31 Reactive Separation of Gallic Acid: Experimentation and ...
Transcript of 31 Reactive Separation of Gallic Acid: Experimentation and ...
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 33
Reactive Separation of Gallic Acid Experimentation and Optimization Using Response Surface Methodology and Artificial Neural Network
K Rewatkar D Z Shende and K L Wasewar
Advanced Separation and Analytical Laboratory Department of Chemical Engineering Visvesvaraya National Institute of Technology (VNIT) Nagpur-440010 Maharashtra INDIA
Gallic acid is a major phenolic pollutant present in the wastewater generated from cork boiling olive mill and pharmaceutical industries Experimental and statistical mod-elling using response surface methodology (RSM) and artificial neural network (ANN) were carried out for reactive separation of gallic acid from aqueous stream using tri-n-butyl phosphate (TBP) in hexanol TBP has a more significant effect on extraction effi-ciency as compared to temperature and pH The optimum conditions of 234 g Lndash1 6565 vv 19 oC and 18 of initial concentration of gallic acid concentration of TBP tem-perature and pH respectively were obtained using RSM Under optimum conditions extraction efficiency of 9945 was obtained for gallic acid The ANN and RSM results were compared with experimental unseen data Error analysis suggested the better per-formance of ANN for extraction efficiency predictions
Key words gallic acid reactive extraction Artificial Neural Network Response Surface Methodology optimization
Introduction
Gallic acid (GA) 345-trihydroxy benzoic acid is considered a major phenolic pollutant pres-ent in the waste streams generated from the cork boiling process1 olive mill2 agro-industries phar-maceuticals and food-processing industries3 The wastewater containing GA is harmful to marine life creates bad odour and results in dark colour forma-tion of water The presence of even a trace amount of this phenolic compound in drinking water im-parts an objectionable taste and odour It is reported that the aromatic ring with three hydroxyl groups and ndashCOOH functional group of GA (Figure 1) are responsible for its toxicity45 The European Com-munity (EC) directive 80779EC prescribes a max-imum tolerance level of phenolic compounds in drinking water at 05 microg Lndash16
Biological processes are generally employed to treat the wastewater because of their process effi-ciency and economic viability The waste streams containing GA are highly acidic and toxic This makes the biological process unsuitable by inhibit-ing the microbial activity17 Hence there has been considerable interest among researchers to focus on the removal of GA from the wastewater due to its
toxicity acidity and unsafe disposal Methods such as oxidation89 colloidal surfactant method1011 pho-tocatalytic degradation12 solvent extraction1314 etc are being employed in the removal of GA Among these the solvent extraction method is found to be the most suitable as it offers higher efficiency15 and GA readily dissolves in most of the organic sol-vents16 Tri-n-butyl phosphate (TBP) an organo-phosphorus compound has very low water solubili-ty (04 g Lndash1 at 25 oC) It is considered an effective extractant for most of the organic acids due to its chemical stability and higher distribution coeffi-cient17 The electronegative atom in phosphoryl group (=PO) of TBP forms the hydrogen bonding with the acid molecule to produce molecular com-plexes in the organic phase in order to improve the extraction efficiency1718
In recent years response surface methodology (RSM) and artificial neural network (ANN) have been used for modelling and optimization of several physical processes19ndash21 Both the models offer huge advantages over the conventionally followed one-factor-at-a-time approach and are considered an ef-fective modelling tool for solving complex nonlin-ear multivariable systems These models develop a functional relationship between the input variable and the output response using the experimental data RSM is a statistical technique generally used
doi 1015255CABEQ2016931Original scientific paper
Received February 9 2016 Accepted March 22 2017
Corresponding author Email k_wasewarrediffmailcom Phone +91-712-2801561 Fax +91-712-2223969
This work is licensed under a Creative Commons Attribution 40
International License
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip33ndash46
34 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
for the design of experiments model development and optimization of process conditions to achieve a target response The ability to optimize a process and interpret the interactive effects of the process variable on the response using a lesser number of experiments is the key attractive feature of the RSM model It requires good prior knowledge or extra preliminary experiments to fix the search criteria and works only for a nonlinear quadratic correla-tion2223 The ANN model is superior over RSM for processing highly nonlinear complex systems and is considered a good technique for both data fitting and prediction ability212425 It does not require a standard experimental design to develop a model and is highly adaptive towards new process condi-tions However ANN needs a large number of data points to construct a significant model26 The devel-opment of the ANN model for the prediction of ex-traction efficiency of gallic acid has not been re-ported in the literature Due to the complex mechanism of reactive extraction of gallic acid it is difficult to analyze the effect of operating parame-ters on the response and hence the mathematical modelling becomes difficult to accomplish The ANN model was developed to predict the removal efficiency of gallic acid from the aqueous solution and the RSM was used to analyse the relative sig-nificance and the interactive effect of the operating parameters and optimization of the process Vari-ous methods like oxidation photocatalytic degrada-tion adsorption Fentonrsquos method electrochemical extraction microbial processes are available for the removal of gallic acid from waste streams but may not be very effective due to various limitations Sol-vent extraction is a clean economical and energy efficient method and the solvents can be reused Hence solvent extraction was selected for the re-moval of gallic acid in the present study
In the present work an attempt has been made to apply the RSM and ANN for optimizing the reac-tive separation of GA using TBP in hexanol The extraction was carried out by varying the concentra-tion of GA in aqueous phase concentration of TBP in organic phase temperature and pH The RSM model comprising a rotatable orthogonal central composite design (RO-CCD) and an ANN model
based on the Levenberg-Marquardt algorithms were developed The model predictions were compared with the experimental efficiency to evaluate the per-formance of both models
Materials and methods
Material
Gallic acid (99 pure) was obtained from Sig-ma-Aldrich Germany Tri-n-butyl phosphate hexa-nol and HCl were obtained from SD Fine-Chem Ltd India with purity of gt98 The HPLC grade water acetonitrile and acetic acid were obtained from Merck India with purity of gt99 All the chemicals were used without purification The pH of aqueous GA solution was adjusted using HCl A stock solution of GA (5 g Lndash1) was prepared using double distilled water and diluted further to obtain the desired concentration An automatic potentio-metric titrator (AT-38C Spectra-lab instruments Pvt Ltd) was used to determine the pH
Experimental method
An equal volume of 20 mL of gallic acid as aqueous phase and TBP with hexanol as organic phase were added in a 150-mL conical flask and shaken at 250 rpm for 2 h using an orbital shaking incubator (REMI Instruments Ltd India) The or-bital shaker was calibrated for solution temperature in the flask using K-type thermocouple (TC-902) Preliminary experiments were performed and it was observed that 2 h was sufficient to reach equilibri-um (refer to supplementary information Figure S1)
In the experimental runs the initial concentra-tion of GA in aqueous phase (CGA0) TBP concentra-tion in organic phase (CTBP) extraction temperature (T) and initial pH of aqueous GA solution (pH0) were varied The experiments were designed with various combinations to study their synergistic in-teractive effects on the extraction efficiency (Table 1) After extraction the equilibrium concentration of GA (CGAaq) in the aqueous phase was determined using HPLC Whereas the equilibrium concentra-tion of organic phase (CGAorg) was determined by material balance as water solubility of TBP is negli-gible The extraction efficiency E () of GA was calculated as
0
() 100GA org
GA
CE
C= sdot (1)
All the runs except centre point were carried out in duplicate to check the reproducibility of ex-traction efficiency E () which were found to be consistent within plusmn1 error
F i g 1 ndash Chemical structure of GA
33
Figures
Figure 1 Chemical structure of GA
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 35
Analytical methods
The equilibrium concentration of GA (CGAaq) in the aqueous phase was determined using HPLC (Agilent 1200 USA) equipped with a quaternary pump diode array detector 20 microL loop and a C18 column (46 mm IDtimes250 mm 5 microm) The mo-bile phase was a mixture of 10 acetonitrile and 90 water and its flow rate was kept constant at 1 mL minndash1 The wavelength of GA was set at 264 nm The average retention time of GA was found to be 392 min A calibration curve for GA
was obtained with the concentration range from 0005 to 03 g Lndash1 (refer to supplementary informa-tion Figure S1) The analysis was repeated twice under identical conditions for each sample and the average value of CGAaq was recorded
Experimental design
Response Surface Methodology (RSM)
RSM is an effective experimental design tech-nique used to determine the main quadratic and in-
Ta b l e 1 ndash Rotatable orthogonal central composite design (RO-CCD) with experimental and predicted extraction efficiency of GA
Run TypeIndependent variables (factors) Response E ()
CGA0
(g Lndash1)x1
CTBP ( vv)
x2 T (oC) x3 pH0 x4 Experimental RSM ANN Remark
1 O1 07 ndash1 30 ndash1 18 ndash1 12 ndash1 9427 9414 9455 Training
2 O2 21 +1 30 ndash1 18 ndash1 12 ndash1 9447 9423 9443 Training
3 O3 07 ndash1 80 +1 18 ndash1 12 ndash1 9937 9896 9937 Training
4 O4 21 +1 80 +1 18 ndash1 12 ndash1 9930 9999 9927 Training
5 O5 07 ndash1 30 ndash1 35 +1 12 ndash1 9204 9228 9202 Training
6 O6 21 +1 30 ndash1 35 +1 12 ndash1 9171 9135 9143 Training
7 O7 07 ndash1 80 +1 35 +1 12 ndash1 9892 9881 9884 Training
8 O8 21 +1 80 +1 35 +1 12 ndash1 9899 9883 9873 Testing
9 O9 07 ndash1 30 ndash1 18 ndash1 26 +1 9153 9188 9218 Training
10 O10 21 +1 30 ndash1 18 ndash1 26 +1 9176 9192 9239 Training
11 O11 07 ndash1 80 +1 18 ndash1 26 +1 9784 9824 9703 Testing
12 O12 21 +1 80 +1 18 ndash1 26 +1 9929 9924 9991 Training
13 O13 07 ndash1 30 ndash1 35 +1 26 +1 9166 9103 9162 Training
14 O14 21 +1 30 ndash1 35 +1 26 +1 8943 9003 9003 Training
15 O15 07 ndash1 80 +1 35 +1 26 +1 9862 9906 9888 Training
16 O16 21 +1 80 +1 35 +1 26 +1 9887 9905 9811 Validation
17 S1 0275 ndashα 55 0 265 0 19 0 9698 9701 9698 Training
18 S2 2525 +α 55 0 265 0 19 0 9747 9706 9726 Training
19 S3 14 0 1483 ndashα 265 0 19 0 8779 8791 8758 Training
20 S4 14 0 9518 +α 265 0 19 0 9955 9904 9928 Training
21 S5 14 0 55 0 128 ndashα 19 0 9829 9790 9826 Training
22 S6 14 0 55 0 402 +α 19 0 9625 9625 9643 Validation
23 S7 14 0 55 0 265 0 0775 ndashα 9765 9804 9772 Training
24 S8 14 0 55 0 265 0 3025 +α 9718 9640 9673 Training
25 C1 14 0 55 0 265 0 19 0 9726 9736 9738 Training
26 C2 14 0 55 0 265 0 19 0 9728 9736 9738 Testing
27 C3 14 0 55 0 265 0 19 0 9722 9736 9738 Training
28 C4 14 0 55 0 265 0 19 0 9717 9736 9738 Training
O orthogonal design points C centre points S star points +ndashα star point value ndash1 low value 0 centre value +1 high value
36 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
teractive effects of operating variables on the re-sponse The rotatable orthogonal central composite design (RO-CCD) was used to develop a quadratic experimental model In the rotatable design the prediction variance of a point depends only on its distance from the centre of the design irrespective of its direction The orthogonality allows all the fac-tors to be estimated independently RO-CCD con-sists of the corner points nc = 2k the star points ns = 2k and the centre points n0
To make CCD both orthogonal and rotatable the centre points n0 were calculated as 4 by substi-tuting k = 4 factors (CGA0 CTBP T pH0) in the Eq (2) below developed by Khuri and Cornell27
1 20 cn (4n ) 4 2k= + minus (2)
The axial distance a was calculated to be 1607 using the expression below
(3)
( )1 421 2 1 2 c
c s 0 cnn n n n 4
a = + + minus
Table 1 shows the corner points nc (Run 1ndash16 O1ndashO16) star points ns (Run 17ndash24 S1ndashS8) and centre points n0 (Run 25ndash28 C1ndashC4) The sum nc + n0 + ns = 28 Hence 28 batch experiments were carried out to construct the RSM model The rela-tionship between independent variables (factors) and the extraction efficiency (response) was estab-lished using a second-order polynomial equation as follows28
(4)
4 4 42
0 1 1 1i i ii i ij i ji i iE x x x xb b b b
= = == + Σ + Σ + Σ Σ
where E is the predicted response (extraction effi-ciency ) and bi bj bij are the regression coeffi-cients representing linear quadratic and interactive effects of factors ie x1 (CGA0) x2 (CTBP) x3 (T) and x4 (pH0) Table 2 shows the independent variables and their levels which were decided based on the trial experiments and literature survey
The regression analysis and process optimiza-tion were performed using the Statistical Software package ldquoDesign Expertrdquo The significance of the regression model was evaluated using the Fischer
distribution (F-value) and null-hypothesis test (p-value) A larger F-value indicates a better fit of model to the experimental extraction efficiency A p-value less than 005 (p lt 005) indicates the de-sign variable of a model contributing less than 5 change in the response Therefore the variable with a larger F-value and p lt 005 was considered signif-icant22
Three-dimensional surface plots were drawn to determine the interactive effect of the process vari-ables on extraction efficiency The numerical opti-mization method was used to identify the combina-tion of variables that jointly optimize a single response E () in the Design-Expert software
Artificial Neural Network (ANN)
ANN is considered a powerful modelling tool to study the process of multivariable complex non-linear systems The single hidden layer is a univer-sal approximation for the multilayer feed forward neural network29 The operation of a single neuron is shown in Figure 2 The input to a neuron is its bias (bj) and the sum of weighted outputs from the previous layer (yindash1) whereas the output depends on the transfer function f(xij) of its input and can be represented as (5)
i-1J ijij i-1k i-1k ijk 1
ij ij
ij ij
( ) for i 1
for i 1 (input
Input
Output
layer)
x w y b
y f xy x
== +
= gt
= =
sum
where xij yij bij are the input output and bias of jth neuron to ith layer respectively and Jindash1 is the to-tal neurons in the (i-1)th layer The ij
i-1kw is the weight between kth neuron of (i-1)th layer and jth neuron of ith layer and f(xij) is a suitable transfer function
The sigmoid transfer function (logsig) for hid-den neurons and the linear transfer function (pure-lin) for output neuron were used as given below30
Ta b l e 2 ndash Levels of independent variables used for rotatable orthogonal central composite design
Independent variables (factors)
Levels of factors
ndashα ndash1 0 +1 +α
CGA0 (g Lndash1) 0275 07 14 21 2525
CTBP ( vv) 1483 30 55 80 9518
T (oC) 1284 18 265 35 4016
pH0 0775 12 19 26 3025 α = 1607
34
Figure 2 Operation of single neuron in ANN
F i g 2 ndash Operation of single neuron in ANN
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 37
(6)
1 ( )1 xlogsig x
eminus=+
( )purelin x x= (7)
Matlab software R2012a was used for ANN modelling The connections between the input layer and the hidden layer and the hidden layer and the output layer consist of weights and biases The weights which express the interactions between the neurons were used in two types of matrix ie input weight matrix (IW) connecting the input layer to the hidden layer and the layer weight matrix (LW) connecting the hidden layer to the output layer Each neuron in the first layer was connected to all the neurons in the next layer with a total of 40 connections between the input and hidden layers The bias determines the threshold to fire a neuron Each neuron has an individual bias value except for the input layer neurons30 The optimal number of neurons in the hidden layer was decided by the trial and error approach by varying it until a minimum mean squared error (MSE) was achieved MSE is an error function which measures the performance of the ANN model and can be calculated as fol-lows21
(8)
i N
ipred iexpi 1
( )MSE
E Esum
where N is the number of experimental data points Eipred and Eiexp are the ANN predicted and experi-mental responses respectively
To compare both constructed models an error analysis was carried out with experimental and the predicted response using coefficient of determina-tion (R2) root mean square error (RMSE) absolute average relative error (AARE) and standard error of prediction (SEP) calculated as follows
(9)
N 2ipred iexp2 i 1
2ipred avgi 1
( )1
( )N
E ER
E E=
=
minus= minus
minussumsum
(10)
N 2ipred iexpi=1
( )RMSE
NE Eminus
= sum
(11)
N ipred iexp
i=1iexp
100AAREN
E E
E
minus= sum
(12) avg
RMSESEP 100E
=
Where Eavg is the average value of the experi-mental response
Results and discussion
The gallic acid is extracted using TBP by form-ing a complex (GATBP) via interfacial reaction (Eq 13) The phosphoryl group of TBP forms a hy-drogen bond with the proton donor gallic acid to form a stable complex in the organic phase This can be represented as
aq org p m orgpGA mTBP (GA TBP )+ (13)
It is evident that the maximum GATBP com-plex formation results in maximum removal effi-ciency of GA that can be obtained only with a par-ticular combination of T pH0 CGA0 and CTBP However a large number of experiments may be required for identifying the suitable process condi-tions Therefore the RSM and ANN models were developed to identify the optimal process conditions required for maximum extraction efficiency of GA
RSM modelling of GA extraction
The design of experiments and experimental extraction efficiency E () are summarized in Table 1 The extraction efficiency for each experiment was calculated using Eq (1) Runs 25ndash28 (C1ndashC4) of centre point were used to determine the experi-mental error and reproducibility of the data The experimental data were regressed to obtain the anal-ysis of variance (ANOVA) as shown in Table 3 The model is highly significant and statistically fit with 9999 confidence level The empirical rela-tionship between the response and the independent variables is expressed in terms of coded variables as follows
(14)
1 2 32 2 2
4 1 2 324 1 2 1 3 1 4
2 3 2 4 3 4
9737 0017 346 051 ndash
051 113 151 011 ndash
0055 024 ndash 025 0011 ndash042 038 024
E x x x
x x x x
x x x x x x xx x x x x x
= + + minus
minus minus minus minus
minus + minus
+ + +
The positive and negative sign of the coeffi-cients indicate the linear increasing or decreasing effect of the corresponding variable on the response respectively The significance of a particular vari-able increases with the sum of square (SS) The ANOVA analysis (Table 3) suggests that CTBP (p = 00001 SS = 25336 F = 88228) has a most signif-icant effect on the response as compared to T (p = 00007 SS = 558 F = 1943) and pH0 (p = 00007 SS = 554 F = 1928) The linear coefficients were more significant than the quadratic or interactive coefficients
The degree of fitness of the model to the exper-imental data was determined by analysing the coef-ficient of determination (R2) The R2 obtained was 098 which indicates that the developed model
38 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
could well explain 98 of total variation in the response and confirms the goodness of fit for the RSM model The R2 (adj) and R2 (pred) values were 097 and 092 respectively which are closer to the R2 value of 0976 suggesting the significance and reasonable prediction ability of the model Ad-equate precision (signal to noise ratio) compares the predicted values with the average prediction error Adequate precision with a ratio greater than 4 is al-ways desirable The value of 3082 indicates an ad-equate signal The coefficient of variation (CV) was calculated to be 056 which indicates better pre-
cision and reliability of the experiments The Stan-dard deviation (S) with 054 indicates strong com-pliance with the predicted response The predicted residual sum of square (PRESS) is a measure of predictive power of the developed model The PRESS value of 2214 suggests that the regression model is significant to predict the response for a new experiment The coded terms of RSM model was then converted into actual variables and repre-sented as follows
(15)
02
0 02 2 2
0
0 0
0 0
0 0
9011 118( ) 029( )
