Lab Quality Analysis in Continuous Ambient October ... - Orsat€¦ · Lab Quality Analysis in...

Thank you so much for this opportunity to share our work with your audience. I am here to talk about PAMS AutoGC Monitoring. My company Orsat, has been working in the state of Texas since 1992 on VOC monitoring required by the Clean Air Act for areas which have failed to meet the ozone attainment goals set by the National Ambient Air Quality Standards or NAAQS rules. We have worked with TCEQ since then to configure, deploy, operate and validate data from the TCEQs AutoGC network.

October, 2018

Separation Science eSeminar

Lab Quality Analysis in Continuous Ambient Air Monitoring

1

1. TCEQ has been operating PAMS AutoGC systems since the mid 90’s after completion of the initial Coastal Oxidant Assessment Study in Texas (COAST) study.

2. The map represents the current AutoGC networks in Texas consisting of over 35 PerkinElmer Ozone Precursor systems and

3. the graph below shows the growth of the networks since 1996. Through various networks the TCEQ receives data from more than 35 continuous monitors collecting speciated concentration data on 48 VOCs hourly year round. This data is available on their website hourly.

October, 2018



2

The systems used in Texas are configured from the PerkinElmer Ozone Precursor System along with additional automation supplied by Orsat. Simple overview of the basic AutoGC system. 1. The sample enters the system through the drier which has a counter flow of dry

air to remove the ambient moisture from the sample prior to trapping. 2. The sample is then pulled through the trap at -30ºC by the sample pump. 3. Once the sample is collected the flow is reversed on the trap and it is rapidly

heated into the gas chromatograph which is equipped with the boiling point and PLOT columns where the C2-C12 HCs are then separated.

4. And the Chromatographic data system records the FID signals, identifies an quantitates the detected peaks. While the 48 minute chromatogram is collected and quantitated by the data system the thermal desorber returns to -30ºC and begins to collect the next sample.

October, 2018



3

1. Our sites have a dilution system which allows the dilution of a 1 ppm standard for2. Automatic introduction of daily Calibration Verification Standard as well as an

analytical or system blank.3. Allows manual multi point calibration curves4. And is capable of diluting either from a 100 ppb or 1 ppm standard

October, 2018



4

This is an example of the separation achieved on the PLOT column of a standard containing all 56 PAMS targets. The standard was generated on a carbon basis and diluted to 4 ppbC. Propane is used to generate a carbon response factor which is applied to all components of the PLOT chromatogram including unidentified totals.

October, 2018



5

Plotted on the same scale as the previous, this is the chromatogram from the boiling point column of the remaining targets in the PAMS 56 standard. Targets in this standard again are 4 ppbC thus all targets have nominally the same area. Benzene is used to generated a carbon response factor which is applied to all the components on the boiling point chromatogram and its totals. While this system has good sensitivity, this can be mitigated by the difficulties of determining the contribution of the system to the measurement. . The boiling point column is responsible for the more difficult separation and due to the complex nature of ambient samples is more likely to exhibit potential interferences.

October, 2018



6

Generating a good system blank is a challenge in its own right. Here is the blank generated from the same dilution system used to dilute the 1 ppm PAMS standard to 5 ppb. This humidified blank represents not only the contribution of the zero gas from the dilution system but also any contribution of the sampling system, trap or columns. The large peak on the PLOT column corresponds to iso-butylene which is not uncommonly seen where systems have parts containing buna o-rings of any type. It along with propylene can accumulate as well in the nafion drier which may require regular replacement and/or cleaning.

October, 2018



7

The TCEQ maintains a rigorous Quality Assurance Project Plan for the operation of their AutoGC network which includes daily calibration checks. For these checks a standard containing 14 of the 56 PAMS targets is dynamically diluted and collected daily. These 14 targets span the boiling point range of the analysis and thus are a good “snap-shot” of the system performance. The check sample shown represents 4 ppbC of each target.

October, 2018



8

By contrast this is a typical PLOT column ambient air sample at over 10 times the scale.

October, 2018



9

And the corresponding boiling point chromatogram at the same scale showing the much lower concentrations generally encountered in the higher boiling targets. Note the many integrated peaks at the end of the run.

1. The expanded portions shows the complex separation issues seen routinely in ambient samples which can contain hundreds of components at low levels.

October, 2018



10

The TCEQ program which currently has 37 AutoGCs collecting hourly data year round, has well defined Operations and Validation operating procedures based around this set of Quality Control checks. With well defined acceptance criteria for each type of quality control, data can be handled accordingly and operations are driven by the quality of the data. AutoGC system generate a large quantity of data which is used by data analysts for modeling as well as regulatory evaluation. The data must meet rigorous quality assurance requirements and thus requires significant review and validation.

