fixed for a specific depth, diffuser and diffuser density. It was considered more important to have a good fit to air flow data than to keep the model parameters fixed. If the model parameters were the same in all zones in the Henriksdal and Käppala models the results were not far away from real data but nevertheless a change was needed. The α factor was kept constant in all zones and was set at 0.6 for all the plants. The α factor has been shown to increase along the tank length due to a reduction of surfac-tants along the process (Rosso et al., 2008).

1–11: 1970–197512. Lars Thofelt: Studies on leaf temperature recorded by direct measurement and by thermo-

graphy. 1975.13. Monica Henricsson: Nutritional studies on Chara globularis Thuill., Chara zeylanica Willd.,

and Chara haitensis Turpin. 1976.14. Göran Kloow: Studies on Regenerated Cellulose by the Fluorescence Depolarization Tech-

nique. 1976.15. Carl-Magnus Backman: A High Pressure Study of the Photolytic Decomposition of Azo-

ethane and Propionyl Peroxide. 1976.16. Lennart Källströmer: The significance of biotin and certain monosaccharides for the growth

of Aspergillus niger on rhamnose medium at elevated temperature. 1977.17. Staffan Renlund: Identification of Oxytocin and Vasopressin in the Bovine Adenohypophysis.

1978.18. Bengt Finnström: Effects of pH, Ionic Strength and Light Intensity on the Flash Photolysis of

L-tryptophan. 1978.19. Thomas C. Amu: Diffusion in Dilute Solutions: An Experimental Study with Special Refer-

ence to the Effect of Size and Shape of Solute and Solvent Molecules. 1978.20. Lars Tegnér: A Flash Photolysis Study of the Thermal Cis-Trans Isomerization of Some

Aromatic Schiff Bases in Solution. 1979.21. Stig Tormod: A High-Speed Stopped Flow Laser Light Scattering Apparatus and its Appli-

cation in a Study of Conformational Changes in Bovine Serum Albumin. 1985.22. Björn Varnestig: Coulomb Excitation of Rotational Nuclei. 1987.23. Frans Lettenström: A study of nuclear effects in deep inelastic muon scattering. 1988.24. Göran Ericsson: Production of Heavy Hypernuclei in Antiproton Annihilation. Study of their

decay in the fission channel. 1988.25. Fang Peng: The Geopotential: Modelling Techniques and Physical Implications with Case

Studies in the South and East China Sea and Fennoscandia. 1989.26. Md. Anowar Hossain: Seismic Refraction Studies in the Baltic Shield along the Fennolora

Profile. 1989.27. Lars Erik Svensson: Coulomb Excitation of Vibrational Nuclei. 1989.28. Bengt Carlsson: Digital differentiating filters and model based fault detection. 1989.29. Alexander Edgar Kavka: Coulomb Excitation. Analytical Methods and Experimental Results

on even Selenium Nuclei. 1989.30. Christopher Juhlin: Seismic Attenuation, Shear Wave Anisotropy and Some Aspects of

Fracturing in the Crystalline Rock of the Siljan Ring Area, Central Sweden. 1990.31. Torbjörn Wigren: Recursive Identification Based on the Nonlinear Wiener Model. 1990.32. Kjell Janson: Experimental investigations of the proton and deuteron structure functions.

1991.33. Suzanne W. Harris: Positive Muons in Crystalline and Amorphous Solids. 1991.34. Jan Blomgren: Experimental Studies of Giant Resonances in Medium-Weight Spherical

Nuclei. 1991.35. Jonas Lindgren: Waveform Inversion of Seismic Reflection Data through Local Optimisation

Methods. 1992.36. Liqi Fang: Dynamic Light Scattering from Polymer Gels and Semidilute Solutions. 1992.37. Raymond Munier: Segmentation, Fragmentation and Jostling of the Baltic Shield with Time.

1993.

Acta Universitatis UpsaliensisUppsala Dissertations from the Faculty of Science

Editor: The Dean of the Faculty of Science

Ammonium Feedback Control in Wastewater Treatment Plants.pdf

Documents

Transcript of Ammonium Feedback Control in Wastewater Treatment Plants.pdf