011( ) 257(pH ) 026( )
00024( ) 00015( ) 011(pH )0014( )( ) 0043( )( )0022( )(pH ) 0002( )( )0022( )(pH ) 0041( )(pH )
GA TBP
GA
TBP
GA TBP GA
GA TBP
TBP
E C C
T C
C TC C C TC C TC T
= + + minus
minus minus minus minus
minus minus minus +
+ minus minus
minus + +
+ +
Response surface analysis process optimization and validation
Figures 3a and 3b illustrate the 3D surface plots for the individual and the interactive effects of independent variables on extraction efficiency In Figure 3a the concentration of extractant CTBP (x2) and its interaction with pH (x2 x4) show a positive effect on extraction efficiency (Table 3) whereas pH alone shows a negative effect A curvilinear ef-fect on extraction efficiency was observed due to the positive linear effect and negative quadratic ef-fect of CTBP This is attributed to the presence of the highly polar phosphoryl group (gtP=O) of TBP making it a strong Lewis base resulting in higher GATBP complex formation during the extraction process18 Hence the higher CTBP resulted in higher extraction The concentration of the undissociated acid always increases at lower pH due to the higher proton concentration TBP extracts only undissoci-ated form of acid31 hence extraction efficiency E () increases at lower pH values A similar effect of pH on the increase in efficiency was also observed in the extraction of other carboxylic acids32
Figure 3b illustrates that CTBP (x2) and T (x3) has a positive and negative effect respectively on ex-traction efficiency E () whereas the combined effect (x2 x3) shows a positive effect (Table 3) With an increase in temperature the formation of GATBP complex decreases The complex is formed via intermolecular hydrogen bonding through pro-ton transfer between acid and extractant The pro-cess of hydrogen bond formation is exothermic in nature which shifted the extraction equilibrium in backward direction at higher temperature thus re-ducing extraction efficiency Similar observations have been made by many researchers3334 Therefore it is concluded that lower temperature and pH with higher CTBP favour the extraction efficiency
Ta b l e 3 ndash Analysis of variance (ANOVA) for the experimen-tal results of the rotatable orthogonal central com-posite design
Source Coded terms
Coef- ficient
Sum of Squares F-value p-value Remarks
Model 3033 7544 00001 Significant
Constant 9737 00001
Linear
CGA0 x1 0017 000604 0021 0887
CTBP x2 346 25336 88228 00001 Significant
T x3 ndash051 558 1943 00007 Significant
pH0 x4 ndash051 554 1928 00007 Significant
Square
CGA0 CGA0 x12 ndash013 022 077 0395
CTBP CTBP x22 ndash151 3024 1053 00001 Significant
T T x32 ndash011 017 058 0461
pH0 pH0 x42 ndash0055 0041 014 0712
Interaction
CGA0 CTBP x1 x2 024 092 319 0097
CGA0 T x1 x3 ndash025 103 351 0081
CGA0 pH0 x1 x4 ndash0011 000181 000629 0940
CTBP T x2 x3 042 288 1003 0007 Significant
CTBP pH0 x2 x4 038 236 823 0013 Significant
T pH0 x3 x4 024 096 333 0091
Statistics
Adequate precision 3082
PRESS 2214
S 054
CV () 056
R2 098
R2(adj) 097
R2(pred) 092
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 39
Optimization of process variables CGA0 CTBP T and pH0 was performed using the numerical optimi-zation method to achieve optimal conditions to maximize the extraction The optimum variables were CGA0 = 234 g Lndash1 CTBP = 6565 (vv) T = 184 oC and pH0 = 18 with the predicted optimum extraction of 998 The experiment at optimum conditions was carried out in duplicate to validate the model predictions at the optimum conditions The experimental extraction efficiency obtained at the optimum conditions was found to be 9943 which is in close agreement with the model predict-
35
(a)
(b)
Fig 3 Three dimensional surface plot showing (a) the effect of CTBP and pH0 on the extraction
of GA at the centre level (T = 265 oC CGA0 = 14 g L-1) and (b) the effect of CTBP and T on
the extraction of GA at the centre level (pH0 = 19 CGA0 = 14 g L-1)
F i g 3 ndash Three dimensional surface plot showing (a) the ef-fect of CTBP and pH0 on the extraction of GA at the centre level (T = 265 oC CGA0 = 14 g Lndash1) and (b) the effect of CTBP and T on the extraction of GA at the centre level (pH0 = 19 CGA0 = 14 g Lndash1)
Ta b l e 4 ndash Additional experimental dataset used for the con-struction of ANN model
RunIndependent variables Response E ()
CGA0 (g Lndash1)
CTBP ( vv)
T (oC) pH0Experi- mental ANN Remark
29 05 65 11 196 9895 9868 Training
30 25 30 11 303 9488 9508 Training
31 14 30 11 302 9304 9325 Training
32 18 53 11 213 9728 9700 Training
33 25 75 11 305 9886 9851 Training
34 12 22 11 298 9013 9099 Validation
35 04 82 11 167 9723 9766 Training
36 028 15 11 225 8572 8704 Validation
37 10 30 15 308 9277 9243 Training
38 25 30 15 303 9473 9429 Validation
39 028 95 15 078 9730 9747 Training
40 14 30 16 301 9293 9305 Training
41 25 30 16 302 9449 9402 Training
42 25 30 18 303 9359 9347 Training
43 05 25 20 197 9076 9036 Training
44 14 30 20 302 9274 9212 Training
45 25 30 25 301 9124 9179 Training
46 028 15 25 078 8485 8482 Training
47 25 30 27 303 9263 9150 Validation
48 14 30 27 302 8768 8934 Training
49 04 85 27 139 9808 9796 Training
50 24 72 27 255 9866 9843 Training
51 12 24 27 22 8931 9044 Testing
52 10 62 28 172 9758 9826 Testing
53 15 72 28 165 9852 9831 Testing
54 25 30 28 303 9171 9136 Training
55 16 32 30 246 9224 9210 Training
56 18 78 30 261 9895 9852 Testing
57 18 27 30 261 9039 8930 Training
58 12 65 30 254 9806 9792 Training
59 25 95 30 078 9864 9879 Training
60 14 30 33 302 8798 8766 Training
61 25 30 35 303 9114 9062 Validation
62 25 30 35 3012 9038 9063 Training
63 14 30 37 302 8733 8729 Training
64 24 80 37 222 9859 9857 Training
65 08 60 37 218 9718 9694 Training
66 24 16 37 209 8515 8536 Training
67 16 32 37 178 9104 9102 Training
68 25 30 41 303 9099 9032 Training
69 10 30 41 150 8883 8890 Training
40 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
ed value In Table 1 Runs no 3 4 12 20 show the extraction efficiency above 992 at the cost of 80ndash95 TBP while the optimum condition gives the same efficiency with 65 TBP
ANN modelling of the GA extraction process
In ANN the three layered feed forward error back propagation (FF-EBP) with four neurons in the input layer (input parameters CGA0 CTBP T pH0) ten neurons in the hidden layer (optimum) one neuron in the output layer (extraction efficien-cy) was selected Figure 4 illustrates the ANN ar-chitecture applied for predicting the extraction effi-ciency of GA
A total of 69 experimental data-points (Table 1 and 4) were used to train the ANN model The input and the output data-points were normalized in the range 0 to 1 and 03 to 07 respectively This keeps the back propagation error within the limits and avoids over fitting of the data due to very large and very small values of weights and speeds up training time The data were normalized between the two points n1 and n0 using the following generalized equation (16)
minn 1 0 0
max min
( )(n n ) n( )
x xxx x
minus= minus + minus
where xn is normalized data-point and x xmin xmax are the actual minimum and maximum of inputoutput data The data were split into subsets of training (80 55 data-points) validation (10 7 data-points) and testing (10 7 data-points) Trainlm a training function that updates weights and bias values based on Levenberg-Marquardt backpropagation (LMB) algorithm was used for training the network LMB is an iterative technique which reduces the performance function in each it-eration and makes trainlm the fastest error back-propagation (EBP) algorithm for a moderate sized network35 The externally normalized input values were forwarded from the input layer to the hidden layer and then to the output layer to predict the re-sponse The error MSE was backpropagated from the output to the hidden layer and thereafter to the input layer to modify the weights (IW LW) and bi-ases (b1 b2) Figure 5 describes the general concept of FF-EBP algorithm used in ANN training to pre-dict extraction efficiency During the series of trials for EBP the training algorithm connection weights transfer function hidden layers and hidden neurons were varied and a network topology with an opti-mal network architecture (4101) was selected based on MSE value
The training was successfully terminated after 12 epochs (iterations) as the performance function
36
Fig 4 ANN architecture used for the prediction of GA extraction efficiency F i g 4 ndash ANN architecture used for the prediction of GA extraction efficiency
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 41
F i g 5 ndash Feed forward backpropagation algorithm used in the ANN training for the prediction of GA extraction efficiency
MSE reached a minimum value of 5middot10ndash4 which is lower than the set goal E0 = 1middot10ndash3 (Figure 6) The similar characteristic curve for the test and the vali-dation were observed suggesting no significant over-fitting The estimated MSE values for training (1middot10ndash4) testing (3middot10ndash4) and validation data (5middot10ndash4) were found to be lower than the set goal which is desirable Table 5 gives the optimal weights and bias Wherein IW = HtimesA b1 = Htimes1 LW =OtimesH ma-trix where H is hidden layer neurons (10) A is in-put layer neurons (4) and O is the output layer neu-ron (1) The bias b2 is the single value associated with a single neuron at the output
Figure 7 describes the ANN regression plot for training validation testing and the overall predic-tion set in the form of network output versus exper-imental extraction efficiency It can be observed that the network output values were close to the ex-
F i g 6 ndash MSE for training validation and test dataset com-puted using LMB (with performance of 5middot10ndash4 at 12 epochs and goal of 1middot10ndash3)
38
Fig 6 MSE for training validation and test dataset computed using LMB (with performance of
5middot10-4 at 12 epochs and goal of 1middot10-3)
42 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
perimental extraction efficiency in all the cases The correlation coefficients lsquoRrsquo for training validation and testing were 0995 0984 0989 respectively
whereas the overall prediction set was 0993 which confirms that the ANN model is satisfactory for in-terpolating the experimental data
Ta b l e 5 ndash Optimal values of weights and biases obtained from ANN training using Levenberg-Marquardt backpropagation algorithm
Input weight matrix IW (Weights to Hidden layer from inputs)
54485 74654 14374 2009934801 10038 71242 0667899763 46322 41336 7966216333 44955 0545 2141845359 42326 56151 16171
8043 27753 20607 0620637705 30939 60152 7216681829 118002 62271 3
minus minusminus minusminus minus
minus minus minusminus minus
minus minus 806645814 08091 62607 2537333117 47426 72301 08914
minusminus
Bias vector (Hidden layer) b1 102912 114372 94153 04321 26838 50968 73116 01728 4231 37916 Tminus minus minus minus
Layer weight vector LW (Weight to output layer from hidden layer)
01575 04295 00956 03633 02071 02006 01154 02645 02656 0313minus minus minus minus minus minus
Bias scalar (output layer) b2 06545
F i g 7 ndash ANN model showing the regression plots for training validation testing and overall prediction set
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 43
The ANN model was also validated using sta-tistical analysis ANOVA The degree of freedom due to regression (DFregression) and residuals (DFresidual) was calculated as follows20
regressionDF M 1= minus (17)
residualDF N M= minus (18)
M H(A O 1) O= + + + (19)
where M is the total number of network connec-tions including weights and biases N is the total number of experiments (N = 69) performed to con-struct and validate the ANN model Table 6 shows the ANOVA for the developed ANN model The F-value and the p-value were obtained from ANO-VA for the ANN model as 104 and 00004 respec-tively which confirms the significance of the ANN model Table 7 shows the values of R2 and R2 (adj) as 0986 and 088 respectively indicating the good fit of the ANN model to the experimental extraction efficiency Therefore the ANN model for predicting extraction efficiency E () for the extraction of GA can be written as follows
n n 1 2 (LW (IW ) )y purelin logsig x b b= times times + + (20)
Comparison of ANN and RSM models
The prediction and generalization capabilities of RSM and ANN models were evaluated based on its R2 RMSE AARE and SEP values Table 1 and 4 indicate that both the models predict well the ex-perimental extraction efficiency and their R2 values are closer to 1 Therefore both models can be con-sidered good for predicting the extraction efficiency of GA The generalization capabilities of both the models were compared only with unseen dataset Table 7 shows the unseen data set which were not used for the model development Figure 8 shows the comparative parity plot for RSM and ANN model predictability for the unseen data set and suggests that both the model predictions fit the ex-perimental data very well The comparative values of R2 RMSE AARE and SEP values for both mod-els using unseen data are shown in Table 8 Thus higher R2 and lower RMSE AARE SEP values for ANN indicate that the ANN model is superior over the RSM model for its generalization capability of the GA extraction process
Ta b l e 6 ndash Analysis of variance (ANOVA) analysis for ANN model
Source DFa SSb MSc F-value p-value R2 R2(adj)
Model or Regression 60 11421 1904 1041 00004 0986 088Residual Error 9 1646 183Total 69 11587aDegree of Freedom bSum of Squares cMean Square
Ta b l e 7 ndash Unseen experimental dataset used in the developed RSM and ANN models for the prediction of ex-traction efficiency of GA
RunIndependent variables Response E ()
CGA0 (g Lndash1)
CTBP ( vv)
T(oC) pH0Experi- mental RSM ANN
1 16 15 11 174 9057 9015 91402 25 30 40 30 9005 8836 90373 25 23 37 234 8888 8723 88534 25 37 27 211 9226 9323 92275 25 45 27 185 9525 9531 94786 25 23 37 234 8798 8723 88537 14 30 40 301 8533 8967 87368 18 20 32 221 8745 8805 86639 10 30 24 26 9256 9169 925710 18 36 32 21 9302 9285 938911 10 30 40 15 8883 9118 8908
Ta b l e 8 ndash Comparison of RSM and ANN models for its gen-eralization capability (using unseen data)
Parameters RSM ANN
R2 061 0915RMSE 173 080
AARE 143 066 SEP 191 088
40
Fig 8 Comparison of RSM and ANN model predictions for the unseen data (data not used in
the model development)
F i g 8 ndash Comparison of RSM and ANN model predictions for the unseen data (data not used in the model development)
44 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
Conclusion
The present work suggests that gallic acid can be efficiently removed from aqueous streams using TBP in hexanol The RSM and ANN models were developed to predict the extraction efficiency of GA by varying the parameters temperature concentra-tion of TBP in aqueous phase pH and initial con-centration of gallic acid in aqueous solution The importance of each parameter and its synergistic interactive effects on the extraction of GA was ex-plained with a minimum number of experiments us-ing the RSM model Both models were efficient in the prediction of GA extraction efficiency having close agreement with the experimental extraction efficiency However ANN showed superiority over RSM in terms of accuracy but the marginal errors between the models were very low The optimal conditions to achieve maximum extraction as pre-dicted by the RSM model were CGA0 = 234 g Lndash1 CTBP = 6565 (vv) T = 184 oC and pH0 = 18 at which an experimental extraction efficiency of 995 was observed The models also showed bet-ter performance with the use of unseen data which confirmed their generalization capabilities Thus the present investigation on experiments model-
ling and optimization of the GA extraction process could be of great significance for the design and evaluation of the performance of similar systems
S u p p l e m e n t a r y I n f o r m a t i o n
Supplementary information for confirmation of 2 h shaking time for reactive extraction of gallic acid and retention time of 39 min in HPLC analysis
The trial experiments were conducted for ob-taining an optimum shaking period Equal volumes (20 mL) of aqueous and organic phases were shak-en for different time intervals (5ndash240 min) at 250 rpm Equilibrium was achieved after half an hour (Figure S1) Hence further experiments were con-ducted with 2 h shaking time (sufficient to achieve equilibrium)
Calibration curve and chromatogram for HPLC analysis of gallic acid was obtained by diluting the mother stock of 5 g Lndash1 to prepare different concen-trations (5ndash300 mg Lndash1) of standard sample solu-tions
UV absorption spectra for gallic acid was ob-tained using UV Spectrophotometer Shimadzu UV-1800
F i g S 1 ndash Experimental extraction efficiency (E) of gallic acid versus time for the operating conditions CGA0 = 2 g Lndash1 CTBP = 40 (vv) T = 30 oC pH0 = 3
F i g S 2 ndash Calibration curve for gallic acid obtained us- ing HPLC
44
Fig S4b HPLC Chromatogram of gallic acid at 264 nm and 396 min
F i g S 3 ndash HPLC Chromatogram of gallic acid at 264 nm and 396 min
Ext
ract
ion
effic
ienc
y E
()
Peak
are
a (m
AU
s)
Time (min) Gallic acid concentration (mg Lndash1)
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 45
N o m e n c l a t u r e
AARE ndash absolute average relative errorCGAaq ndash equilibrium concentration of GA in aqueous
phase g Lndash1
CGAorg ndash equilibrium concentration of GA in organic phase g Lndash1
CGA0 ndash initial concentration of GA in aqueous phase g Lndash1
CTBP ndash TBP concentration in organic phase DF ndash degree of freedom in ANOVA analysisE ndash extraction efficiency response F-value ndash Fischer distribution value (ratio of variances)H A O ndash number of neurons in hidden layer input lay-
er and output layer respectivelyIW ndash input weight matrixk ndash number of input variables or factors in
RO-CCDlogsig ndash sigmoidal transfer function for neural net-
workLW ndash layer weight matrixM ndash number of network connections including
weights and biasesMS ndash mean square in ANOVA analysisMSE ndash mean squared errorN ndash total number of experiments to construct
ANN modeln0 ndash centre points in RO-CCDnc ndash corner points or factorial points in RO-CCDns ndash star or axial points in RO-CCDpH0 ndash initial pH of aqueous GA solutionPRESS ndash predicted residual sum of squarespurelin ndash linear transfer function for neural networkp-value ndash null-hypothesis testR ndash coefficient of correlation
R2 ndash coefficient of determinationR2 (adj) ndash adjusted R-squaredR2 (pred) ndash predicted R-squaredRMSE ndash root mean square errorSEP ndash standard error of predictionSS ndash sum of squares in ANOVA analysisT ndash extraction temperature oCw ndash connection weight between layersx1 x2 x3 x4 ndash coded levels of independent vaiables in
RSMxij yij bij ndash network input output and bias of jth neu-
ron to ith layer respectivelya ndash distance of each star point from centre in
RO-CCDbi bj bij ndash regression coefficients representing linear
quadratic and interactive effects in RSM
R e f e r e n c e s
1 Benitez F J Real F J Acero J L Leal A I Garcia C Gallic acid degradation in aqueous solutions by UVH2O2 treatment Fentonrsquos reagent and the photo-Fenton sys-tem J Hazard Mater 126 (2005) 31doi httpsdoiorg101016jjhazmat200504040
2 El-Gohary F A Badawy M I El-Khateeb M A El-Kalliny A S Integrated treatment of olive mill wastewater (OMW) by the combination of fentonrsquos reaction and anaer-obic treatment J Hazard Mater 162 (2009) 1536doi httpsdoiorg101016jjhazmat200806098
3 Herrero M Cifuentes A Ibanez E Sub- and supercriti-cal fluid extraction of functional ingredients from different natural sources plants food-by-products algae and mi-croalgae A Review Food Chem 98 (2006) 136doi httpsdoiorg101016jfoodchem200505058
4 Kougan G B Tabopda T Kuete V Verpoorte R Sim-ple phenols phenolic acids and related esters from the me-dicinal plants of africa in Kuet V (First Ed) Medicinal plant research in Africa Pharmacology and chemistry El-sevier Inc London 2013 225ndash249
45
UV absorption spectra for gallic acid was obtained using UV Spectrophotometer Shimadzu UV-1800
Fig S3 Absorption spectra of gallic acid showing max = 264 nm
--
F i g S 4 ndash Absorption spectra of gallic acid showing lmax = 264 nm
46 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
5 Kapellakis I E Tsagarakis K P Crowther J C Olive oil history production and by-product management Rev Environ Sci Biotechnol 7 (2008) 1doi httpsdoiorg101007s11157-007-9120-9
6 Calace N Nardi E Petronio B Pietroletti M Adsorp-tion of phenols by papermill sludges Environ Pollut 118 (2002) 315doi httpsdoiorg101016S0269-7491(01)00303-7
7 Fki I Allouche N Sayadi S The use of polyphenolic extract purified hydroxytyrosol and 34-dihydroxyphenyl acetic acid from olive mill wastewater for the stabilization of refined oils A potential alternative to synthetic antioxi-dants Food Chem 93 (2005) 197doi httpsdoiorg101016jfoodchem200409014
8 Ollis D F Pelizzetti E Serpone N Photocatalyzed de-struction of water contaminants Environ Sci Technol 25 (1991) 1522doi httpsdoiorg101021es00021a001
9 Masten S J Davies S H The use of ozonation to de-grade organic contaminants in wastewaters Environ Sci Technol 28 (1994) 180Adoi httpsdoiorg101021es00053a001