October, 2018



11

Orsat has developed a cloud-based database for the review of AutoGC data called the Merlin AutoGC Xplorer or MAX. It was specifically developed because other existing systems did not adequately allow the review of QC data, generation of control charts and easy identification of data not meeting data quality objectives. MAX not only allows for tracking quality control data it also allows configuration of networks with differing pass/fail criteria as well as tracking of calibration standards over time.

October, 2018



12

MAX can facilitation PAMS VOC data verification and Validation which is broken into several levels1. Verification includes routine

• operator review for general instrument operation, data completeness and pass/fail criteria for QC samples

• And Technical review by technical staff of QC data, operator logs for deviations and well as flagging

2. Validation is broken into several levels and focuses on review of ambient data for outliers and comparison with historical data where it exists.

MAX shows missing data as well as data collected but not valid due to timing issues and calculates data completeness

October, 2018



13

The Daily QuickLook by amount shows the ppbC amounts for all targets of all samples as well as a bar-chart representation of each sample. This is data for the light gases on the PLOT column. High values for each hour are highlighted.

October, 2018



14

Scrolling down on the QuickLook shows results from both chromatographic columns. This is the C6+ data. Filenames link directly to the information from each ASCII text file which was ingested into the system.

October, 2018



15

Status flags show details of each sample hour loaded such as Validity based on timing – this shows the calculation of the sample time from the data acquisition time from the chromatographic data system. If the sample time is outside the +/- 25% criteria the data will be flagged as invalid.

October, 2018



16

Additional status flags include whether or not the data was reprocessed

October, 2018



17

If the data was re-uploaded

October, 2018



18

And if the filename does not match the actual acquisition time – which can occur if the system sequence has drifted off time due to sampling issues.

October, 2018



19

Below the Quicklook the daily QC sample recoveries are calculated so operators can have immediate pass/fail information. This shows both the daily CVS recoveries based on +/- 25% criteria and weekly second source standard recoveries based on +/-30% Recoveries.

October, 2018



20

If a retention time standard was run and a dilution supplied in the network setup the Quicklook will calculate the recoveries of this standard as well.

October, 2018



21

Daily blanks show any targets which occur in the daily blank and are highlighted based on the network pass/fail criteria. It also shows totals.

October, 2018



22

Unlike some other systems MAX was originally designed specifically to allow the charting of QC data. This is a graph of the nightly check sample over almost 2 months. MAX data supplies roll-over text which shows the recovery, amount, time and date as well as the actual filename so validators or operators can quickly locate the actual datafile for chromatographic review if necessary.

October, 2018



23

This is the control chart for the second source standard which is only run weekly over the same time period.

October, 2018



24

Because MAX can be setup for variations in QC it is capable of allowing for nightly checks with all targets or only a limited number.

October, 2018



25

Blanks can also be charted across time.

October, 2018



26

All charted data can be exported to csv or excel format. This is useful when attempting to generate MDLs using the new Method Update Rule requirements.

October, 2018



27

MAX can also facilitate validation of ambient air using time series.

October, 2018



28

Time series of related species such at the trimethylbenzenes can facilitate location of outliers and the actual datafiles associated with outlier data.

October, 2018



29

By clicking on the data point the actual ASCII file can be opened.

October, 2018



30

Showing the actual data associated with that datapoint.

October, 2018



31

Time series can be used to review expected diurnal patterns for a single site or

October, 2018



32

Or across multiple sites.

October, 2018



33

A unique feature of MAX is the ability to graph the retention times of targets. This is helpful to review closely eluting peaks such as the methylpentanes at the end of the PLOT column. This will show potential misidentifications.

October, 2018



34

Here are retention time plots of 4 closely eluting targets on the boiling point column.

October, 2018



35

MAX can also generate scatter plot relationships between targets.

October, 2018



36

1. MAX cloud services are designed to allow the uploading and review of data for validation prior to generation of AQS data sets.

• It requires a specific text output currently for PE or Agilent data• It allows uploading using zipped files• Allows reloaded data after reprocessing

2. It is an SQL database• Capable of holding years of data• Graphing large amount of data• Automatic association with AQS parameter codes• Allows user defined target lists and pass/fail criteria• Tracks C of A concentrations for recovery calculations

3. It is on a secure server • Has user defined access levels for operators, validators and administrators• As well as user defined site access.