10 Spigno G Jauregi P Recovery of gallic acid with colloi-dal gas aphrons (CGA) Int J Food Eng 1 (2005) 1doi httpsdoiorg1022021556-37581038
11 Spigno G Dermiki M Pastori C Casanova F Jauregi P Recovery of gallic acid with colloidal gas aphrons gen-erated from a cationic surfactant Sep Purif Technol 71 (2010) 56doi httpsdoiorg101016jseppur200911002
12 Quici N Litter M I Heterogeneous photocatalytic degra-dation of gallic acid under different experimental condi-tions Photochem Photobiol Sci 8 (2009) 975doi httpsdoiorg101039b901904a
13 Singh M Jha A Kumar A Hettiarachchy N Rai A K Sharma D Influence of the solvents on the extraction of major phenolic compounds (punicalagin ellagic acid and gallic acid) and their antioxidant activities in pomegranate aril J Food Sci Technol 51 (2014) 2070doi httpsdoiorg101007s13197-014-1267-0
14 Claudio A F M Ferreira A M Freire C S R Silves-tre A J D Freire M G Coutinho J A P Optimization of the gallic acid extraction using ionic-liquid-based aque-ous two-phase systems Sep Purif Technol 97 (2012) 142doi httpsdoiorg101016jseppur201202036
15 Li Y Wang Y Li Y Dai Y Extraction of glyoxylic acid glycolic acid acrylic acid and benzoic acid with trialkyl-phosphine oxide J Chem Eng Data 48 (2003) 621doi httpsdoiorg101021je020174r
16 Daneshfar A Ghaziaskar H S Homayoun N Solubility of gallic acid in methanol ethanol water and ethyl acetate J Chem Eng Data 53 (2008) 776doi httpsdoiorg101021je700633w
17 Wasewar K L Shende D Z Reactive extraction of caproic acid using tri- n -butyl phosphate in hexanol octa-nol and decanol J Chem Eng Data 56 (2011) 288doi httpsdoiorg101021je100974f
18 Keshav A Chand S Wasewar K L Reactive extraction of acrylic acid using tri- n -butyl phosphate in different di-luents J Chem Eng Data 54 (2009) 1782doi httpsdoiorg101021je800856e
19 Zhong K Wang Q Optimization of ultrasonic extraction of polysaccharides from dried longan pulp using response surface methodology Carbohydr Polym 80 (2010) 19doi httpsdoiorg101016jcarbpol200910066
20 Marchitan N Cojocaru C Mereuta A Duca G Crete-scu I Gonta M Modeling and optimization of tartaric acid reactive extraction from aqueous solutions A compar-ison between response surface methodology and artificial neural network Sep Purif Technol 75 (2010) 273doi httpsdoiorg101016jseppur201008016
21 Messikh N Samar M H Messikh L Neural network analysis of liquidndashliquid extraction of phenol from waste-water using tbp solvent Desalination 208 (2007) 42doi httpsdoiorg101016jdesal200604073
22 Montgomery D Design and analysis of experiments (Fifth Ed) John Wiley amp Sons Inc New Jersey 2001
23 Box G E P Hunter J S H W G Statistics for experi-menters design innovation and discovery (second ed) Vol 48 John Wiley ampSons New Jersey 2006
24 Dornier M Decloux M Trystram G Lebert A Dynam-ic modeling of crossflow microfiltration using neural net-works J Memb Sci 98 (1995) 263doi httpsdoiorg1010160376-7388(94)00195-5
25 Niemi H Bulsari A Palosaari S Simulation of mem-brane separation by neural networks J Memb Sci 102 (1995) 185doi httpsdoiorg1010160376-7388(94)00314-O
26 Bourquin J Schmidli H van Hoogevest P Leuenberger H Pitfalls of artificial neural networks (ANN) modelling technique for data sets containing outlier measurements us-ing a study on mixture properties of a direct compressed dosage form Eur J Pharm Sci 7 (1998) 17doi httpsdoiorg101016S0928-0987(97)10027-6
27 Cornell A I Khuri J A Response surfaces Designs and analyses (second ed) Marcel Dekker Inc New York 1996
28 Bezerra M A Santelli R E Oliveira E P Villar L S Escaleira L A Response surface methodology (RSM) as a tool for optimization in analytical chemistry Talanta 76 (2008) 965doi httpsdoiorg101016jtalanta200805019
29 Hornik K Stinchcombe M White H Multilayer feedfor-ward networks are universal approximators Neural Net-works 2 (1989) 359doi httpsdoiorg1010160893-6080(89)90020-8
30 Hagan M T Demuth H B Beale M H Neural network design PWS Publishing Company Boston 1995
31 Keshav A Wasewar K L Chand S Extraction of propi-onic acid using different extractants (tri-n-butylphosphate tri-n-octylamine and Aliquat 336) Ind Eng Chem Res 47 (2008) 6192doi httpsdoiorg101021ie800006r
32 Yang S T White S A Hsu S T Extraction of carboxyl-ic acids with tertiary and quaternary amines Effect of pH Ind Eng Chem Res 30 (1991) 1335doi httpsdoiorg101021ie00054a040
33 Keshav A Wasewar K L Chand S Extraction of Acryl-ic Propionic and butyric acid using Aliquat 336 in oleyl alcohol Equilibria and effect of temperature Ind Eng Chem Res 48 (2009) 888doi httpsdoiorg101021ie8010337
34 Uslu H Kırbaslar S I Effect of temperature and initial acid concentration on the reactive extraction of carboxylic acids J Chem Eng Data 58 (2013) 1822doi httpsdoiorg101021je4002202
35 Pham D T Sagiroglu S Training multilayered percep-trons for pattern recognition A comparative study of four training algorithms Int J Mach Tools Manuf 41 (2001) 419doi httpsdoiorg101016S0890-6955(00)00073-0
34 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
for the design of experiments model development and optimization of process conditions to achieve a target response The ability to optimize a process and interpret the interactive effects of the process variable on the response using a lesser number of experiments is the key attractive feature of the RSM model It requires good prior knowledge or extra preliminary experiments to fix the search criteria and works only for a nonlinear quadratic correla-tion2223 The ANN model is superior over RSM for processing highly nonlinear complex systems and is considered a good technique for both data fitting and prediction ability212425 It does not require a standard experimental design to develop a model and is highly adaptive towards new process condi-tions However ANN needs a large number of data points to construct a significant model26 The devel-opment of the ANN model for the prediction of ex-traction efficiency of gallic acid has not been re-ported in the literature Due to the complex mechanism of reactive extraction of gallic acid it is difficult to analyze the effect of operating parame-ters on the response and hence the mathematical modelling becomes difficult to accomplish The ANN model was developed to predict the removal efficiency of gallic acid from the aqueous solution and the RSM was used to analyse the relative sig-nificance and the interactive effect of the operating parameters and optimization of the process Vari-ous methods like oxidation photocatalytic degrada-tion adsorption Fentonrsquos method electrochemical extraction microbial processes are available for the removal of gallic acid from waste streams but may not be very effective due to various limitations Sol-vent extraction is a clean economical and energy efficient method and the solvents can be reused Hence solvent extraction was selected for the re-moval of gallic acid in the present study
In the present work an attempt has been made to apply the RSM and ANN for optimizing the reac-tive separation of GA using TBP in hexanol The extraction was carried out by varying the concentra-tion of GA in aqueous phase concentration of TBP in organic phase temperature and pH The RSM model comprising a rotatable orthogonal central composite design (RO-CCD) and an ANN model
based on the Levenberg-Marquardt algorithms were developed The model predictions were compared with the experimental efficiency to evaluate the per-formance of both models
Materials and methods
Material
Gallic acid (99 pure) was obtained from Sig-ma-Aldrich Germany Tri-n-butyl phosphate hexa-nol and HCl were obtained from SD Fine-Chem Ltd India with purity of gt98 The HPLC grade water acetonitrile and acetic acid were obtained from Merck India with purity of gt99 All the chemicals were used without purification The pH of aqueous GA solution was adjusted using HCl A stock solution of GA (5 g Lndash1) was prepared using double distilled water and diluted further to obtain the desired concentration An automatic potentio-metric titrator (AT-38C Spectra-lab instruments Pvt Ltd) was used to determine the pH
Experimental method
An equal volume of 20 mL of gallic acid as aqueous phase and TBP with hexanol as organic phase were added in a 150-mL conical flask and shaken at 250 rpm for 2 h using an orbital shaking incubator (REMI Instruments Ltd India) The or-bital shaker was calibrated for solution temperature in the flask using K-type thermocouple (TC-902) Preliminary experiments were performed and it was observed that 2 h was sufficient to reach equilibri-um (refer to supplementary information Figure S1)
In the experimental runs the initial concentra-tion of GA in aqueous phase (CGA0) TBP concentra-tion in organic phase (CTBP) extraction temperature (T) and initial pH of aqueous GA solution (pH0) were varied The experiments were designed with various combinations to study their synergistic in-teractive effects on the extraction efficiency (Table 1) After extraction the equilibrium concentration of GA (CGAaq) in the aqueous phase was determined using HPLC Whereas the equilibrium concentra-tion of organic phase (CGAorg) was determined by material balance as water solubility of TBP is negli-gible The extraction efficiency E () of GA was calculated as
0
() 100GA org
GA
CE
C= sdot (1)
All the runs except centre point were carried out in duplicate to check the reproducibility of ex-traction efficiency E () which were found to be consistent within plusmn1 error
F i g 1 ndash Chemical structure of GA
33
Figures
Figure 1 Chemical structure of GA
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 35
Analytical methods
The equilibrium concentration of GA (CGAaq) in the aqueous phase was determined using HPLC (Agilent 1200 USA) equipped with a quaternary pump diode array detector 20 microL loop and a C18 column (46 mm IDtimes250 mm 5 microm) The mo-bile phase was a mixture of 10 acetonitrile and 90 water and its flow rate was kept constant at 1 mL minndash1 The wavelength of GA was set at 264 nm The average retention time of GA was found to be 392 min A calibration curve for GA
was obtained with the concentration range from 0005 to 03 g Lndash1 (refer to supplementary informa-tion Figure S1) The analysis was repeated twice under identical conditions for each sample and the average value of CGAaq was recorded
Experimental design
Response Surface Methodology (RSM)
RSM is an effective experimental design tech-nique used to determine the main quadratic and in-
Ta b l e 1 ndash Rotatable orthogonal central composite design (RO-CCD) with experimental and predicted extraction efficiency of GA
Run TypeIndependent variables (factors) Response E ()
CGA0
(g Lndash1)x1
CTBP ( vv)
x2 T (oC) x3 pH0 x4 Experimental RSM ANN Remark
1 O1 07 ndash1 30 ndash1 18 ndash1 12 ndash1 9427 9414 9455 Training
2 O2 21 +1 30 ndash1 18 ndash1 12 ndash1 9447 9423 9443 Training
3 O3 07 ndash1 80 +1 18 ndash1 12 ndash1 9937 9896 9937 Training
4 O4 21 +1 80 +1 18 ndash1 12 ndash1 9930 9999 9927 Training
5 O5 07 ndash1 30 ndash1 35 +1 12 ndash1 9204 9228 9202 Training
6 O6 21 +1 30 ndash1 35 +1 12 ndash1 9171 9135 9143 Training
7 O7 07 ndash1 80 +1 35 +1 12 ndash1 9892 9881 9884 Training
8 O8 21 +1 80 +1 35 +1 12 ndash1 9899 9883 9873 Testing
9 O9 07 ndash1 30 ndash1 18 ndash1 26 +1 9153 9188 9218 Training
10 O10 21 +1 30 ndash1 18 ndash1 26 +1 9176 9192 9239 Training
11 O11 07 ndash1 80 +1 18 ndash1 26 +1 9784 9824 9703 Testing
12 O12 21 +1 80 +1 18 ndash1 26 +1 9929 9924 9991 Training
13 O13 07 ndash1 30 ndash1 35 +1 26 +1 9166 9103 9162 Training
14 O14 21 +1 30 ndash1 35 +1 26 +1 8943 9003 9003 Training
15 O15 07 ndash1 80 +1 35 +1 26 +1 9862 9906 9888 Training
16 O16 21 +1 80 +1 35 +1 26 +1 9887 9905 9811 Validation
17 S1 0275 ndashα 55 0 265 0 19 0 9698 9701 9698 Training
18 S2 2525 +α 55 0 265 0 19 0 9747 9706 9726 Training
19 S3 14 0 1483 ndashα 265 0 19 0 8779 8791 8758 Training
20 S4 14 0 9518 +α 265 0 19 0 9955 9904 9928 Training
21 S5 14 0 55 0 128 ndashα 19 0 9829 9790 9826 Training
22 S6 14 0 55 0 402 +α 19 0 9625 9625 9643 Validation
23 S7 14 0 55 0 265 0 0775 ndashα 9765 9804 9772 Training
24 S8 14 0 55 0 265 0 3025 +α 9718 9640 9673 Training
25 C1 14 0 55 0 265 0 19 0 9726 9736 9738 Training
26 C2 14 0 55 0 265 0 19 0 9728 9736 9738 Testing
27 C3 14 0 55 0 265 0 19 0 9722 9736 9738 Training
28 C4 14 0 55 0 265 0 19 0 9717 9736 9738 Training
O orthogonal design points C centre points S star points +ndashα star point value ndash1 low value 0 centre value +1 high value
36 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
teractive effects of operating variables on the re-sponse The rotatable orthogonal central composite design (RO-CCD) was used to develop a quadratic experimental model In the rotatable design the prediction variance of a point depends only on its distance from the centre of the design irrespective of its direction The orthogonality allows all the fac-tors to be estimated independently RO-CCD con-sists of the corner points nc = 2k the star points ns = 2k and the centre points n0
To make CCD both orthogonal and rotatable the centre points n0 were calculated as 4 by substi-tuting k = 4 factors (CGA0 CTBP T pH0) in the Eq (2) below developed by Khuri and Cornell27
1 20 cn (4n ) 4 2k= + minus (2)
The axial distance a was calculated to be 1607 using the expression below
(3)
( )1 421 2 1 2 c
c s 0 cnn n n n 4
a = + + minus
Table 1 shows the corner points nc (Run 1ndash16 O1ndashO16) star points ns (Run 17ndash24 S1ndashS8) and centre points n0 (Run 25ndash28 C1ndashC4) The sum nc + n0 + ns = 28 Hence 28 batch experiments were carried out to construct the RSM model The rela-tionship between independent variables (factors) and the extraction efficiency (response) was estab-lished using a second-order polynomial equation as follows28
(4)
4 4 42
0 1 1 1i i ii i ij i ji i iE x x x xb b b b
= = == + Σ + Σ + Σ Σ
where E is the predicted response (extraction effi-ciency ) and bi bj bij are the regression coeffi-cients representing linear quadratic and interactive effects of factors ie x1 (CGA0) x2 (CTBP) x3 (T) and x4 (pH0) Table 2 shows the independent variables and their levels which were decided based on the trial experiments and literature survey
The regression analysis and process optimiza-tion were performed using the Statistical Software package ldquoDesign Expertrdquo The significance of the regression model was evaluated using the Fischer
distribution (F-value) and null-hypothesis test (p-value) A larger F-value indicates a better fit of model to the experimental extraction efficiency A p-value less than 005 (p lt 005) indicates the de-sign variable of a model contributing less than 5 change in the response Therefore the variable with a larger F-value and p lt 005 was considered signif-icant22
Three-dimensional surface plots were drawn to determine the interactive effect of the process vari-ables on extraction efficiency The numerical opti-mization method was used to identify the combina-tion of variables that jointly optimize a single response E () in the Design-Expert software
Artificial Neural Network (ANN)
ANN is considered a powerful modelling tool to study the process of multivariable complex non-linear systems The single hidden layer is a univer-sal approximation for the multilayer feed forward neural network29 The operation of a single neuron is shown in Figure 2 The input to a neuron is its bias (bj) and the sum of weighted outputs from the previous layer (yindash1) whereas the output depends on the transfer function f(xij) of its input and can be represented as (5)
i-1J ijij i-1k i-1k ijk 1
ij ij
ij ij
( ) for i 1
for i 1 (input
Input
Output
layer)
x w y b
y f xy x
== +
= gt
= =
sum
where xij yij bij are the input output and bias of jth neuron to ith layer respectively and Jindash1 is the to-tal neurons in the (i-1)th layer The ij
i-1kw is the weight between kth neuron of (i-1)th layer and jth neuron of ith layer and f(xij) is a suitable transfer function
The sigmoid transfer function (logsig) for hid-den neurons and the linear transfer function (pure-lin) for output neuron were used as given below30
Ta b l e 2 ndash Levels of independent variables used for rotatable orthogonal central composite design
Independent variables (factors)
Levels of factors
ndashα ndash1 0 +1 +α
CGA0 (g Lndash1) 0275 07 14 21 2525
CTBP ( vv) 1483 30 55 80 9518
T (oC) 1284 18 265 35 4016
pH0 0775 12 19 26 3025 α = 1607
34
Figure 2 Operation of single neuron in ANN
F i g 2 ndash Operation of single neuron in ANN
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 37
(6)
1 ( )1 xlogsig x
eminus=+
( )purelin x x= (7)
Matlab software R2012a was used for ANN modelling The connections between the input layer and the hidden layer and the hidden layer and the output layer consist of weights and biases The weights which express the interactions between the neurons were used in two types of matrix ie input weight matrix (IW) connecting the input layer to the hidden layer and the layer weight matrix (LW) connecting the hidden layer to the output layer Each neuron in the first layer was connected to all the neurons in the next layer with a total of 40 connections between the input and hidden layers The bias determines the threshold to fire a neuron Each neuron has an individual bias value except for the input layer neurons30 The optimal number of neurons in the hidden layer was decided by the trial and error approach by varying it until a minimum mean squared error (MSE) was achieved MSE is an error function which measures the performance of the ANN model and can be calculated as fol-lows21
(8)
i N
ipred iexpi 1
( )MSE
E Esum
where N is the number of experimental data points Eipred and Eiexp are the ANN predicted and experi-mental responses respectively
To compare both constructed models an error analysis was carried out with experimental and the predicted response using coefficient of determina-tion (R2) root mean square error (RMSE) absolute average relative error (AARE) and standard error of prediction (SEP) calculated as follows
(9)
N 2ipred iexp2 i 1
2ipred avgi 1
( )1
( )N
E ER
E E=
=
minus= minus
minussumsum
(10)
N 2ipred iexpi=1
( )RMSE
NE Eminus
= sum
(11)
N ipred iexp
i=1iexp
100AAREN
E E
E
minus= sum
(12) avg
RMSESEP 100E
=
Where Eavg is the average value of the experi-mental response
Results and discussion
The gallic acid is extracted using TBP by form-ing a complex (GATBP) via interfacial reaction (Eq 13) The phosphoryl group of TBP forms a hy-drogen bond with the proton donor gallic acid to form a stable complex in the organic phase This can be represented as
aq org p m orgpGA mTBP (GA TBP )+ (13)
It is evident that the maximum GATBP com-plex formation results in maximum removal effi-ciency of GA that can be obtained only with a par-ticular combination of T pH0 CGA0 and CTBP However a large number of experiments may be required for identifying the suitable process condi-tions Therefore the RSM and ANN models were developed to identify the optimal process conditions required for maximum extraction efficiency of GA
RSM modelling of GA extraction
The design of experiments and experimental extraction efficiency E () are summarized in Table 1 The extraction efficiency for each experiment was calculated using Eq (1) Runs 25ndash28 (C1ndashC4) of centre point were used to determine the experi-mental error and reproducibility of the data The experimental data were regressed to obtain the anal-ysis of variance (ANOVA) as shown in Table 3 The model is highly significant and statistically fit with 9999 confidence level The empirical rela-tionship between the response and the independent variables is expressed in terms of coded variables as follows
(14)
1 2 32 2 2
4 1 2 324 1 2 1 3 1 4
2 3 2 4 3 4
9737 0017 346 051 ndash
051 113 151 011 ndash
0055 024 ndash 025 0011 ndash042 038 024
E x x x
x x x x
x x x x x x xx x x x x x
= + + minus
minus minus minus minus
minus + minus
+ + +
The positive and negative sign of the coeffi-cients indicate the linear increasing or decreasing effect of the corresponding variable on the response respectively The significance of a particular vari-able increases with the sum of square (SS) The ANOVA analysis (Table 3) suggests that CTBP (p = 00001 SS = 25336 F = 88228) has a most signif-icant effect on the response as compared to T (p = 00007 SS = 558 F = 1943) and pH0 (p = 00007 SS = 554 F = 1928) The linear coefficients were more significant than the quadratic or interactive coefficients
The degree of fitness of the model to the exper-imental data was determined by analysing the coef-ficient of determination (R2) The R2 obtained was 098 which indicates that the developed model
38 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
could well explain 98 of total variation in the response and confirms the goodness of fit for the RSM model The R2 (adj) and R2 (pred) values were 097 and 092 respectively which are closer to the R2 value of 0976 suggesting the significance and reasonable prediction ability of the model Ad-equate precision (signal to noise ratio) compares the predicted values with the average prediction error Adequate precision with a ratio greater than 4 is al-ways desirable The value of 3082 indicates an ad-equate signal The coefficient of variation (CV) was calculated to be 056 which indicates better pre-
cision and reliability of the experiments The Stan-dard deviation (S) with 054 indicates strong com-pliance with the predicted response The predicted residual sum of square (PRESS) is a measure of predictive power of the developed model The PRESS value of 2214 suggests that the regression model is significant to predict the response for a new experiment The coded terms of RSM model was then converted into actual variables and repre-sented as follows
(15)
02
0 02 2 2
0
0 0
0 0
0 0
9011 118( ) 029( )
011( ) 257(pH ) 026( )
00024( ) 00015( ) 011(pH )0014( )( ) 0043( )( )0022( )(pH ) 0002( )( )0022( )(pH ) 0041( )(pH )
GA TBP
GA
TBP
GA TBP GA
GA TBP
TBP
E C C
T C
C TC C C TC C TC T
= + + minus
minus minus minus minus
minus minus minus +
+ minus minus
minus + +
+ +
Response surface analysis process optimization and validation
Figures 3a and 3b illustrate the 3D surface plots for the individual and the interactive effects of independent variables on extraction efficiency In Figure 3a the concentration of extractant CTBP (x2) and its interaction with pH (x2 x4) show a positive effect on extraction efficiency (Table 3) whereas pH alone shows a negative effect A curvilinear ef-fect on extraction efficiency was observed due to the positive linear effect and negative quadratic ef-fect of CTBP This is attributed to the presence of the highly polar phosphoryl group (gtP=O) of TBP making it a strong Lewis base resulting in higher GATBP complex formation during the extraction process18 Hence the higher CTBP resulted in higher extraction The concentration of the undissociated acid always increases at lower pH due to the higher proton concentration TBP extracts only undissoci-ated form of acid31 hence extraction efficiency E () increases at lower pH values A similar effect of pH on the increase in efficiency was also observed in the extraction of other carboxylic acids32
Figure 3b illustrates that CTBP (x2) and T (x3) has a positive and negative effect respectively on ex-traction efficiency E () whereas the combined effect (x2 x3) shows a positive effect (Table 3) With an increase in temperature the formation of GATBP complex decreases The complex is formed via intermolecular hydrogen bonding through pro-ton transfer between acid and extractant The pro-cess of hydrogen bond formation is exothermic in nature which shifted the extraction equilibrium in backward direction at higher temperature thus re-ducing extraction efficiency Similar observations have been made by many researchers3334 Therefore it is concluded that lower temperature and pH with higher CTBP favour the extraction efficiency
Ta b l e 3 ndash Analysis of variance (ANOVA) for the experimen-tal results of the rotatable orthogonal central com-posite design
Source Coded terms
Coef- ficient
Sum of Squares F-value p-value Remarks
Model 3033 7544 00001 Significant
Constant 9737 00001
Linear
CGA0 x1 0017 000604 0021 0887
CTBP x2 346 25336 88228 00001 Significant
T x3 ndash051 558 1943 00007 Significant
pH0 x4 ndash051 554 1928 00007 Significant
Square
CGA0 CGA0 x12 ndash013 022 077 0395
CTBP CTBP x22 ndash151 3024 1053 00001 Significant
T T x32 ndash011 017 058 0461
pH0 pH0 x42 ndash0055 0041 014 0712
Interaction
CGA0 CTBP x1 x2 024 092 319 0097
CGA0 T x1 x3 ndash025 103 351 0081
CGA0 pH0 x1 x4 ndash0011 000181 000629 0940
CTBP T x2 x3 042 288 1003 0007 Significant
CTBP pH0 x2 x4 038 236 823 0013 Significant
T pH0 x3 x4 024 096 333 0091
Statistics
Adequate precision 3082
PRESS 2214
S 054
CV () 056
R2 098
R2(adj) 097
R2(pred) 092
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 39
Optimization of process variables CGA0 CTBP T and pH0 was performed using the numerical optimi-zation method to achieve optimal conditions to maximize the extraction The optimum variables were CGA0 = 234 g Lndash1 CTBP = 6565 (vv) T = 184 oC and pH0 = 18 with the predicted optimum extraction of 998 The experiment at optimum conditions was carried out in duplicate to validate the model predictions at the optimum conditions The experimental extraction efficiency obtained at the optimum conditions was found to be 9943 which is in close agreement with the model predict-
35
(a)
(b)
Fig 3 Three dimensional surface plot showing (a) the effect of CTBP and pH0 on the extraction
of GA at the centre level (T = 265 oC CGA0 = 14 g L-1) and (b) the effect of CTBP and T on
the extraction of GA at the centre level (pH0 = 19 CGA0 = 14 g L-1)
F i g 3 ndash Three dimensional surface plot showing (a) the ef-fect of CTBP and pH0 on the extraction of GA at the centre level (T = 265 oC CGA0 = 14 g Lndash1) and (b) the effect of CTBP and T on the extraction of GA at the centre level (pH0 = 19 CGA0 = 14 g Lndash1)
Ta b l e 4 ndash Additional experimental dataset used for the con-struction of ANN model
RunIndependent variables Response E ()
CGA0 (g Lndash1)
CTBP ( vv)
T (oC) pH0Experi- mental ANN Remark
29 05 65 11 196 9895 9868 Training
30 25 30 11 303 9488 9508 Training
31 14 30 11 302 9304 9325 Training
32 18 53 11 213 9728 9700 Training
33 25 75 11 305 9886 9851 Training
34 12 22 11 298 9013 9099 Validation
35 04 82 11 167 9723 9766 Training
36 028 15 11 225 8572 8704 Validation
37 10 30 15 308 9277 9243 Training
38 25 30 15 303 9473 9429 Validation
39 028 95 15 078 9730 9747 Training
40 14 30 16 301 9293 9305 Training
41 25 30 16 302 9449 9402 Training
42 25 30 18 303 9359 9347 Training
43 05 25 20 197 9076 9036 Training
44 14 30 20 302 9274 9212 Training
45 25 30 25 301 9124 9179 Training
46 028 15 25 078 8485 8482 Training
47 25 30 27 303 9263 9150 Validation
48 14 30 27 302 8768 8934 Training
49 04 85 27 139 9808 9796 Training
50 24 72 27 255 9866 9843 Training
51 12 24 27 22 8931 9044 Testing
52 10 62 28 172 9758 9826 Testing
53 15 72 28 165 9852 9831 Testing
54 25 30 28 303 9171 9136 Training
55 16 32 30 246 9224 9210 Training
56 18 78 30 261 9895 9852 Testing
57 18 27 30 261 9039 8930 Training
58 12 65 30 254 9806 9792 Training
59 25 95 30 078 9864 9879 Training
60 14 30 33 302 8798 8766 Training
61 25 30 35 303 9114 9062 Validation
62 25 30 35 3012 9038 9063 Training
63 14 30 37 302 8733 8729 Training
64 24 80 37 222 9859 9857 Training
65 08 60 37 218 9718 9694 Training
66 24 16 37 209 8515 8536 Training
67 16 32 37 178 9104 9102 Training
68 25 30 41 303 9099 9032 Training
69 10 30 41 150 8883 8890 Training
40 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
ed value In Table 1 Runs no 3 4 12 20 show the extraction efficiency above 992 at the cost of 80ndash95 TBP while the optimum condition gives the same efficiency with 65 TBP
ANN modelling of the GA extraction process
In ANN the three layered feed forward error back propagation (FF-EBP) with four neurons in the input layer (input parameters CGA0 CTBP T pH0) ten neurons in the hidden layer (optimum) one neuron in the output layer (extraction efficien-cy) was selected Figure 4 illustrates the ANN ar-chitecture applied for predicting the extraction effi-ciency of GA
A total of 69 experimental data-points (Table 1 and 4) were used to train the ANN model The input and the output data-points were normalized in the range 0 to 1 and 03 to 07 respectively This keeps the back propagation error within the limits and avoids over fitting of the data due to very large and very small values of weights and speeds up training time The data were normalized between the two points n1 and n0 using the following generalized equation (16)
minn 1 0 0
max min
( )(n n ) n( )
x xxx x
minus= minus + minus
where xn is normalized data-point and x xmin xmax are the actual minimum and maximum of inputoutput data The data were split into subsets of training (80 55 data-points) validation (10 7 data-points) and testing (10 7 data-points) Trainlm a training function that updates weights and bias values based on Levenberg-Marquardt backpropagation (LMB) algorithm was used for training the network LMB is an iterative technique which reduces the performance function in each it-eration and makes trainlm the fastest error back-propagation (EBP) algorithm for a moderate sized network35 The externally normalized input values were forwarded from the input layer to the hidden layer and then to the output layer to predict the re-sponse The error MSE was backpropagated from the output to the hidden layer and thereafter to the input layer to modify the weights (IW LW) and bi-ases (b1 b2) Figure 5 describes the general concept of FF-EBP algorithm used in ANN training to pre-dict extraction efficiency During the series of trials for EBP the training algorithm connection weights transfer function hidden layers and hidden neurons were varied and a network topology with an opti-mal network architecture (4101) was selected based on MSE value
The training was successfully terminated after 12 epochs (iterations) as the performance function
36
Fig 4 ANN architecture used for the prediction of GA extraction efficiency F i g 4 ndash ANN architecture used for the prediction of GA extraction efficiency
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 41
F i g 5 ndash Feed forward backpropagation algorithm used in the ANN training for the prediction of GA extraction efficiency
MSE reached a minimum value of 5middot10ndash4 which is lower than the set goal E0 = 1middot10ndash3 (Figure 6) The similar characteristic curve for the test and the vali-dation were observed suggesting no significant over-fitting The estimated MSE values for training (1middot10ndash4) testing (3middot10ndash4) and validation data (5middot10ndash4) were found to be lower than the set goal which is desirable Table 5 gives the optimal weights and bias Wherein IW = HtimesA b1 = Htimes1 LW =OtimesH ma-trix where H is hidden layer neurons (10) A is in-put layer neurons (4) and O is the output layer neu-ron (1) The bias b2 is the single value associated with a single neuron at the output
Figure 7 describes the ANN regression plot for training validation testing and the overall predic-tion set in the form of network output versus exper-imental extraction efficiency It can be observed that the network output values were close to the ex-
F i g 6 ndash MSE for training validation and test dataset com-puted using LMB (with performance of 5middot10ndash4 at 12 epochs and goal of 1middot10ndash3)
38
Fig 6 MSE for training validation and test dataset computed using LMB (with performance of
5middot10-4 at 12 epochs and goal of 1middot10-3)
42 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
perimental extraction efficiency in all the cases The correlation coefficients lsquoRrsquo for training validation and testing were 0995 0984 0989 respectively
whereas the overall prediction set was 0993 which confirms that the ANN model is satisfactory for in-terpolating the experimental data
Ta b l e 5 ndash Optimal values of weights and biases obtained from ANN training using Levenberg-Marquardt backpropagation algorithm
Input weight matrix IW (Weights to Hidden layer from inputs)
54485 74654 14374 2009934801 10038 71242 0667899763 46322 41336 7966216333 44955 0545 2141845359 42326 56151 16171
8043 27753 20607 0620637705 30939 60152 7216681829 118002 62271 3
minus minusminus minusminus minus
minus minus minusminus minus
minus minus 806645814 08091 62607 2537333117 47426 72301 08914
minusminus
Bias vector (Hidden layer) b1 102912 114372 94153 04321 26838 50968 73116 01728 4231 37916 Tminus minus minus minus
Layer weight vector LW (Weight to output layer from hidden layer)
01575 04295 00956 03633 02071 02006 01154 02645 02656 0313minus minus minus minus minus minus
Bias scalar (output layer) b2 06545
F i g 7 ndash ANN model showing the regression plots for training validation testing and overall prediction set
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 43
The ANN model was also validated using sta-tistical analysis ANOVA The degree of freedom due to regression (DFregression) and residuals (DFresidual) was calculated as follows20
regressionDF M 1= minus (17)
residualDF N M= minus (18)
M H(A O 1) O= + + + (19)
where M is the total number of network connec-tions including weights and biases N is the total number of experiments (N = 69) performed to con-struct and validate the ANN model Table 6 shows the ANOVA for the developed ANN model The F-value and the p-value were obtained from ANO-VA for the ANN model as 104 and 00004 respec-tively which confirms the significance of the ANN model Table 7 shows the values of R2 and R2 (adj) as 0986 and 088 respectively indicating the good fit of the ANN model to the experimental extraction efficiency Therefore the ANN model for predicting extraction efficiency E () for the extraction of GA can be written as follows
n n 1 2 (LW (IW ) )y purelin logsig x b b= times times + + (20)
Comparison of ANN and RSM models
The prediction and generalization capabilities of RSM and ANN models were evaluated based on its R2 RMSE AARE and SEP values Table 1 and 4 indicate that both the models predict well the ex-perimental extraction efficiency and their R2 values are closer to 1 Therefore both models can be con-sidered good for predicting the extraction efficiency of GA The generalization capabilities of both the models were compared only with unseen dataset Table 7 shows the unseen data set which were not used for the model development Figure 8 shows the comparative parity plot for RSM and ANN model predictability for the unseen data set and suggests that both the model predictions fit the ex-perimental data very well The comparative values of R2 RMSE AARE and SEP values for both mod-els using unseen data are shown in Table 8 Thus higher R2 and lower RMSE AARE SEP values for ANN indicate that the ANN model is superior over the RSM model for its generalization capability of the GA extraction process
Ta b l e 6 ndash Analysis of variance (ANOVA) analysis for ANN model
Source DFa SSb MSc F-value p-value R2 R2(adj)
Model or Regression 60 11421 1904 1041 00004 0986 088Residual Error 9 1646 183Total 69 11587aDegree of Freedom bSum of Squares cMean Square
Ta b l e 7 ndash Unseen experimental dataset used in the developed RSM and ANN models for the prediction of ex-traction efficiency of GA
RunIndependent variables Response E ()
CGA0 (g Lndash1)
CTBP ( vv)
T(oC) pH0Experi- mental RSM ANN
1 16 15 11 174 9057 9015 91402 25 30 40 30 9005 8836 90373 25 23 37 234 8888 8723 88534 25 37 27 211 9226 9323 92275 25 45 27 185 9525 9531 94786 25 23 37 234 8798 8723 88537 14 30 40 301 8533 8967 87368 18 20 32 221 8745 8805 86639 10 30 24 26 9256 9169 925710 18 36 32 21 9302 9285 938911 10 30 40 15 8883 9118 8908
Ta b l e 8 ndash Comparison of RSM and ANN models for its gen-eralization capability (using unseen data)
Parameters RSM ANN
R2 061 0915RMSE 173 080
AARE 143 066 SEP 191 088
40
Fig 8 Comparison of RSM and ANN model predictions for the unseen data (data not used in
the model development)
F i g 8 ndash Comparison of RSM and ANN model predictions for the unseen data (data not used in the model development)
44 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
Conclusion
The present work suggests that gallic acid can be efficiently removed from aqueous streams using TBP in hexanol The RSM and ANN models were developed to predict the extraction efficiency of GA by varying the parameters temperature concentra-tion of TBP in aqueous phase pH and initial con-centration of gallic acid in aqueous solution The importance of each parameter and its synergistic interactive effects on the extraction of GA was ex-plained with a minimum number of experiments us-ing the RSM model Both models were efficient in the prediction of GA extraction efficiency having close agreement with the experimental extraction efficiency However ANN showed superiority over RSM in terms of accuracy but the marginal errors between the models were very low The optimal conditions to achieve maximum extraction as pre-dicted by the RSM model were CGA0 = 234 g Lndash1 CTBP = 6565 (vv) T = 184 oC and pH0 = 18 at which an experimental extraction efficiency of 995 was observed The models also showed bet-ter performance with the use of unseen data which confirmed their generalization capabilities Thus the present investigation on experiments model-
ling and optimization of the GA extraction process could be of great significance for the design and evaluation of the performance of similar systems
S u p p l e m e n t a r y I n f o r m a t i o n
Supplementary information for confirmation of 2 h shaking time for reactive extraction of gallic acid and retention time of 39 min in HPLC analysis
The trial experiments were conducted for ob-taining an optimum shaking period Equal volumes (20 mL) of aqueous and organic phases were shak-en for different time intervals (5ndash240 min) at 250 rpm Equilibrium was achieved after half an hour (Figure S1) Hence further experiments were con-ducted with 2 h shaking time (sufficient to achieve equilibrium)
Calibration curve and chromatogram for HPLC analysis of gallic acid was obtained by diluting the mother stock of 5 g Lndash1 to prepare different concen-trations (5ndash300 mg Lndash1) of standard sample solu-tions
UV absorption spectra for gallic acid was ob-tained using UV Spectrophotometer Shimadzu UV-1800
F i g S 1 ndash Experimental extraction efficiency (E) of gallic acid versus time for the operating conditions CGA0 = 2 g Lndash1 CTBP = 40 (vv) T = 30 oC pH0 = 3
F i g S 2 ndash Calibration curve for gallic acid obtained us- ing HPLC
44
Fig S4b HPLC Chromatogram of gallic acid at 264 nm and 396 min
F i g S 3 ndash HPLC Chromatogram of gallic acid at 264 nm and 396 min
Ext
ract
ion
effic
ienc
y E
()
Peak
are
a (m
AU
s)
Time (min) Gallic acid concentration (mg Lndash1)
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 45
N o m e n c l a t u r e
AARE ndash absolute average relative errorCGAaq ndash equilibrium concentration of GA in aqueous
phase g Lndash1
CGAorg ndash equilibrium concentration of GA in organic phase g Lndash1
CGA0 ndash initial concentration of GA in aqueous phase g Lndash1
CTBP ndash TBP concentration in organic phase DF ndash degree of freedom in ANOVA analysisE ndash extraction efficiency response F-value ndash Fischer distribution value (ratio of variances)H A O ndash number of neurons in hidden layer input lay-
er and output layer respectivelyIW ndash input weight matrixk ndash number of input variables or factors in
RO-CCDlogsig ndash sigmoidal transfer function for neural net-
workLW ndash layer weight matrixM ndash number of network connections including
weights and biasesMS ndash mean square in ANOVA analysisMSE ndash mean squared errorN ndash total number of experiments to construct
ANN modeln0 ndash centre points in RO-CCDnc ndash corner points or factorial points in RO-CCDns ndash star or axial points in RO-CCDpH0 ndash initial pH of aqueous GA solutionPRESS ndash predicted residual sum of squarespurelin ndash linear transfer function for neural networkp-value ndash null-hypothesis testR ndash coefficient of correlation
R2 ndash coefficient of determinationR2 (adj) ndash adjusted R-squaredR2 (pred) ndash predicted R-squaredRMSE ndash root mean square errorSEP ndash standard error of predictionSS ndash sum of squares in ANOVA analysisT ndash extraction temperature oCw ndash connection weight between layersx1 x2 x3 x4 ndash coded levels of independent vaiables in
RSMxij yij bij ndash network input output and bias of jth neu-
ron to ith layer respectivelya ndash distance of each star point from centre in
RO-CCDbi bj bij ndash regression coefficients representing linear
quadratic and interactive effects in RSM
R e f e r e n c e s
1 Benitez F J Real F J Acero J L Leal A I Garcia C Gallic acid degradation in aqueous solutions by UVH2O2 treatment Fentonrsquos reagent and the photo-Fenton sys-tem J Hazard Mater 126 (2005) 31doi httpsdoiorg101016jjhazmat200504040
2 El-Gohary F A Badawy M I El-Khateeb M A El-Kalliny A S Integrated treatment of olive mill wastewater (OMW) by the combination of fentonrsquos reaction and anaer-obic treatment J Hazard Mater 162 (2009) 1536doi httpsdoiorg101016jjhazmat200806098
3 Herrero M Cifuentes A Ibanez E Sub- and supercriti-cal fluid extraction of functional ingredients from different natural sources plants food-by-products algae and mi-croalgae A Review Food Chem 98 (2006) 136doi httpsdoiorg101016jfoodchem200505058
4 Kougan G B Tabopda T Kuete V Verpoorte R Sim-ple phenols phenolic acids and related esters from the me-dicinal plants of africa in Kuet V (First Ed) Medicinal plant research in Africa Pharmacology and chemistry El-sevier Inc London 2013 225ndash249
45
UV absorption spectra for gallic acid was obtained using UV Spectrophotometer Shimadzu UV-1800
Fig S3 Absorption spectra of gallic acid showing max = 264 nm
--
F i g S 4 ndash Absorption spectra of gallic acid showing lmax = 264 nm
46 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
5 Kapellakis I E Tsagarakis K P Crowther J C Olive oil history production and by-product management Rev Environ Sci Biotechnol 7 (2008) 1doi httpsdoiorg101007s11157-007-9120-9
6 Calace N Nardi E Petronio B Pietroletti M Adsorp-tion of phenols by papermill sludges Environ Pollut 118 (2002) 315doi httpsdoiorg101016S0269-7491(01)00303-7