October, 2018



37

The value of a strong Quality Assurance Project Plan for AutoGC networks allows agencies to:

1. Maintain consistent operating procedures and methodsThis will insure that All instruments are operating under similar methods and conditionsAnd that All instrument performance is evaluated with the same quality controls

2. Quality Controls are designed insure consistent data quality; soQC failures trigger actions which will insure against poor quality dataHaving Common QC rules for all instruments will insure similar quality of data for all sites

3. Finally Measurements of Network performance should includeNetwork-wide Audits to verify inter-site performance within the networkAs well as Evaluations of site data over time for intra-site performance.

October, 2018



38

The PAMS AutoGC protocol uses a simple carbon response across all components. This shows the typical system response across all 56 PAMS targets components using a simple carbon response factor generated as an average response factor from a 3-point calibration curve based on propane and benzene response from 0.5 ppbv to 60 ppbv. This graph represents the daily check standard generated by the dilution of a 100 ppbv 56 component standard diluted to 0.5 ppbv was run daily. This graph shows the distribution of the % recovery of all targets in a daily check standard over 2.5 months. Significant deviations include:1. Propylene – values which are high due to common contamination of nafion driers2. Acetylene - poorly adsorbed and often lost in the sampling or analytical system3. Hexane – again values high and larger deviations due to integration errors

associated with the dean’s switch.4. Generally Losses of heavier targets due to adsorption

October, 2018



39

The application of consistent quality control across so many systems is reflected in the population of data. Although there is some inherent bias in the calibration method originally set forth. With fine tuned quality control procedures, these systems can be operated to produce extremely uniform results. 1. This graph shows the distribution of recoveries on 13 targets used in the TCEQ

daily check standard collected over a week across 25 AutoGC sites. The small quartile limits shows that the precision across these 25 sites is good.

2. The second graph shows the same data for the statically diluted second source weekly standard across a 6 week period for the same 25 sites and again the low quartile spread indicates that these systems are generating similar quality data.

October, 2018



40

Network-wide performance audits will reveal the consistency of the data from all sites. Performance Audits are performed by a number of different groups across the AutoGC Networks in Texas.

The TCEQ Performance audit requires two separate canisters due to the distance between sites. Canisters contained all 48 targets diluted to nominally 7 ppbv. These results represent the average blended concentration as well as the average of the pre and post laboratory analysis. With the exception of acetylene the analytical bias was generally less than 20% on pre and post lab analysis with one canister being consistently less than 10%. Box plots show the distribution of 75% of the results and whiskers represent the minimum and maximum values observed. Red shadow represents 30% bias from the theoretical value expected. The bias on average AutoGC results with the exception of acetylene was less than 30%

October, 2018



41

Aecom which operates the North Texas Network performes audits quarterly. This data represents the results of two separate audit test series. Both test canisters contained all targets at nominally 4 ppbv. The bias on the blends were less than 5% and laboratory results represent the averaged values for the two separate canisters which had a RPD of generally less than 30% with exceptions including acetylene, ethylene, 2-methylpentane, n-decane, 1,2,3-trimethylbenzene and n-Undecane. However bias in AutoGC results were generally within 30% with exceptions being acetylene and n-undecane

October, 2018



42

This performance audit of the 7 AutoGCs in the Houston/Galveston Extended Industry Monitoring Network was based on an audit canister concentration of nominally 5 ppbv. This data represents two separate audits whose blends were not as close as with other tests shown, blend error was about 12%. %Bias in laboratory results was generally less than 30% with the exception of acetylene, n-nonane and n-undecane. These AutoGCs are some of the oldest in the network with over 10 years of continuous service. %Bias of AutoGC results was less than 30% with the sole exception of n-undecane.

October, 2018



43

In order to compare all sites regardless of network the results were normalized as percent bias from the theoretical values for each test. This represents bias across 32 AutoGC sites on a single audit test series (5 sites not included in audit program). The graph also shows the plus and minus 30% control limits.

October, 2018



44

Minimum Detection Limits are calculated on each site yearly to assure that all systems are performing well after yearly maintenance has been performed. Detection limits across all networks are maintained by rigorous attention to instrument setup and operation insuring that responses are maintained within specified limits at the time of maintenance. Since the consumption of gas and time required to run 7 duplicate runs at an acceptable low level presents a significant challenge, our procedure involves “spiking” several minutes of a 40 ppbC PAMS standard on to the analytical trap followed by the remainder of the 40 minute sample period with blank humidified air from our dilution system. This can all be done remotely and automatically so operators are not required to blend canisters and spend hours running samples.