7 Fki I Allouche N Sayadi S The use of polyphenolic extract purified hydroxytyrosol and 34-dihydroxyphenyl acetic acid from olive mill wastewater for the stabilization of refined oils A potential alternative to synthetic antioxi-dants Food Chem 93 (2005) 197doi httpsdoiorg101016jfoodchem200409014
8 Ollis D F Pelizzetti E Serpone N Photocatalyzed de-struction of water contaminants Environ Sci Technol 25 (1991) 1522doi httpsdoiorg101021es00021a001
9 Masten S J Davies S H The use of ozonation to de-grade organic contaminants in wastewaters Environ Sci Technol 28 (1994) 180Adoi httpsdoiorg101021es00053a001
10 Spigno G Jauregi P Recovery of gallic acid with colloi-dal gas aphrons (CGA) Int J Food Eng 1 (2005) 1doi httpsdoiorg1022021556-37581038
11 Spigno G Dermiki M Pastori C Casanova F Jauregi P Recovery of gallic acid with colloidal gas aphrons gen-erated from a cationic surfactant Sep Purif Technol 71 (2010) 56doi httpsdoiorg101016jseppur200911002
12 Quici N Litter M I Heterogeneous photocatalytic degra-dation of gallic acid under different experimental condi-tions Photochem Photobiol Sci 8 (2009) 975doi httpsdoiorg101039b901904a
13 Singh M Jha A Kumar A Hettiarachchy N Rai A K Sharma D Influence of the solvents on the extraction of major phenolic compounds (punicalagin ellagic acid and gallic acid) and their antioxidant activities in pomegranate aril J Food Sci Technol 51 (2014) 2070doi httpsdoiorg101007s13197-014-1267-0
14 Claudio A F M Ferreira A M Freire C S R Silves-tre A J D Freire M G Coutinho J A P Optimization of the gallic acid extraction using ionic-liquid-based aque-ous two-phase systems Sep Purif Technol 97 (2012) 142doi httpsdoiorg101016jseppur201202036
15 Li Y Wang Y Li Y Dai Y Extraction of glyoxylic acid glycolic acid acrylic acid and benzoic acid with trialkyl-phosphine oxide J Chem Eng Data 48 (2003) 621doi httpsdoiorg101021je020174r
16 Daneshfar A Ghaziaskar H S Homayoun N Solubility of gallic acid in methanol ethanol water and ethyl acetate J Chem Eng Data 53 (2008) 776doi httpsdoiorg101021je700633w
17 Wasewar K L Shende D Z Reactive extraction of caproic acid using tri- n -butyl phosphate in hexanol octa-nol and decanol J Chem Eng Data 56 (2011) 288doi httpsdoiorg101021je100974f
18 Keshav A Chand S Wasewar K L Reactive extraction of acrylic acid using tri- n -butyl phosphate in different di-luents J Chem Eng Data 54 (2009) 1782doi httpsdoiorg101021je800856e
19 Zhong K Wang Q Optimization of ultrasonic extraction of polysaccharides from dried longan pulp using response surface methodology Carbohydr Polym 80 (2010) 19doi httpsdoiorg101016jcarbpol200910066
20 Marchitan N Cojocaru C Mereuta A Duca G Crete-scu I Gonta M Modeling and optimization of tartaric acid reactive extraction from aqueous solutions A compar-ison between response surface methodology and artificial neural network Sep Purif Technol 75 (2010) 273doi httpsdoiorg101016jseppur201008016
21 Messikh N Samar M H Messikh L Neural network analysis of liquidndashliquid extraction of phenol from waste-water using tbp solvent Desalination 208 (2007) 42doi httpsdoiorg101016jdesal200604073
22 Montgomery D Design and analysis of experiments (Fifth Ed) John Wiley amp Sons Inc New Jersey 2001
23 Box G E P Hunter J S H W G Statistics for experi-menters design innovation and discovery (second ed) Vol 48 John Wiley ampSons New Jersey 2006
24 Dornier M Decloux M Trystram G Lebert A Dynam-ic modeling of crossflow microfiltration using neural net-works J Memb Sci 98 (1995) 263doi httpsdoiorg1010160376-7388(94)00195-5
25 Niemi H Bulsari A Palosaari S Simulation of mem-brane separation by neural networks J Memb Sci 102 (1995) 185doi httpsdoiorg1010160376-7388(94)00314-O
26 Bourquin J Schmidli H van Hoogevest P Leuenberger H Pitfalls of artificial neural networks (ANN) modelling technique for data sets containing outlier measurements us-ing a study on mixture properties of a direct compressed dosage form Eur J Pharm Sci 7 (1998) 17doi httpsdoiorg101016S0928-0987(97)10027-6
27 Cornell A I Khuri J A Response surfaces Designs and analyses (second ed) Marcel Dekker Inc New York 1996
28 Bezerra M A Santelli R E Oliveira E P Villar L S Escaleira L A Response surface methodology (RSM) as a tool for optimization in analytical chemistry Talanta 76 (2008) 965doi httpsdoiorg101016jtalanta200805019
29 Hornik K Stinchcombe M White H Multilayer feedfor-ward networks are universal approximators Neural Net-works 2 (1989) 359doi httpsdoiorg1010160893-6080(89)90020-8
30 Hagan M T Demuth H B Beale M H Neural network design PWS Publishing Company Boston 1995
31 Keshav A Wasewar K L Chand S Extraction of propi-onic acid using different extractants (tri-n-butylphosphate tri-n-octylamine and Aliquat 336) Ind Eng Chem Res 47 (2008) 6192doi httpsdoiorg101021ie800006r
32 Yang S T White S A Hsu S T Extraction of carboxyl-ic acids with tertiary and quaternary amines Effect of pH Ind Eng Chem Res 30 (1991) 1335doi httpsdoiorg101021ie00054a040
33 Keshav A Wasewar K L Chand S Extraction of Acryl-ic Propionic and butyric acid using Aliquat 336 in oleyl alcohol Equilibria and effect of temperature Ind Eng Chem Res 48 (2009) 888doi httpsdoiorg101021ie8010337
34 Uslu H Kırbaslar S I Effect of temperature and initial acid concentration on the reactive extraction of carboxylic acids J Chem Eng Data 58 (2013) 1822doi httpsdoiorg101021je4002202
35 Pham D T Sagiroglu S Training multilayered percep-trons for pattern recognition A comparative study of four training algorithms Int J Mach Tools Manuf 41 (2001) 419doi httpsdoiorg101016S0890-6955(00)00073-0
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 35
Analytical methods
The equilibrium concentration of GA (CGAaq) in the aqueous phase was determined using HPLC (Agilent 1200 USA) equipped with a quaternary pump diode array detector 20 microL loop and a C18 column (46 mm IDtimes250 mm 5 microm) The mo-bile phase was a mixture of 10 acetonitrile and 90 water and its flow rate was kept constant at 1 mL minndash1 The wavelength of GA was set at 264 nm The average retention time of GA was found to be 392 min A calibration curve for GA
was obtained with the concentration range from 0005 to 03 g Lndash1 (refer to supplementary informa-tion Figure S1) The analysis was repeated twice under identical conditions for each sample and the average value of CGAaq was recorded
Experimental design
Response Surface Methodology (RSM)
RSM is an effective experimental design tech-nique used to determine the main quadratic and in-
Ta b l e 1 ndash Rotatable orthogonal central composite design (RO-CCD) with experimental and predicted extraction efficiency of GA
Run TypeIndependent variables (factors) Response E ()
CGA0
(g Lndash1)x1
CTBP ( vv)
x2 T (oC) x3 pH0 x4 Experimental RSM ANN Remark
1 O1 07 ndash1 30 ndash1 18 ndash1 12 ndash1 9427 9414 9455 Training
2 O2 21 +1 30 ndash1 18 ndash1 12 ndash1 9447 9423 9443 Training
3 O3 07 ndash1 80 +1 18 ndash1 12 ndash1 9937 9896 9937 Training
4 O4 21 +1 80 +1 18 ndash1 12 ndash1 9930 9999 9927 Training
5 O5 07 ndash1 30 ndash1 35 +1 12 ndash1 9204 9228 9202 Training
6 O6 21 +1 30 ndash1 35 +1 12 ndash1 9171 9135 9143 Training
7 O7 07 ndash1 80 +1 35 +1 12 ndash1 9892 9881 9884 Training
8 O8 21 +1 80 +1 35 +1 12 ndash1 9899 9883 9873 Testing
9 O9 07 ndash1 30 ndash1 18 ndash1 26 +1 9153 9188 9218 Training
10 O10 21 +1 30 ndash1 18 ndash1 26 +1 9176 9192 9239 Training
11 O11 07 ndash1 80 +1 18 ndash1 26 +1 9784 9824 9703 Testing
12 O12 21 +1 80 +1 18 ndash1 26 +1 9929 9924 9991 Training
13 O13 07 ndash1 30 ndash1 35 +1 26 +1 9166 9103 9162 Training
14 O14 21 +1 30 ndash1 35 +1 26 +1 8943 9003 9003 Training
15 O15 07 ndash1 80 +1 35 +1 26 +1 9862 9906 9888 Training
16 O16 21 +1 80 +1 35 +1 26 +1 9887 9905 9811 Validation
17 S1 0275 ndashα 55 0 265 0 19 0 9698 9701 9698 Training
18 S2 2525 +α 55 0 265 0 19 0 9747 9706 9726 Training
19 S3 14 0 1483 ndashα 265 0 19 0 8779 8791 8758 Training
20 S4 14 0 9518 +α 265 0 19 0 9955 9904 9928 Training
21 S5 14 0 55 0 128 ndashα 19 0 9829 9790 9826 Training
22 S6 14 0 55 0 402 +α 19 0 9625 9625 9643 Validation
23 S7 14 0 55 0 265 0 0775 ndashα 9765 9804 9772 Training
24 S8 14 0 55 0 265 0 3025 +α 9718 9640 9673 Training
25 C1 14 0 55 0 265 0 19 0 9726 9736 9738 Training
26 C2 14 0 55 0 265 0 19 0 9728 9736 9738 Testing
27 C3 14 0 55 0 265 0 19 0 9722 9736 9738 Training
28 C4 14 0 55 0 265 0 19 0 9717 9736 9738 Training
O orthogonal design points C centre points S star points +ndashα star point value ndash1 low value 0 centre value +1 high value
36 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
teractive effects of operating variables on the re-sponse The rotatable orthogonal central composite design (RO-CCD) was used to develop a quadratic experimental model In the rotatable design the prediction variance of a point depends only on its distance from the centre of the design irrespective of its direction The orthogonality allows all the fac-tors to be estimated independently RO-CCD con-sists of the corner points nc = 2k the star points ns = 2k and the centre points n0
To make CCD both orthogonal and rotatable the centre points n0 were calculated as 4 by substi-tuting k = 4 factors (CGA0 CTBP T pH0) in the Eq (2) below developed by Khuri and Cornell27
1 20 cn (4n ) 4 2k= + minus (2)
The axial distance a was calculated to be 1607 using the expression below
(3)
( )1 421 2 1 2 c
c s 0 cnn n n n 4
a = + + minus
Table 1 shows the corner points nc (Run 1ndash16 O1ndashO16) star points ns (Run 17ndash24 S1ndashS8) and centre points n0 (Run 25ndash28 C1ndashC4) The sum nc + n0 + ns = 28 Hence 28 batch experiments were carried out to construct the RSM model The rela-tionship between independent variables (factors) and the extraction efficiency (response) was estab-lished using a second-order polynomial equation as follows28
(4)
4 4 42
0 1 1 1i i ii i ij i ji i iE x x x xb b b b
= = == + Σ + Σ + Σ Σ
where E is the predicted response (extraction effi-ciency ) and bi bj bij are the regression coeffi-cients representing linear quadratic and interactive effects of factors ie x1 (CGA0) x2 (CTBP) x3 (T) and x4 (pH0) Table 2 shows the independent variables and their levels which were decided based on the trial experiments and literature survey
The regression analysis and process optimiza-tion were performed using the Statistical Software package ldquoDesign Expertrdquo The significance of the regression model was evaluated using the Fischer
distribution (F-value) and null-hypothesis test (p-value) A larger F-value indicates a better fit of model to the experimental extraction efficiency A p-value less than 005 (p lt 005) indicates the de-sign variable of a model contributing less than 5 change in the response Therefore the variable with a larger F-value and p lt 005 was considered signif-icant22
Three-dimensional surface plots were drawn to determine the interactive effect of the process vari-ables on extraction efficiency The numerical opti-mization method was used to identify the combina-tion of variables that jointly optimize a single response E () in the Design-Expert software
Artificial Neural Network (ANN)
ANN is considered a powerful modelling tool to study the process of multivariable complex non-linear systems The single hidden layer is a univer-sal approximation for the multilayer feed forward neural network29 The operation of a single neuron is shown in Figure 2 The input to a neuron is its bias (bj) and the sum of weighted outputs from the previous layer (yindash1) whereas the output depends on the transfer function f(xij) of its input and can be represented as (5)
i-1J ijij i-1k i-1k ijk 1
ij ij
ij ij
( ) for i 1
for i 1 (input
Input
Output
layer)
x w y b
y f xy x
== +
= gt
= =
sum
where xij yij bij are the input output and bias of jth neuron to ith layer respectively and Jindash1 is the to-tal neurons in the (i-1)th layer The ij
i-1kw is the weight between kth neuron of (i-1)th layer and jth neuron of ith layer and f(xij) is a suitable transfer function
The sigmoid transfer function (logsig) for hid-den neurons and the linear transfer function (pure-lin) for output neuron were used as given below30
Ta b l e 2 ndash Levels of independent variables used for rotatable orthogonal central composite design
Independent variables (factors)
Levels of factors
ndashα ndash1 0 +1 +α
CGA0 (g Lndash1) 0275 07 14 21 2525
CTBP ( vv) 1483 30 55 80 9518
T (oC) 1284 18 265 35 4016
pH0 0775 12 19 26 3025 α = 1607
34
Figure 2 Operation of single neuron in ANN
F i g 2 ndash Operation of single neuron in ANN
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 37
(6)
1 ( )1 xlogsig x
eminus=+
( )purelin x x= (7)
Matlab software R2012a was used for ANN modelling The connections between the input layer and the hidden layer and the hidden layer and the output layer consist of weights and biases The weights which express the interactions between the neurons were used in two types of matrix ie input weight matrix (IW) connecting the input layer to the hidden layer and the layer weight matrix (LW) connecting the hidden layer to the output layer Each neuron in the first layer was connected to all the neurons in the next layer with a total of 40 connections between the input and hidden layers The bias determines the threshold to fire a neuron Each neuron has an individual bias value except for the input layer neurons30 The optimal number of neurons in the hidden layer was decided by the trial and error approach by varying it until a minimum mean squared error (MSE) was achieved MSE is an error function which measures the performance of the ANN model and can be calculated as fol-lows21
(8)
i N
ipred iexpi 1
( )MSE
E Esum
where N is the number of experimental data points Eipred and Eiexp are the ANN predicted and experi-mental responses respectively
To compare both constructed models an error analysis was carried out with experimental and the predicted response using coefficient of determina-tion (R2) root mean square error (RMSE) absolute average relative error (AARE) and standard error of prediction (SEP) calculated as follows
(9)
N 2ipred iexp2 i 1
2ipred avgi 1
( )1
( )N
E ER
E E=
=
minus= minus
minussumsum
(10)
N 2ipred iexpi=1
( )RMSE
NE Eminus
= sum
(11)
N ipred iexp
i=1iexp
100AAREN
E E
E
minus= sum
(12) avg
RMSESEP 100E
=
Where Eavg is the average value of the experi-mental response
Results and discussion
The gallic acid is extracted using TBP by form-ing a complex (GATBP) via interfacial reaction (Eq 13) The phosphoryl group of TBP forms a hy-drogen bond with the proton donor gallic acid to form a stable complex in the organic phase This can be represented as
aq org p m orgpGA mTBP (GA TBP )+ (13)
It is evident that the maximum GATBP com-plex formation results in maximum removal effi-ciency of GA that can be obtained only with a par-ticular combination of T pH0 CGA0 and CTBP However a large number of experiments may be required for identifying the suitable process condi-tions Therefore the RSM and ANN models were developed to identify the optimal process conditions required for maximum extraction efficiency of GA
RSM modelling of GA extraction
The design of experiments and experimental extraction efficiency E () are summarized in Table 1 The extraction efficiency for each experiment was calculated using Eq (1) Runs 25ndash28 (C1ndashC4) of centre point were used to determine the experi-mental error and reproducibility of the data The experimental data were regressed to obtain the anal-ysis of variance (ANOVA) as shown in Table 3 The model is highly significant and statistically fit with 9999 confidence level The empirical rela-tionship between the response and the independent variables is expressed in terms of coded variables as follows
(14)
1 2 32 2 2
4 1 2 324 1 2 1 3 1 4
2 3 2 4 3 4
9737 0017 346 051 ndash
051 113 151 011 ndash
0055 024 ndash 025 0011 ndash042 038 024
E x x x
x x x x
x x x x x x xx x x x x x
= + + minus
minus minus minus minus
minus + minus
+ + +
The positive and negative sign of the coeffi-cients indicate the linear increasing or decreasing effect of the corresponding variable on the response respectively The significance of a particular vari-able increases with the sum of square (SS) The ANOVA analysis (Table 3) suggests that CTBP (p = 00001 SS = 25336 F = 88228) has a most signif-icant effect on the response as compared to T (p = 00007 SS = 558 F = 1943) and pH0 (p = 00007 SS = 554 F = 1928) The linear coefficients were more significant than the quadratic or interactive coefficients
The degree of fitness of the model to the exper-imental data was determined by analysing the coef-ficient of determination (R2) The R2 obtained was 098 which indicates that the developed model
38 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
could well explain 98 of total variation in the response and confirms the goodness of fit for the RSM model The R2 (adj) and R2 (pred) values were 097 and 092 respectively which are closer to the R2 value of 0976 suggesting the significance and reasonable prediction ability of the model Ad-equate precision (signal to noise ratio) compares the predicted values with the average prediction error Adequate precision with a ratio greater than 4 is al-ways desirable The value of 3082 indicates an ad-equate signal The coefficient of variation (CV) was calculated to be 056 which indicates better pre-
cision and reliability of the experiments The Stan-dard deviation (S) with 054 indicates strong com-pliance with the predicted response The predicted residual sum of square (PRESS) is a measure of predictive power of the developed model The PRESS value of 2214 suggests that the regression model is significant to predict the response for a new experiment The coded terms of RSM model was then converted into actual variables and repre-sented as follows
(15)
02
0 02 2 2
0
0 0
0 0
0 0
9011 118( ) 029( )
011( ) 257(pH ) 026( )
00024( ) 00015( ) 011(pH )0014( )( ) 0043( )( )0022( )(pH ) 0002( )( )0022( )(pH ) 0041( )(pH )
GA TBP
GA
TBP
GA TBP GA
GA TBP
TBP
E C C
T C
C TC C C TC C TC T
= + + minus
minus minus minus minus
minus minus minus +
+ minus minus
minus + +
+ +
Response surface analysis process optimization and validation
Figures 3a and 3b illustrate the 3D surface plots for the individual and the interactive effects of independent variables on extraction efficiency In Figure 3a the concentration of extractant CTBP (x2) and its interaction with pH (x2 x4) show a positive effect on extraction efficiency (Table 3) whereas pH alone shows a negative effect A curvilinear ef-fect on extraction efficiency was observed due to the positive linear effect and negative quadratic ef-fect of CTBP This is attributed to the presence of the highly polar phosphoryl group (gtP=O) of TBP making it a strong Lewis base resulting in higher GATBP complex formation during the extraction process18 Hence the higher CTBP resulted in higher extraction The concentration of the undissociated acid always increases at lower pH due to the higher proton concentration TBP extracts only undissoci-ated form of acid31 hence extraction efficiency E () increases at lower pH values A similar effect of pH on the increase in efficiency was also observed in the extraction of other carboxylic acids32
Figure 3b illustrates that CTBP (x2) and T (x3) has a positive and negative effect respectively on ex-traction efficiency E () whereas the combined effect (x2 x3) shows a positive effect (Table 3) With an increase in temperature the formation of GATBP complex decreases The complex is formed via intermolecular hydrogen bonding through pro-ton transfer between acid and extractant The pro-cess of hydrogen bond formation is exothermic in nature which shifted the extraction equilibrium in backward direction at higher temperature thus re-ducing extraction efficiency Similar observations have been made by many researchers3334 Therefore it is concluded that lower temperature and pH with higher CTBP favour the extraction efficiency
Ta b l e 3 ndash Analysis of variance (ANOVA) for the experimen-tal results of the rotatable orthogonal central com-posite design
Source Coded terms
Coef- ficient
Sum of Squares F-value p-value Remarks
Model 3033 7544 00001 Significant
Constant 9737 00001
Linear
CGA0 x1 0017 000604 0021 0887
CTBP x2 346 25336 88228 00001 Significant
T x3 ndash051 558 1943 00007 Significant
pH0 x4 ndash051 554 1928 00007 Significant
Square
CGA0 CGA0 x12 ndash013 022 077 0395
CTBP CTBP x22 ndash151 3024 1053 00001 Significant
T T x32 ndash011 017 058 0461
pH0 pH0 x42 ndash0055 0041 014 0712
Interaction
CGA0 CTBP x1 x2 024 092 319 0097
CGA0 T x1 x3 ndash025 103 351 0081
CGA0 pH0 x1 x4 ndash0011 000181 000629 0940
CTBP T x2 x3 042 288 1003 0007 Significant
CTBP pH0 x2 x4 038 236 823 0013 Significant
T pH0 x3 x4 024 096 333 0091
Statistics
Adequate precision 3082
PRESS 2214
S 054
CV () 056
R2 098
R2(adj) 097
R2(pred) 092
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 39
Optimization of process variables CGA0 CTBP T and pH0 was performed using the numerical optimi-zation method to achieve optimal conditions to maximize the extraction The optimum variables were CGA0 = 234 g Lndash1 CTBP = 6565 (vv) T = 184 oC and pH0 = 18 with the predicted optimum extraction of 998 The experiment at optimum conditions was carried out in duplicate to validate the model predictions at the optimum conditions The experimental extraction efficiency obtained at the optimum conditions was found to be 9943 which is in close agreement with the model predict-