This graph shows the results of the MDL method across 34 AutoGC sites showing that all detection limits are below 1 ppbC and most below the 0.4 ppbC TCEQ quality objective. (3 sites had not had MDL’s for the year pending maintenance)

October, 2018



45

Rigorous attention to instrumentation settings and response allows these systems to perform consistently not only across multiple networks but also over the years. This data represents the results of the same MDL method on a single instrument over a 10 year period. The blue squares are the average MDL value at this site with error bars representing the minimum and maximum values seen over that period. The Red diamond represents the average for all sites from the previous data set. Again detection limits are well below 1 ppbC.

October, 2018



46

The importance of maintaining MDL is made more obvious by this representation of the typical distribution of ambient data concentrations across 3 ½ months.

1. This log scale clearly shows that approximately 35% of the ambient data falls below 1 ppbC. Thus it is crucial that instrumentation be maintained and operated with MDL’s in mind. Generation of synthetic concentrations of low levels sufficient to truly measure MDLs is difficult at best given the limitations of canisters and humidification requirements. However some test results suggest that our ability to accurately measure instrument detection limits may be limited by our ability to make sufficient challenge samples from humidified canisters.

October, 2018



47

To summarize, the general requirements for successful PAMS AutoGC operations include1. A good Chromatographic data system capable of identification and quantitation of

complex samples, a robust and simple calibration strategy, an output format for easy review of data and event control for the automation of routine quality control checks.

2. Strong uniform Standard operating procedures for both operations and validation to maintain the operations within the necessary control limits and uniformly flag data which falls outside those limits.

3. A strong set of data quality objectives with well defined control limits and a system for identifying and correcting failures.

October, 2018



48

Orsat has been configuring, operating and helping to develop quality controls to produce robust PAMS AutoGC systems since 1992. TCEQ currently collects data from 37 PerkinElmer PAMS AutoGC systems in Texas which are operated year round. These systems post hourly data to the TCEQ website real-time with percent data recoveries of 90-95% and are operated based on the original PAMS technical assistance document which outlined using the average carbon response factor.

October, 2018



49

I would like to acknowledge the folks who have been instrumental in the various AutoGC networks in Texas1. AECOM previously URS has operated the Houston/Galveston Extended Industry

Monitoring Network for over 10 years and I would like to thank Marty Hale for his performance audit data and Program director Scott Jenkins.

2. The University of Texas Center for Energy and Environmental Research has operated the Corpus Christi Monitoring Network as well as several of the initial systems in the Eagle Ford Shale monitoring activities. We want to thank Dave Sullivan for his continued help and support.

3. And all the Monitoring Groups at the TCEQ from the director Cory Chism down have worked tirelessly to maintain the quality of these activities. Our continued gratitude go to Cindy Maresh and Melanie Hotchkiss who struggle with the ongoing quality issues of such a large network and who have helped to develop a strong QAPP and supporting documents that have allowed us to maintain a robust network.

October, 2018



50

Thank you for your attention and I would be happy to entertain Questions:

Why is the TCEQ network so large? Are these all PAMS sites?No the 37 sites are not all PAMS sites. The TCEQ only has 8 required PAMS sites however, over the years they have used the AutoGC data for their Community Air Toxics program because the data is a good representative “snap-shot” of overall air quality and the communities value the ability to see the hourly data for their communities. The North Texas Network of 13 AutoGCs was setup based on the community demand for monitoring when the Barnett Shale Development began in the relatively populated areas near Fort Worth. This data goes a long way to determining the impact of such activities on local communities over years and because of the time resolution is a better representation than canister data.

What are the biggest sources of data loss?I would have to say the most common source of data loss is when the system stops and has to be restarted. This can happen due to power outages at the site which cause any number of issues depending on the length of time the system is down but most often requires the operator to travel to the site to reset the system. Thunderstorms often are an issue and we advise our operators to check their sites often during periods of thunderstorms. Sometimes we can restart the system remotely. The other source of data loss in Texas anyway is the loss of the air conditioning in the trailers. These systems require trailer temperatures of 75-78 and

October, 2018



51

if the air conditioner fails the system will not recycle and stops.

October, 2018

Separation Science eSeminar 51


Lab Quality Analysis in Continuous Ambient October ... - Orsat€¦ · Lab Quality Analysis in...

Documents

Transcript of Lab Quality Analysis in Continuous Ambient October ... - Orsat€¦ · Lab Quality Analysis in...