35
(a)
(b)
Fig 3 Three dimensional surface plot showing (a) the effect of CTBP and pH0 on the extraction
of GA at the centre level (T = 265 oC CGA0 = 14 g L-1) and (b) the effect of CTBP and T on
the extraction of GA at the centre level (pH0 = 19 CGA0 = 14 g L-1)
F i g 3 ndash Three dimensional surface plot showing (a) the ef-fect of CTBP and pH0 on the extraction of GA at the centre level (T = 265 oC CGA0 = 14 g Lndash1) and (b) the effect of CTBP and T on the extraction of GA at the centre level (pH0 = 19 CGA0 = 14 g Lndash1)
Ta b l e 4 ndash Additional experimental dataset used for the con-struction of ANN model
RunIndependent variables Response E ()
CGA0 (g Lndash1)
CTBP ( vv)
T (oC) pH0Experi- mental ANN Remark
29 05 65 11 196 9895 9868 Training
30 25 30 11 303 9488 9508 Training
31 14 30 11 302 9304 9325 Training
32 18 53 11 213 9728 9700 Training
33 25 75 11 305 9886 9851 Training
34 12 22 11 298 9013 9099 Validation
35 04 82 11 167 9723 9766 Training
36 028 15 11 225 8572 8704 Validation
37 10 30 15 308 9277 9243 Training
38 25 30 15 303 9473 9429 Validation
39 028 95 15 078 9730 9747 Training
40 14 30 16 301 9293 9305 Training
41 25 30 16 302 9449 9402 Training
42 25 30 18 303 9359 9347 Training
43 05 25 20 197 9076 9036 Training
44 14 30 20 302 9274 9212 Training
45 25 30 25 301 9124 9179 Training
46 028 15 25 078 8485 8482 Training
47 25 30 27 303 9263 9150 Validation
48 14 30 27 302 8768 8934 Training
49 04 85 27 139 9808 9796 Training
50 24 72 27 255 9866 9843 Training
51 12 24 27 22 8931 9044 Testing
52 10 62 28 172 9758 9826 Testing
53 15 72 28 165 9852 9831 Testing
54 25 30 28 303 9171 9136 Training
55 16 32 30 246 9224 9210 Training
56 18 78 30 261 9895 9852 Testing
57 18 27 30 261 9039 8930 Training
58 12 65 30 254 9806 9792 Training
59 25 95 30 078 9864 9879 Training
60 14 30 33 302 8798 8766 Training
61 25 30 35 303 9114 9062 Validation
62 25 30 35 3012 9038 9063 Training
63 14 30 37 302 8733 8729 Training
64 24 80 37 222 9859 9857 Training
65 08 60 37 218 9718 9694 Training
66 24 16 37 209 8515 8536 Training
67 16 32 37 178 9104 9102 Training
68 25 30 41 303 9099 9032 Training
69 10 30 41 150 8883 8890 Training
40 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
ed value In Table 1 Runs no 3 4 12 20 show the extraction efficiency above 992 at the cost of 80ndash95 TBP while the optimum condition gives the same efficiency with 65 TBP
ANN modelling of the GA extraction process
In ANN the three layered feed forward error back propagation (FF-EBP) with four neurons in the input layer (input parameters CGA0 CTBP T pH0) ten neurons in the hidden layer (optimum) one neuron in the output layer (extraction efficien-cy) was selected Figure 4 illustrates the ANN ar-chitecture applied for predicting the extraction effi-ciency of GA
A total of 69 experimental data-points (Table 1 and 4) were used to train the ANN model The input and the output data-points were normalized in the range 0 to 1 and 03 to 07 respectively This keeps the back propagation error within the limits and avoids over fitting of the data due to very large and very small values of weights and speeds up training time The data were normalized between the two points n1 and n0 using the following generalized equation (16)
minn 1 0 0
max min
( )(n n ) n( )
x xxx x
minus= minus + minus
where xn is normalized data-point and x xmin xmax are the actual minimum and maximum of inputoutput data The data were split into subsets of training (80 55 data-points) validation (10 7 data-points) and testing (10 7 data-points) Trainlm a training function that updates weights and bias values based on Levenberg-Marquardt backpropagation (LMB) algorithm was used for training the network LMB is an iterative technique which reduces the performance function in each it-eration and makes trainlm the fastest error back-propagation (EBP) algorithm for a moderate sized network35 The externally normalized input values were forwarded from the input layer to the hidden layer and then to the output layer to predict the re-sponse The error MSE was backpropagated from the output to the hidden layer and thereafter to the input layer to modify the weights (IW LW) and bi-ases (b1 b2) Figure 5 describes the general concept of FF-EBP algorithm used in ANN training to pre-dict extraction efficiency During the series of trials for EBP the training algorithm connection weights transfer function hidden layers and hidden neurons were varied and a network topology with an opti-mal network architecture (4101) was selected based on MSE value
The training was successfully terminated after 12 epochs (iterations) as the performance function
36
Fig 4 ANN architecture used for the prediction of GA extraction efficiency F i g 4 ndash ANN architecture used for the prediction of GA extraction efficiency
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 41
F i g 5 ndash Feed forward backpropagation algorithm used in the ANN training for the prediction of GA extraction efficiency
MSE reached a minimum value of 5middot10ndash4 which is lower than the set goal E0 = 1middot10ndash3 (Figure 6) The similar characteristic curve for the test and the vali-dation were observed suggesting no significant over-fitting The estimated MSE values for training (1middot10ndash4) testing (3middot10ndash4) and validation data (5middot10ndash4) were found to be lower than the set goal which is desirable Table 5 gives the optimal weights and bias Wherein IW = HtimesA b1 = Htimes1 LW =OtimesH ma-trix where H is hidden layer neurons (10) A is in-put layer neurons (4) and O is the output layer neu-ron (1) The bias b2 is the single value associated with a single neuron at the output
Figure 7 describes the ANN regression plot for training validation testing and the overall predic-tion set in the form of network output versus exper-imental extraction efficiency It can be observed that the network output values were close to the ex-
F i g 6 ndash MSE for training validation and test dataset com-puted using LMB (with performance of 5middot10ndash4 at 12 epochs and goal of 1middot10ndash3)
38
Fig 6 MSE for training validation and test dataset computed using LMB (with performance of
5middot10-4 at 12 epochs and goal of 1middot10-3)
42 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
perimental extraction efficiency in all the cases The correlation coefficients lsquoRrsquo for training validation and testing were 0995 0984 0989 respectively
whereas the overall prediction set was 0993 which confirms that the ANN model is satisfactory for in-terpolating the experimental data
Ta b l e 5 ndash Optimal values of weights and biases obtained from ANN training using Levenberg-Marquardt backpropagation algorithm
Input weight matrix IW (Weights to Hidden layer from inputs)
54485 74654 14374 2009934801 10038 71242 0667899763 46322 41336 7966216333 44955 0545 2141845359 42326 56151 16171
8043 27753 20607 0620637705 30939 60152 7216681829 118002 62271 3
minus minusminus minusminus minus
minus minus minusminus minus
minus minus 806645814 08091 62607 2537333117 47426 72301 08914
minusminus
Bias vector (Hidden layer) b1 102912 114372 94153 04321 26838 50968 73116 01728 4231 37916 Tminus minus minus minus
Layer weight vector LW (Weight to output layer from hidden layer)
01575 04295 00956 03633 02071 02006 01154 02645 02656 0313minus minus minus minus minus minus
Bias scalar (output layer) b2 06545
F i g 7 ndash ANN model showing the regression plots for training validation testing and overall prediction set
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 43
The ANN model was also validated using sta-tistical analysis ANOVA The degree of freedom due to regression (DFregression) and residuals (DFresidual) was calculated as follows20
regressionDF M 1= minus (17)
residualDF N M= minus (18)
M H(A O 1) O= + + + (19)
where M is the total number of network connec-tions including weights and biases N is the total number of experiments (N = 69) performed to con-struct and validate the ANN model Table 6 shows the ANOVA for the developed ANN model The F-value and the p-value were obtained from ANO-VA for the ANN model as 104 and 00004 respec-tively which confirms the significance of the ANN model Table 7 shows the values of R2 and R2 (adj) as 0986 and 088 respectively indicating the good fit of the ANN model to the experimental extraction efficiency Therefore the ANN model for predicting extraction efficiency E () for the extraction of GA can be written as follows
n n 1 2 (LW (IW ) )y purelin logsig x b b= times times + + (20)
Comparison of ANN and RSM models
The prediction and generalization capabilities of RSM and ANN models were evaluated based on its R2 RMSE AARE and SEP values Table 1 and 4 indicate that both the models predict well the ex-perimental extraction efficiency and their R2 values are closer to 1 Therefore both models can be con-sidered good for predicting the extraction efficiency of GA The generalization capabilities of both the models were compared only with unseen dataset Table 7 shows the unseen data set which were not used for the model development Figure 8 shows the comparative parity plot for RSM and ANN model predictability for the unseen data set and suggests that both the model predictions fit the ex-perimental data very well The comparative values of R2 RMSE AARE and SEP values for both mod-els using unseen data are shown in Table 8 Thus higher R2 and lower RMSE AARE SEP values for ANN indicate that the ANN model is superior over the RSM model for its generalization capability of the GA extraction process
Ta b l e 6 ndash Analysis of variance (ANOVA) analysis for ANN model
Source DFa SSb MSc F-value p-value R2 R2(adj)
Model or Regression 60 11421 1904 1041 00004 0986 088Residual Error 9 1646 183Total 69 11587aDegree of Freedom bSum of Squares cMean Square
Ta b l e 7 ndash Unseen experimental dataset used in the developed RSM and ANN models for the prediction of ex-traction efficiency of GA
RunIndependent variables Response E ()
CGA0 (g Lndash1)
CTBP ( vv)
T(oC) pH0Experi- mental RSM ANN
1 16 15 11 174 9057 9015 91402 25 30 40 30 9005 8836 90373 25 23 37 234 8888 8723 88534 25 37 27 211 9226 9323 92275 25 45 27 185 9525 9531 94786 25 23 37 234 8798 8723 88537 14 30 40 301 8533 8967 87368 18 20 32 221 8745 8805 86639 10 30 24 26 9256 9169 925710 18 36 32 21 9302 9285 938911 10 30 40 15 8883 9118 8908
Ta b l e 8 ndash Comparison of RSM and ANN models for its gen-eralization capability (using unseen data)
Parameters RSM ANN
R2 061 0915RMSE 173 080
AARE 143 066 SEP 191 088
40
Fig 8 Comparison of RSM and ANN model predictions for the unseen data (data not used in
the model development)
F i g 8 ndash Comparison of RSM and ANN model predictions for the unseen data (data not used in the model development)
44 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
Conclusion
The present work suggests that gallic acid can be efficiently removed from aqueous streams using TBP in hexanol The RSM and ANN models were developed to predict the extraction efficiency of GA by varying the parameters temperature concentra-tion of TBP in aqueous phase pH and initial con-centration of gallic acid in aqueous solution The importance of each parameter and its synergistic interactive effects on the extraction of GA was ex-plained with a minimum number of experiments us-ing the RSM model Both models were efficient in the prediction of GA extraction efficiency having close agreement with the experimental extraction efficiency However ANN showed superiority over RSM in terms of accuracy but the marginal errors between the models were very low The optimal conditions to achieve maximum extraction as pre-dicted by the RSM model were CGA0 = 234 g Lndash1 CTBP = 6565 (vv) T = 184 oC and pH0 = 18 at which an experimental extraction efficiency of 995 was observed The models also showed bet-ter performance with the use of unseen data which confirmed their generalization capabilities Thus the present investigation on experiments model-
ling and optimization of the GA extraction process could be of great significance for the design and evaluation of the performance of similar systems
S u p p l e m e n t a r y I n f o r m a t i o n
Supplementary information for confirmation of 2 h shaking time for reactive extraction of gallic acid and retention time of 39 min in HPLC analysis
The trial experiments were conducted for ob-taining an optimum shaking period Equal volumes (20 mL) of aqueous and organic phases were shak-en for different time intervals (5ndash240 min) at 250 rpm Equilibrium was achieved after half an hour (Figure S1) Hence further experiments were con-ducted with 2 h shaking time (sufficient to achieve equilibrium)
Calibration curve and chromatogram for HPLC analysis of gallic acid was obtained by diluting the mother stock of 5 g Lndash1 to prepare different concen-trations (5ndash300 mg Lndash1) of standard sample solu-tions
UV absorption spectra for gallic acid was ob-tained using UV Spectrophotometer Shimadzu UV-1800
F i g S 1 ndash Experimental extraction efficiency (E) of gallic acid versus time for the operating conditions CGA0 = 2 g Lndash1 CTBP = 40 (vv) T = 30 oC pH0 = 3
F i g S 2 ndash Calibration curve for gallic acid obtained us- ing HPLC
44
Fig S4b HPLC Chromatogram of gallic acid at 264 nm and 396 min
F i g S 3 ndash HPLC Chromatogram of gallic acid at 264 nm and 396 min
Ext
ract
ion
effic
ienc
y E
()
Peak
are
a (m
AU
s)
Time (min) Gallic acid concentration (mg Lndash1)
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 45
N o m e n c l a t u r e
AARE ndash absolute average relative errorCGAaq ndash equilibrium concentration of GA in aqueous
phase g Lndash1
CGAorg ndash equilibrium concentration of GA in organic phase g Lndash1
CGA0 ndash initial concentration of GA in aqueous phase g Lndash1
CTBP ndash TBP concentration in organic phase DF ndash degree of freedom in ANOVA analysisE ndash extraction efficiency response F-value ndash Fischer distribution value (ratio of variances)H A O ndash number of neurons in hidden layer input lay-
er and output layer respectivelyIW ndash input weight matrixk ndash number of input variables or factors in
RO-CCDlogsig ndash sigmoidal transfer function for neural net-
workLW ndash layer weight matrixM ndash number of network connections including
weights and biasesMS ndash mean square in ANOVA analysisMSE ndash mean squared errorN ndash total number of experiments to construct
ANN modeln0 ndash centre points in RO-CCDnc ndash corner points or factorial points in RO-CCDns ndash star or axial points in RO-CCDpH0 ndash initial pH of aqueous GA solutionPRESS ndash predicted residual sum of squarespurelin ndash linear transfer function for neural networkp-value ndash null-hypothesis testR ndash coefficient of correlation
R2 ndash coefficient of determinationR2 (adj) ndash adjusted R-squaredR2 (pred) ndash predicted R-squaredRMSE ndash root mean square errorSEP ndash standard error of predictionSS ndash sum of squares in ANOVA analysisT ndash extraction temperature oCw ndash connection weight between layersx1 x2 x3 x4 ndash coded levels of independent vaiables in
RSMxij yij bij ndash network input output and bias of jth neu-
ron to ith layer respectivelya ndash distance of each star point from centre in
RO-CCDbi bj bij ndash regression coefficients representing linear
quadratic and interactive effects in RSM
R e f e r e n c e s
1 Benitez F J Real F J Acero J L Leal A I Garcia C Gallic acid degradation in aqueous solutions by UVH2O2 treatment Fentonrsquos reagent and the photo-Fenton sys-tem J Hazard Mater 126 (2005) 31doi httpsdoiorg101016jjhazmat200504040
2 El-Gohary F A Badawy M I El-Khateeb M A El-Kalliny A S Integrated treatment of olive mill wastewater (OMW) by the combination of fentonrsquos reaction and anaer-obic treatment J Hazard Mater 162 (2009) 1536doi httpsdoiorg101016jjhazmat200806098
3 Herrero M Cifuentes A Ibanez E Sub- and supercriti-cal fluid extraction of functional ingredients from different natural sources plants food-by-products algae and mi-croalgae A Review Food Chem 98 (2006) 136doi httpsdoiorg101016jfoodchem200505058
4 Kougan G B Tabopda T Kuete V Verpoorte R Sim-ple phenols phenolic acids and related esters from the me-dicinal plants of africa in Kuet V (First Ed) Medicinal plant research in Africa Pharmacology and chemistry El-sevier Inc London 2013 225ndash249
45
UV absorption spectra for gallic acid was obtained using UV Spectrophotometer Shimadzu UV-1800
Fig S3 Absorption spectra of gallic acid showing max = 264 nm
--
F i g S 4 ndash Absorption spectra of gallic acid showing lmax = 264 nm
46 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
5 Kapellakis I E Tsagarakis K P Crowther J C Olive oil history production and by-product management Rev Environ Sci Biotechnol 7 (2008) 1doi httpsdoiorg101007s11157-007-9120-9
6 Calace N Nardi E Petronio B Pietroletti M Adsorp-tion of phenols by papermill sludges Environ Pollut 118 (2002) 315doi httpsdoiorg101016S0269-7491(01)00303-7
7 Fki I Allouche N Sayadi S The use of polyphenolic extract purified hydroxytyrosol and 34-dihydroxyphenyl acetic acid from olive mill wastewater for the stabilization of refined oils A potential alternative to synthetic antioxi-dants Food Chem 93 (2005) 197doi httpsdoiorg101016jfoodchem200409014
8 Ollis D F Pelizzetti E Serpone N Photocatalyzed de-struction of water contaminants Environ Sci Technol 25 (1991) 1522doi httpsdoiorg101021es00021a001
9 Masten S J Davies S H The use of ozonation to de-grade organic contaminants in wastewaters Environ Sci Technol 28 (1994) 180Adoi httpsdoiorg101021es00053a001
10 Spigno G Jauregi P Recovery of gallic acid with colloi-dal gas aphrons (CGA) Int J Food Eng 1 (2005) 1doi httpsdoiorg1022021556-37581038
11 Spigno G Dermiki M Pastori C Casanova F Jauregi P Recovery of gallic acid with colloidal gas aphrons gen-erated from a cationic surfactant Sep Purif Technol 71 (2010) 56doi httpsdoiorg101016jseppur200911002
12 Quici N Litter M I Heterogeneous photocatalytic degra-dation of gallic acid under different experimental condi-tions Photochem Photobiol Sci 8 (2009) 975doi httpsdoiorg101039b901904a
13 Singh M Jha A Kumar A Hettiarachchy N Rai A K Sharma D Influence of the solvents on the extraction of major phenolic compounds (punicalagin ellagic acid and gallic acid) and their antioxidant activities in pomegranate aril J Food Sci Technol 51 (2014) 2070doi httpsdoiorg101007s13197-014-1267-0
14 Claudio A F M Ferreira A M Freire C S R Silves-tre A J D Freire M G Coutinho J A P Optimization of the gallic acid extraction using ionic-liquid-based aque-ous two-phase systems Sep Purif Technol 97 (2012) 142doi httpsdoiorg101016jseppur201202036
15 Li Y Wang Y Li Y Dai Y Extraction of glyoxylic acid glycolic acid acrylic acid and benzoic acid with trialkyl-phosphine oxide J Chem Eng Data 48 (2003) 621doi httpsdoiorg101021je020174r
16 Daneshfar A Ghaziaskar H S Homayoun N Solubility of gallic acid in methanol ethanol water and ethyl acetate J Chem Eng Data 53 (2008) 776doi httpsdoiorg101021je700633w
17 Wasewar K L Shende D Z Reactive extraction of caproic acid using tri- n -butyl phosphate in hexanol octa-nol and decanol J Chem Eng Data 56 (2011) 288doi httpsdoiorg101021je100974f
18 Keshav A Chand S Wasewar K L Reactive extraction of acrylic acid using tri- n -butyl phosphate in different di-luents J Chem Eng Data 54 (2009) 1782doi httpsdoiorg101021je800856e
19 Zhong K Wang Q Optimization of ultrasonic extraction of polysaccharides from dried longan pulp using response surface methodology Carbohydr Polym 80 (2010) 19doi httpsdoiorg101016jcarbpol200910066
20 Marchitan N Cojocaru C Mereuta A Duca G Crete-scu I Gonta M Modeling and optimization of tartaric acid reactive extraction from aqueous solutions A compar-ison between response surface methodology and artificial neural network Sep Purif Technol 75 (2010) 273doi httpsdoiorg101016jseppur201008016
21 Messikh N Samar M H Messikh L Neural network analysis of liquidndashliquid extraction of phenol from waste-water using tbp solvent Desalination 208 (2007) 42doi httpsdoiorg101016jdesal200604073
22 Montgomery D Design and analysis of experiments (Fifth Ed) John Wiley amp Sons Inc New Jersey 2001
23 Box G E P Hunter J S H W G Statistics for experi-menters design innovation and discovery (second ed) Vol 48 John Wiley ampSons New Jersey 2006
24 Dornier M Decloux M Trystram G Lebert A Dynam-ic modeling of crossflow microfiltration using neural net-works J Memb Sci 98 (1995) 263doi httpsdoiorg1010160376-7388(94)00195-5
25 Niemi H Bulsari A Palosaari S Simulation of mem-brane separation by neural networks J Memb Sci 102 (1995) 185doi httpsdoiorg1010160376-7388(94)00314-O
26 Bourquin J Schmidli H van Hoogevest P Leuenberger H Pitfalls of artificial neural networks (ANN) modelling technique for data sets containing outlier measurements us-ing a study on mixture properties of a direct compressed dosage form Eur J Pharm Sci 7 (1998) 17doi httpsdoiorg101016S0928-0987(97)10027-6
27 Cornell A I Khuri J A Response surfaces Designs and analyses (second ed) Marcel Dekker Inc New York 1996
28 Bezerra M A Santelli R E Oliveira E P Villar L S Escaleira L A Response surface methodology (RSM) as a tool for optimization in analytical chemistry Talanta 76 (2008) 965doi httpsdoiorg101016jtalanta200805019
29 Hornik K Stinchcombe M White H Multilayer feedfor-ward networks are universal approximators Neural Net-works 2 (1989) 359doi httpsdoiorg1010160893-6080(89)90020-8
30 Hagan M T Demuth H B Beale M H Neural network design PWS Publishing Company Boston 1995
31 Keshav A Wasewar K L Chand S Extraction of propi-onic acid using different extractants (tri-n-butylphosphate tri-n-octylamine and Aliquat 336) Ind Eng Chem Res 47 (2008) 6192doi httpsdoiorg101021ie800006r
32 Yang S T White S A Hsu S T Extraction of carboxyl-ic acids with tertiary and quaternary amines Effect of pH Ind Eng Chem Res 30 (1991) 1335doi httpsdoiorg101021ie00054a040
33 Keshav A Wasewar K L Chand S Extraction of Acryl-ic Propionic and butyric acid using Aliquat 336 in oleyl alcohol Equilibria and effect of temperature Ind Eng Chem Res 48 (2009) 888doi httpsdoiorg101021ie8010337
34 Uslu H Kırbaslar S I Effect of temperature and initial acid concentration on the reactive extraction of carboxylic acids J Chem Eng Data 58 (2013) 1822doi httpsdoiorg101021je4002202
35 Pham D T Sagiroglu S Training multilayered percep-trons for pattern recognition A comparative study of four training algorithms Int J Mach Tools Manuf 41 (2001) 419doi httpsdoiorg101016S0890-6955(00)00073-0
36 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
teractive effects of operating variables on the re-sponse The rotatable orthogonal central composite design (RO-CCD) was used to develop a quadratic experimental model In the rotatable design the prediction variance of a point depends only on its distance from the centre of the design irrespective of its direction The orthogonality allows all the fac-tors to be estimated independently RO-CCD con-sists of the corner points nc = 2k the star points ns = 2k and the centre points n0
To make CCD both orthogonal and rotatable the centre points n0 were calculated as 4 by substi-tuting k = 4 factors (CGA0 CTBP T pH0) in the Eq (2) below developed by Khuri and Cornell27
1 20 cn (4n ) 4 2k= + minus (2)
The axial distance a was calculated to be 1607 using the expression below
(3)
( )1 421 2 1 2 c
c s 0 cnn n n n 4
a = + + minus
Table 1 shows the corner points nc (Run 1ndash16 O1ndashO16) star points ns (Run 17ndash24 S1ndashS8) and centre points n0 (Run 25ndash28 C1ndashC4) The sum nc + n0 + ns = 28 Hence 28 batch experiments were carried out to construct the RSM model The rela-tionship between independent variables (factors) and the extraction efficiency (response) was estab-lished using a second-order polynomial equation as follows28
(4)
4 4 42
0 1 1 1i i ii i ij i ji i iE x x x xb b b b
= = == + Σ + Σ + Σ Σ
where E is the predicted response (extraction effi-ciency ) and bi bj bij are the regression coeffi-cients representing linear quadratic and interactive effects of factors ie x1 (CGA0) x2 (CTBP) x3 (T) and x4 (pH0) Table 2 shows the independent variables and their levels which were decided based on the trial experiments and literature survey
The regression analysis and process optimiza-tion were performed using the Statistical Software package ldquoDesign Expertrdquo The significance of the regression model was evaluated using the Fischer
distribution (F-value) and null-hypothesis test (p-value) A larger F-value indicates a better fit of model to the experimental extraction efficiency A p-value less than 005 (p lt 005) indicates the de-sign variable of a model contributing less than 5 change in the response Therefore the variable with a larger F-value and p lt 005 was considered signif-icant22
Three-dimensional surface plots were drawn to determine the interactive effect of the process vari-ables on extraction efficiency The numerical opti-mization method was used to identify the combina-tion of variables that jointly optimize a single response E () in the Design-Expert software
Artificial Neural Network (ANN)
ANN is considered a powerful modelling tool to study the process of multivariable complex non-linear systems The single hidden layer is a univer-sal approximation for the multilayer feed forward neural network29 The operation of a single neuron is shown in Figure 2 The input to a neuron is its bias (bj) and the sum of weighted outputs from the previous layer (yindash1) whereas the output depends on the transfer function f(xij) of its input and can be represented as (5)
i-1J ijij i-1k i-1k ijk 1
ij ij
ij ij
( ) for i 1
for i 1 (input
Input
Output
layer)
x w y b
y f xy x
== +
= gt
= =
sum
where xij yij bij are the input output and bias of jth neuron to ith layer respectively and Jindash1 is the to-tal neurons in the (i-1)th layer The ij
i-1kw is the weight between kth neuron of (i-1)th layer and jth neuron of ith layer and f(xij) is a suitable transfer function
The sigmoid transfer function (logsig) for hid-den neurons and the linear transfer function (pure-lin) for output neuron were used as given below30
Ta b l e 2 ndash Levels of independent variables used for rotatable orthogonal central composite design
Independent variables (factors)
Levels of factors
ndashα ndash1 0 +1 +α
CGA0 (g Lndash1) 0275 07 14 21 2525
CTBP ( vv) 1483 30 55 80 9518
T (oC) 1284 18 265 35 4016
pH0 0775 12 19 26 3025 α = 1607
34
Figure 2 Operation of single neuron in ANN
F i g 2 ndash Operation of single neuron in ANN
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 37
(6)
1 ( )1 xlogsig x
eminus=+
( )purelin x x= (7)
Matlab software R2012a was used for ANN modelling The connections between the input layer and the hidden layer and the hidden layer and the output layer consist of weights and biases The weights which express the interactions between the neurons were used in two types of matrix ie input weight matrix (IW) connecting the input layer to the hidden layer and the layer weight matrix (LW) connecting the hidden layer to the output layer Each neuron in the first layer was connected to all the neurons in the next layer with a total of 40 connections between the input and hidden layers The bias determines the threshold to fire a neuron Each neuron has an individual bias value except for the input layer neurons30 The optimal number of neurons in the hidden layer was decided by the trial and error approach by varying it until a minimum mean squared error (MSE) was achieved MSE is an error function which measures the performance of the ANN model and can be calculated as fol-lows21
(8)
i N
ipred iexpi 1
( )MSE
E Esum
where N is the number of experimental data points Eipred and Eiexp are the ANN predicted and experi-mental responses respectively
To compare both constructed models an error analysis was carried out with experimental and the predicted response using coefficient of determina-tion (R2) root mean square error (RMSE) absolute average relative error (AARE) and standard error of prediction (SEP) calculated as follows
(9)
N 2ipred iexp2 i 1
2ipred avgi 1
( )1
( )N
E ER
E E=
=
minus= minus
minussumsum
(10)
N 2ipred iexpi=1
( )RMSE
NE Eminus
= sum
(11)
N ipred iexp
i=1iexp
100AAREN
E E
E
minus= sum
(12) avg
RMSESEP 100E
=
Where Eavg is the average value of the experi-mental response
Results and discussion
The gallic acid is extracted using TBP by form-ing a complex (GATBP) via interfacial reaction (Eq 13) The phosphoryl group of TBP forms a hy-drogen bond with the proton donor gallic acid to form a stable complex in the organic phase This can be represented as
aq org p m orgpGA mTBP (GA TBP )+ (13)
It is evident that the maximum GATBP com-plex formation results in maximum removal effi-ciency of GA that can be obtained only with a par-ticular combination of T pH0 CGA0 and CTBP However a large number of experiments may be required for identifying the suitable process condi-tions Therefore the RSM and ANN models were developed to identify the optimal process conditions required for maximum extraction efficiency of GA
RSM modelling of GA extraction
The design of experiments and experimental extraction efficiency E () are summarized in Table 1 The extraction efficiency for each experiment was calculated using Eq (1) Runs 25ndash28 (C1ndashC4) of centre point were used to determine the experi-mental error and reproducibility of the data The experimental data were regressed to obtain the anal-ysis of variance (ANOVA) as shown in Table 3 The model is highly significant and statistically fit with 9999 confidence level The empirical rela-tionship between the response and the independent variables is expressed in terms of coded variables as follows
(14)
1 2 32 2 2
4 1 2 324 1 2 1 3 1 4
2 3 2 4 3 4
9737 0017 346 051 ndash
051 113 151 011 ndash
0055 024 ndash 025 0011 ndash042 038 024
E x x x
x x x x
x x x x x x xx x x x x x
= + + minus
minus minus minus minus
minus + minus
+ + +
The positive and negative sign of the coeffi-cients indicate the linear increasing or decreasing effect of the corresponding variable on the response respectively The significance of a particular vari-able increases with the sum of square (SS) The ANOVA analysis (Table 3) suggests that CTBP (p = 00001 SS = 25336 F = 88228) has a most signif-icant effect on the response as compared to T (p = 00007 SS = 558 F = 1943) and pH0 (p = 00007 SS = 554 F = 1928) The linear coefficients were more significant than the quadratic or interactive coefficients
The degree of fitness of the model to the exper-imental data was determined by analysing the coef-ficient of determination (R2) The R2 obtained was 098 which indicates that the developed model
38 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
could well explain 98 of total variation in the response and confirms the goodness of fit for the RSM model The R2 (adj) and R2 (pred) values were 097 and 092 respectively which are closer to the R2 value of 0976 suggesting the significance and reasonable prediction ability of the model Ad-equate precision (signal to noise ratio) compares the predicted values with the average prediction error Adequate precision with a ratio greater than 4 is al-ways desirable The value of 3082 indicates an ad-equate signal The coefficient of variation (CV) was calculated to be 056 which indicates better pre-
cision and reliability of the experiments The Stan-dard deviation (S) with 054 indicates strong com-pliance with the predicted response The predicted residual sum of square (PRESS) is a measure of predictive power of the developed model The PRESS value of 2214 suggests that the regression model is significant to predict the response for a new experiment The coded terms of RSM model was then converted into actual variables and repre-sented as follows
(15)
02
0 02 2 2
0
0 0
0 0
0 0
9011 118( ) 029( )
011( ) 257(pH ) 026( )
00024( ) 00015( ) 011(pH )0014( )( ) 0043( )( )0022( )(pH ) 0002( )( )0022( )(pH ) 0041( )(pH )
GA TBP
GA
TBP
GA TBP GA
GA TBP
TBP
E C C
T C
C TC C C TC C TC T
= + + minus
minus minus minus minus
minus minus minus +
+ minus minus
minus + +
+ +
Response surface analysis process optimization and validation
Figures 3a and 3b illustrate the 3D surface plots for the individual and the interactive effects of independent variables on extraction efficiency In Figure 3a the concentration of extractant CTBP (x2) and its interaction with pH (x2 x4) show a positive effect on extraction efficiency (Table 3) whereas pH alone shows a negative effect A curvilinear ef-fect on extraction efficiency was observed due to the positive linear effect and negative quadratic ef-fect of CTBP This is attributed to the presence of the highly polar phosphoryl group (gtP=O) of TBP making it a strong Lewis base resulting in higher GATBP complex formation during the extraction process18 Hence the higher CTBP resulted in higher extraction The concentration of the undissociated acid always increases at lower pH due to the higher proton concentration TBP extracts only undissoci-ated form of acid31 hence extraction efficiency E () increases at lower pH values A similar effect of pH on the increase in efficiency was also observed in the extraction of other carboxylic acids32
Figure 3b illustrates that CTBP (x2) and T (x3) has a positive and negative effect respectively on ex-traction efficiency E () whereas the combined effect (x2 x3) shows a positive effect (Table 3) With an increase in temperature the formation of GATBP complex decreases The complex is formed via intermolecular hydrogen bonding through pro-ton transfer between acid and extractant The pro-cess of hydrogen bond formation is exothermic in nature which shifted the extraction equilibrium in backward direction at higher temperature thus re-ducing extraction efficiency Similar observations have been made by many researchers3334 Therefore it is concluded that lower temperature and pH with higher CTBP favour the extraction efficiency
Ta b l e 3 ndash Analysis of variance (ANOVA) for the experimen-tal results of the rotatable orthogonal central com-posite design
Source Coded terms
Coef- ficient
Sum of Squares F-value p-value Remarks
Model 3033 7544 00001 Significant
Constant 9737 00001
Linear
CGA0 x1 0017 000604 0021 0887
CTBP x2 346 25336 88228 00001 Significant
T x3 ndash051 558 1943 00007 Significant
pH0 x4 ndash051 554 1928 00007 Significant
Square
CGA0 CGA0 x12 ndash013 022 077 0395
CTBP CTBP x22 ndash151 3024 1053 00001 Significant
T T x32 ndash011 017 058 0461
pH0 pH0 x42 ndash0055 0041 014 0712
Interaction
CGA0 CTBP x1 x2 024 092 319 0097
CGA0 T x1 x3 ndash025 103 351 0081
CGA0 pH0 x1 x4 ndash0011 000181 000629 0940
CTBP T x2 x3 042 288 1003 0007 Significant
CTBP pH0 x2 x4 038 236 823 0013 Significant
T pH0 x3 x4 024 096 333 0091
Statistics
Adequate precision 3082
PRESS 2214
S 054
CV () 056
R2 098
R2(adj) 097
R2(pred) 092
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 39
Optimization of process variables CGA0 CTBP T and pH0 was performed using the numerical optimi-zation method to achieve optimal conditions to maximize the extraction The optimum variables were CGA0 = 234 g Lndash1 CTBP = 6565 (vv) T = 184 oC and pH0 = 18 with the predicted optimum extraction of 998 The experiment at optimum conditions was carried out in duplicate to validate the model predictions at the optimum conditions The experimental extraction efficiency obtained at the optimum conditions was found to be 9943 which is in close agreement with the model predict-
35
(a)
(b)
Fig 3 Three dimensional surface plot showing (a) the effect of CTBP and pH0 on the extraction
of GA at the centre level (T = 265 oC CGA0 = 14 g L-1) and (b) the effect of CTBP and T on
the extraction of GA at the centre level (pH0 = 19 CGA0 = 14 g L-1)
F i g 3 ndash Three dimensional surface plot showing (a) the ef-fect of CTBP and pH0 on the extraction of GA at the centre level (T = 265 oC CGA0 = 14 g Lndash1) and (b) the effect of CTBP and T on the extraction of GA at the centre level (pH0 = 19 CGA0 = 14 g Lndash1)
Ta b l e 4 ndash Additional experimental dataset used for the con-struction of ANN model
RunIndependent variables Response E ()
CGA0 (g Lndash1)
CTBP ( vv)
T (oC) pH0Experi- mental ANN Remark
29 05 65 11 196 9895 9868 Training
30 25 30 11 303 9488 9508 Training
31 14 30 11 302 9304 9325 Training
32 18 53 11 213 9728 9700 Training
33 25 75 11 305 9886 9851 Training
34 12 22 11 298 9013 9099 Validation
35 04 82 11 167 9723 9766 Training
36 028 15 11 225 8572 8704 Validation
37 10 30 15 308 9277 9243 Training
38 25 30 15 303 9473 9429 Validation
39 028 95 15 078 9730 9747 Training
40 14 30 16 301 9293 9305 Training
41 25 30 16 302 9449 9402 Training
42 25 30 18 303 9359 9347 Training
43 05 25 20 197 9076 9036 Training
44 14 30 20 302 9274 9212 Training
45 25 30 25 301 9124 9179 Training
46 028 15 25 078 8485 8482 Training
47 25 30 27 303 9263 9150 Validation
48 14 30 27 302 8768 8934 Training
49 04 85 27 139 9808 9796 Training
50 24 72 27 255 9866 9843 Training
51 12 24 27 22 8931 9044 Testing
52 10 62 28 172 9758 9826 Testing
53 15 72 28 165 9852 9831 Testing
54 25 30 28 303 9171 9136 Training
55 16 32 30 246 9224 9210 Training
56 18 78 30 261 9895 9852 Testing
57 18 27 30 261 9039 8930 Training
58 12 65 30 254 9806 9792 Training
59 25 95 30 078 9864 9879 Training
60 14 30 33 302 8798 8766 Training
61 25 30 35 303 9114 9062 Validation
62 25 30 35 3012 9038 9063 Training
63 14 30 37 302 8733 8729 Training
64 24 80 37 222 9859 9857 Training
65 08 60 37 218 9718 9694 Training
66 24 16 37 209 8515 8536 Training
67 16 32 37 178 9104 9102 Training
68 25 30 41 303 9099 9032 Training
69 10 30 41 150 8883 8890 Training
40 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
ed value In Table 1 Runs no 3 4 12 20 show the extraction efficiency above 992 at the cost of 80ndash95 TBP while the optimum condition gives the same efficiency with 65 TBP
ANN modelling of the GA extraction process
In ANN the three layered feed forward error back propagation (FF-EBP) with four neurons in the input layer (input parameters CGA0 CTBP T pH0) ten neurons in the hidden layer (optimum) one neuron in the output layer (extraction efficien-cy) was selected Figure 4 illustrates the ANN ar-chitecture applied for predicting the extraction effi-ciency of GA
A total of 69 experimental data-points (Table 1 and 4) were used to train the ANN model The input and the output data-points were normalized in the range 0 to 1 and 03 to 07 respectively This keeps the back propagation error within the limits and avoids over fitting of the data due to very large and very small values of weights and speeds up training time The data were normalized between the two points n1 and n0 using the following generalized equation (16)
minn 1 0 0
max min
( )(n n ) n( )
x xxx x
minus= minus + minus
where xn is normalized data-point and x xmin xmax are the actual minimum and maximum of inputoutput data The data were split into subsets of training (80 55 data-points) validation (10 7 data-points) and testing (10 7 data-points) Trainlm a training function that updates weights and bias values based on Levenberg-Marquardt backpropagation (LMB) algorithm was used for training the network LMB is an iterative technique which reduces the performance function in each it-eration and makes trainlm the fastest error back-propagation (EBP) algorithm for a moderate sized network35 The externally normalized input values were forwarded from the input layer to the hidden layer and then to the output layer to predict the re-sponse The error MSE was backpropagated from the output to the hidden layer and thereafter to the input layer to modify the weights (IW LW) and bi-ases (b1 b2) Figure 5 describes the general concept of FF-EBP algorithm used in ANN training to pre-dict extraction efficiency During the series of trials for EBP the training algorithm connection weights transfer function hidden layers and hidden neurons were varied and a network topology with an opti-mal network architecture (4101) was selected based on MSE value
The training was successfully terminated after 12 epochs (iterations) as the performance function
36
Fig 4 ANN architecture used for the prediction of GA extraction efficiency F i g 4 ndash ANN architecture used for the prediction of GA extraction efficiency
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 41
F i g 5 ndash Feed forward backpropagation algorithm used in the ANN training for the prediction of GA extraction efficiency
MSE reached a minimum value of 5middot10ndash4 which is lower than the set goal E0 = 1middot10ndash3 (Figure 6) The similar characteristic curve for the test and the vali-dation were observed suggesting no significant over-fitting The estimated MSE values for training (1middot10ndash4) testing (3middot10ndash4) and validation data (5middot10ndash4) were found to be lower than the set goal which is desirable Table 5 gives the optimal weights and bias Wherein IW = HtimesA b1 = Htimes1 LW =OtimesH ma-trix where H is hidden layer neurons (10) A is in-put layer neurons (4) and O is the output layer neu-ron (1) The bias b2 is the single value associated with a single neuron at the output
Figure 7 describes the ANN regression plot for training validation testing and the overall predic-tion set in the form of network output versus exper-imental extraction efficiency It can be observed that the network output values were close to the ex-
F i g 6 ndash MSE for training validation and test dataset com-puted using LMB (with performance of 5middot10ndash4 at 12 epochs and goal of 1middot10ndash3)
38
Fig 6 MSE for training validation and test dataset computed using LMB (with performance of
5middot10-4 at 12 epochs and goal of 1middot10-3)
42 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
perimental extraction efficiency in all the cases The correlation coefficients lsquoRrsquo for training validation and testing were 0995 0984 0989 respectively
whereas the overall prediction set was 0993 which confirms that the ANN model is satisfactory for in-terpolating the experimental data
Ta b l e 5 ndash Optimal values of weights and biases obtained from ANN training using Levenberg-Marquardt backpropagation algorithm
Input weight matrix IW (Weights to Hidden layer from inputs)
54485 74654 14374 2009934801 10038 71242 0667899763 46322 41336 7966216333 44955 0545 2141845359 42326 56151 16171
8043 27753 20607 0620637705 30939 60152 7216681829 118002 62271 3
minus minusminus minusminus minus
minus minus minusminus minus
minus minus 806645814 08091 62607 2537333117 47426 72301 08914
minusminus
Bias vector (Hidden layer) b1 102912 114372 94153 04321 26838 50968 73116 01728 4231 37916 Tminus minus minus minus
Layer weight vector LW (Weight to output layer from hidden layer)
01575 04295 00956 03633 02071 02006 01154 02645 02656 0313minus minus minus minus minus minus
Bias scalar (output layer) b2 06545
F i g 7 ndash ANN model showing the regression plots for training validation testing and overall prediction set
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 43
The ANN model was also validated using sta-tistical analysis ANOVA The degree of freedom due to regression (DFregression) and residuals (DFresidual) was calculated as follows20
regressionDF M 1= minus (17)
residualDF N M= minus (18)
M H(A O 1) O= + + + (19)
where M is the total number of network connec-tions including weights and biases N is the total number of experiments (N = 69) performed to con-struct and validate the ANN model Table 6 shows the ANOVA for the developed ANN model The F-value and the p-value were obtained from ANO-VA for the ANN model as 104 and 00004 respec-tively which confirms the significance of the ANN model Table 7 shows the values of R2 and R2 (adj) as 0986 and 088 respectively indicating the good fit of the ANN model to the experimental extraction efficiency Therefore the ANN model for predicting extraction efficiency E () for the extraction of GA can be written as follows
n n 1 2 (LW (IW ) )y purelin logsig x b b= times times + + (20)
Comparison of ANN and RSM models
The prediction and generalization capabilities of RSM and ANN models were evaluated based on its R2 RMSE AARE and SEP values Table 1 and 4 indicate that both the models predict well the ex-perimental extraction efficiency and their R2 values are closer to 1 Therefore both models can be con-sidered good for predicting the extraction efficiency of GA The generalization capabilities of both the models were compared only with unseen dataset Table 7 shows the unseen data set which were not used for the model development Figure 8 shows the comparative parity plot for RSM and ANN model predictability for the unseen data set and suggests that both the model predictions fit the ex-perimental data very well The comparative values of R2 RMSE AARE and SEP values for both mod-els using unseen data are shown in Table 8 Thus higher R2 and lower RMSE AARE SEP values for ANN indicate that the ANN model is superior over the RSM model for its generalization capability of the GA extraction process
Ta b l e 6 ndash Analysis of variance (ANOVA) analysis for ANN model
Source DFa SSb MSc F-value p-value R2 R2(adj)
Model or Regression 60 11421 1904 1041 00004 0986 088Residual Error 9 1646 183Total 69 11587aDegree of Freedom bSum of Squares cMean Square
Ta b l e 7 ndash Unseen experimental dataset used in the developed RSM and ANN models for the prediction of ex-traction efficiency of GA
RunIndependent variables Response E ()
CGA0 (g Lndash1)
CTBP ( vv)
T(oC) pH0Experi- mental RSM ANN
1 16 15 11 174 9057 9015 91402 25 30 40 30 9005 8836 90373 25 23 37 234 8888 8723 88534 25 37 27 211 9226 9323 92275 25 45 27 185 9525 9531 94786 25 23 37 234 8798 8723 88537 14 30 40 301 8533 8967 87368 18 20 32 221 8745 8805 86639 10 30 24 26 9256 9169 925710 18 36 32 21 9302 9285 938911 10 30 40 15 8883 9118 8908
Ta b l e 8 ndash Comparison of RSM and ANN models for its gen-eralization capability (using unseen data)
Parameters RSM ANN
R2 061 0915RMSE 173 080
AARE 143 066 SEP 191 088
40
Fig 8 Comparison of RSM and ANN model predictions for the unseen data (data not used in
the model development)
F i g 8 ndash Comparison of RSM and ANN model predictions for the unseen data (data not used in the model development)
44 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
Conclusion
The present work suggests that gallic acid can be efficiently removed from aqueous streams using TBP in hexanol The RSM and ANN models were developed to predict the extraction efficiency of GA by varying the parameters temperature concentra-tion of TBP in aqueous phase pH and initial con-centration of gallic acid in aqueous solution The importance of each parameter and its synergistic interactive effects on the extraction of GA was ex-plained with a minimum number of experiments us-ing the RSM model Both models were efficient in the prediction of GA extraction efficiency having close agreement with the experimental extraction efficiency However ANN showed superiority over RSM in terms of accuracy but the marginal errors between the models were very low The optimal conditions to achieve maximum extraction as pre-dicted by the RSM model were CGA0 = 234 g Lndash1 CTBP = 6565 (vv) T = 184 oC and pH0 = 18 at which an experimental extraction efficiency of 995 was observed The models also showed bet-ter performance with the use of unseen data which confirmed their generalization capabilities Thus the present investigation on experiments model-
ling and optimization of the GA extraction process could be of great significance for the design and evaluation of the performance of similar systems
S u p p l e m e n t a r y I n f o r m a t i o n
Supplementary information for confirmation of 2 h shaking time for reactive extraction of gallic acid and retention time of 39 min in HPLC analysis
The trial experiments were conducted for ob-taining an optimum shaking period Equal volumes (20 mL) of aqueous and organic phases were shak-en for different time intervals (5ndash240 min) at 250 rpm Equilibrium was achieved after half an hour (Figure S1) Hence further experiments were con-ducted with 2 h shaking time (sufficient to achieve equilibrium)
Calibration curve and chromatogram for HPLC analysis of gallic acid was obtained by diluting the mother stock of 5 g Lndash1 to prepare different concen-trations (5ndash300 mg Lndash1) of standard sample solu-tions
UV absorption spectra for gallic acid was ob-tained using UV Spectrophotometer Shimadzu UV-1800
F i g S 1 ndash Experimental extraction efficiency (E) of gallic acid versus time for the operating conditions CGA0 = 2 g Lndash1 CTBP = 40 (vv) T = 30 oC pH0 = 3
F i g S 2 ndash Calibration curve for gallic acid obtained us- ing HPLC
44
Fig S4b HPLC Chromatogram of gallic acid at 264 nm and 396 min
F i g S 3 ndash HPLC Chromatogram of gallic acid at 264 nm and 396 min
Ext
ract
ion
effic
ienc
y E
()
Peak
are
a (m
AU
s)
Time (min) Gallic acid concentration (mg Lndash1)
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 45
N o m e n c l a t u r e
AARE ndash absolute average relative errorCGAaq ndash equilibrium concentration of GA in aqueous
phase g Lndash1
CGAorg ndash equilibrium concentration of GA in organic phase g Lndash1
CGA0 ndash initial concentration of GA in aqueous phase g Lndash1
CTBP ndash TBP concentration in organic phase DF ndash degree of freedom in ANOVA analysisE ndash extraction efficiency response F-value ndash Fischer distribution value (ratio of variances)H A O ndash number of neurons in hidden layer input lay-
er and output layer respectivelyIW ndash input weight matrixk ndash number of input variables or factors in
RO-CCDlogsig ndash sigmoidal transfer function for neural net-
workLW ndash layer weight matrixM ndash number of network connections including
weights and biasesMS ndash mean square in ANOVA analysisMSE ndash mean squared errorN ndash total number of experiments to construct
ANN modeln0 ndash centre points in RO-CCDnc ndash corner points or factorial points in RO-CCDns ndash star or axial points in RO-CCDpH0 ndash initial pH of aqueous GA solutionPRESS ndash predicted residual sum of squarespurelin ndash linear transfer function for neural networkp-value ndash null-hypothesis testR ndash coefficient of correlation
R2 ndash coefficient of determinationR2 (adj) ndash adjusted R-squaredR2 (pred) ndash predicted R-squaredRMSE ndash root mean square errorSEP ndash standard error of predictionSS ndash sum of squares in ANOVA analysisT ndash extraction temperature oCw ndash connection weight between layersx1 x2 x3 x4 ndash coded levels of independent vaiables in
RSMxij yij bij ndash network input output and bias of jth neu-
ron to ith layer respectivelya ndash distance of each star point from centre in
RO-CCDbi bj bij ndash regression coefficients representing linear
quadratic and interactive effects in RSM
R e f e r e n c e s
1 Benitez F J Real F J Acero J L Leal A I Garcia C Gallic acid degradation in aqueous solutions by UVH2O2 treatment Fentonrsquos reagent and the photo-Fenton sys-tem J Hazard Mater 126 (2005) 31doi httpsdoiorg101016jjhazmat200504040
2 El-Gohary F A Badawy M I El-Khateeb M A El-Kalliny A S Integrated treatment of olive mill wastewater (OMW) by the combination of fentonrsquos reaction and anaer-obic treatment J Hazard Mater 162 (2009) 1536doi httpsdoiorg101016jjhazmat200806098
3 Herrero M Cifuentes A Ibanez E Sub- and supercriti-cal fluid extraction of functional ingredients from different natural sources plants food-by-products algae and mi-croalgae A Review Food Chem 98 (2006) 136doi httpsdoiorg101016jfoodchem200505058
4 Kougan G B Tabopda T Kuete V Verpoorte R Sim-ple phenols phenolic acids and related esters from the me-dicinal plants of africa in Kuet V (First Ed) Medicinal plant research in Africa Pharmacology and chemistry El-sevier Inc London 2013 225ndash249
45
UV absorption spectra for gallic acid was obtained using UV Spectrophotometer Shimadzu UV-1800
Fig S3 Absorption spectra of gallic acid showing max = 264 nm
--
F i g S 4 ndash Absorption spectra of gallic acid showing lmax = 264 nm
46 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
5 Kapellakis I E Tsagarakis K P Crowther J C Olive oil history production and by-product management Rev Environ Sci Biotechnol 7 (2008) 1doi httpsdoiorg101007s11157-007-9120-9
6 Calace N Nardi E Petronio B Pietroletti M Adsorp-tion of phenols by papermill sludges Environ Pollut 118 (2002) 315doi httpsdoiorg101016S0269-7491(01)00303-7
7 Fki I Allouche N Sayadi S The use of polyphenolic extract purified hydroxytyrosol and 34-dihydroxyphenyl acetic acid from olive mill wastewater for the stabilization of refined oils A potential alternative to synthetic antioxi-dants Food Chem 93 (2005) 197doi httpsdoiorg101016jfoodchem200409014
8 Ollis D F Pelizzetti E Serpone N Photocatalyzed de-struction of water contaminants Environ Sci Technol 25 (1991) 1522doi httpsdoiorg101021es00021a001
9 Masten S J Davies S H The use of ozonation to de-grade organic contaminants in wastewaters Environ Sci Technol 28 (1994) 180Adoi httpsdoiorg101021es00053a001
10 Spigno G Jauregi P Recovery of gallic acid with colloi-dal gas aphrons (CGA) Int J Food Eng 1 (2005) 1doi httpsdoiorg1022021556-37581038
11 Spigno G Dermiki M Pastori C Casanova F Jauregi P Recovery of gallic acid with colloidal gas aphrons gen-erated from a cationic surfactant Sep Purif Technol 71 (2010) 56doi httpsdoiorg101016jseppur200911002
12 Quici N Litter M I Heterogeneous photocatalytic degra-dation of gallic acid under different experimental condi-tions Photochem Photobiol Sci 8 (2009) 975doi httpsdoiorg101039b901904a
13 Singh M Jha A Kumar A Hettiarachchy N Rai A K Sharma D Influence of the solvents on the extraction of major phenolic compounds (punicalagin ellagic acid and gallic acid) and their antioxidant activities in pomegranate aril J Food Sci Technol 51 (2014) 2070doi httpsdoiorg101007s13197-014-1267-0
14 Claudio A F M Ferreira A M Freire C S R Silves-tre A J D Freire M G Coutinho J A P Optimization of the gallic acid extraction using ionic-liquid-based aque-ous two-phase systems Sep Purif Technol 97 (2012) 142doi httpsdoiorg101016jseppur201202036
15 Li Y Wang Y Li Y Dai Y Extraction of glyoxylic acid glycolic acid acrylic acid and benzoic acid with trialkyl-phosphine oxide J Chem Eng Data 48 (2003) 621doi httpsdoiorg101021je020174r
16 Daneshfar A Ghaziaskar H S Homayoun N Solubility of gallic acid in methanol ethanol water and ethyl acetate J Chem Eng Data 53 (2008) 776doi httpsdoiorg101021je700633w
17 Wasewar K L Shende D Z Reactive extraction of caproic acid using tri- n -butyl phosphate in hexanol octa-nol and decanol J Chem Eng Data 56 (2011) 288doi httpsdoiorg101021je100974f
18 Keshav A Chand S Wasewar K L Reactive extraction of acrylic acid using tri- n -butyl phosphate in different di-luents J Chem Eng Data 54 (2009) 1782doi httpsdoiorg101021je800856e
19 Zhong K Wang Q Optimization of ultrasonic extraction of polysaccharides from dried longan pulp using response surface methodology Carbohydr Polym 80 (2010) 19doi httpsdoiorg101016jcarbpol200910066
20 Marchitan N Cojocaru C Mereuta A Duca G Crete-scu I Gonta M Modeling and optimization of tartaric acid reactive extraction from aqueous solutions A compar-ison between response surface methodology and artificial neural network Sep Purif Technol 75 (2010) 273doi httpsdoiorg101016jseppur201008016
21 Messikh N Samar M H Messikh L Neural network analysis of liquidndashliquid extraction of phenol from waste-water using tbp solvent Desalination 208 (2007) 42doi httpsdoiorg101016jdesal200604073
22 Montgomery D Design and analysis of experiments (Fifth Ed) John Wiley amp Sons Inc New Jersey 2001
23 Box G E P Hunter J S H W G Statistics for experi-menters design innovation and discovery (second ed) Vol 48 John Wiley ampSons New Jersey 2006
24 Dornier M Decloux M Trystram G Lebert A Dynam-ic modeling of crossflow microfiltration using neural net-works J Memb Sci 98 (1995) 263doi httpsdoiorg1010160376-7388(94)00195-5
25 Niemi H Bulsari A Palosaari S Simulation of mem-brane separation by neural networks J Memb Sci 102 (1995) 185doi httpsdoiorg1010160376-7388(94)00314-O
26 Bourquin J Schmidli H van Hoogevest P Leuenberger H Pitfalls of artificial neural networks (ANN) modelling technique for data sets containing outlier measurements us-ing a study on mixture properties of a direct compressed dosage form Eur J Pharm Sci 7 (1998) 17doi httpsdoiorg101016S0928-0987(97)10027-6
27 Cornell A I Khuri J A Response surfaces Designs and analyses (second ed) Marcel Dekker Inc New York 1996
28 Bezerra M A Santelli R E Oliveira E P Villar L S Escaleira L A Response surface methodology (RSM) as a tool for optimization in analytical chemistry Talanta 76 (2008) 965doi httpsdoiorg101016jtalanta200805019
29 Hornik K Stinchcombe M White H Multilayer feedfor-ward networks are universal approximators Neural Net-works 2 (1989) 359doi httpsdoiorg1010160893-6080(89)90020-8
30 Hagan M T Demuth H B Beale M H Neural network design PWS Publishing Company Boston 1995
31 Keshav A Wasewar K L Chand S Extraction of propi-onic acid using different extractants (tri-n-butylphosphate tri-n-octylamine and Aliquat 336) Ind Eng Chem Res 47 (2008) 6192doi httpsdoiorg101021ie800006r
32 Yang S T White S A Hsu S T Extraction of carboxyl-ic acids with tertiary and quaternary amines Effect of pH Ind Eng Chem Res 30 (1991) 1335doi httpsdoiorg101021ie00054a040
33 Keshav A Wasewar K L Chand S Extraction of Acryl-ic Propionic and butyric acid using Aliquat 336 in oleyl alcohol Equilibria and effect of temperature Ind Eng Chem Res 48 (2009) 888doi httpsdoiorg101021ie8010337
34 Uslu H Kırbaslar S I Effect of temperature and initial acid concentration on the reactive extraction of carboxylic acids J Chem Eng Data 58 (2013) 1822doi httpsdoiorg101021je4002202
35 Pham D T Sagiroglu S Training multilayered percep-trons for pattern recognition A comparative study of four training algorithms Int J Mach Tools Manuf 41 (2001) 419doi httpsdoiorg101016S0890-6955(00)00073-0
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 37
(6)
1 ( )1 xlogsig x
eminus=+
( )purelin x x= (7)
Matlab software R2012a was used for ANN modelling The connections between the input layer and the hidden layer and the hidden layer and the output layer consist of weights and biases The weights which express the interactions between the neurons were used in two types of matrix ie input weight matrix (IW) connecting the input layer to the hidden layer and the layer weight matrix (LW) connecting the hidden layer to the output layer Each neuron in the first layer was connected to all the neurons in the next layer with a total of 40 connections between the input and hidden layers The bias determines the threshold to fire a neuron Each neuron has an individual bias value except for the input layer neurons30 The optimal number of neurons in the hidden layer was decided by the trial and error approach by varying it until a minimum mean squared error (MSE) was achieved MSE is an error function which measures the performance of the ANN model and can be calculated as fol-lows21
(8)
i N
ipred iexpi 1
( )MSE
E Esum
where N is the number of experimental data points Eipred and Eiexp are the ANN predicted and experi-mental responses respectively
To compare both constructed models an error analysis was carried out with experimental and the predicted response using coefficient of determina-tion (R2) root mean square error (RMSE) absolute average relative error (AARE) and standard error of prediction (SEP) calculated as follows
(9)
N 2ipred iexp2 i 1
2ipred avgi 1
( )1
( )N
E ER
E E=
=
minus= minus
minussumsum
(10)
N 2ipred iexpi=1
( )RMSE
NE Eminus
= sum
(11)
N ipred iexp
i=1iexp
100AAREN
E E
E
minus= sum
(12) avg
RMSESEP 100E
=
Where Eavg is the average value of the experi-mental response
Results and discussion
The gallic acid is extracted using TBP by form-ing a complex (GATBP) via interfacial reaction (Eq 13) The phosphoryl group of TBP forms a hy-drogen bond with the proton donor gallic acid to form a stable complex in the organic phase This can be represented as
aq org p m orgpGA mTBP (GA TBP )+ (13)
It is evident that the maximum GATBP com-plex formation results in maximum removal effi-ciency of GA that can be obtained only with a par-ticular combination of T pH0 CGA0 and CTBP However a large number of experiments may be required for identifying the suitable process condi-tions Therefore the RSM and ANN models were developed to identify the optimal process conditions required for maximum extraction efficiency of GA
RSM modelling of GA extraction
The design of experiments and experimental extraction efficiency E () are summarized in Table 1 The extraction efficiency for each experiment was calculated using Eq (1) Runs 25ndash28 (C1ndashC4) of centre point were used to determine the experi-mental error and reproducibility of the data The experimental data were regressed to obtain the anal-ysis of variance (ANOVA) as shown in Table 3 The model is highly significant and statistically fit with 9999 confidence level The empirical rela-tionship between the response and the independent variables is expressed in terms of coded variables as follows
(14)
1 2 32 2 2
4 1 2 324 1 2 1 3 1 4
2 3 2 4 3 4
9737 0017 346 051 ndash
051 113 151 011 ndash
0055 024 ndash 025 0011 ndash042 038 024
E x x x
x x x x
x x x x x x xx x x x x x
= + + minus
minus minus minus minus
minus + minus
+ + +
The positive and negative sign of the coeffi-cients indicate the linear increasing or decreasing effect of the corresponding variable on the response respectively The significance of a particular vari-able increases with the sum of square (SS) The ANOVA analysis (Table 3) suggests that CTBP (p = 00001 SS = 25336 F = 88228) has a most signif-icant effect on the response as compared to T (p = 00007 SS = 558 F = 1943) and pH0 (p = 00007 SS = 554 F = 1928) The linear coefficients were more significant than the quadratic or interactive coefficients
The degree of fitness of the model to the exper-imental data was determined by analysing the coef-ficient of determination (R2) The R2 obtained was 098 which indicates that the developed model
38 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
could well explain 98 of total variation in the response and confirms the goodness of fit for the RSM model The R2 (adj) and R2 (pred) values were 097 and 092 respectively which are closer to the R2 value of 0976 suggesting the significance and reasonable prediction ability of the model Ad-equate precision (signal to noise ratio) compares the predicted values with the average prediction error Adequate precision with a ratio greater than 4 is al-ways desirable The value of 3082 indicates an ad-equate signal The coefficient of variation (CV) was calculated to be 056 which indicates better pre-
cision and reliability of the experiments The Stan-dard deviation (S) with 054 indicates strong com-pliance with the predicted response The predicted residual sum of square (PRESS) is a measure of predictive power of the developed model The PRESS value of 2214 suggests that the regression model is significant to predict the response for a new experiment The coded terms of RSM model was then converted into actual variables and repre-sented as follows
(15)
02
0 02 2 2
0
0 0
0 0
0 0
9011 118( ) 029( )
011( ) 257(pH ) 026( )
00024( ) 00015( ) 011(pH )0014( )( ) 0043( )( )0022( )(pH ) 0002( )( )0022( )(pH ) 0041( )(pH )
GA TBP
GA
TBP
GA TBP GA
GA TBP
TBP
E C C
T C
C TC C C TC C TC T
= + + minus
minus minus minus minus
minus minus minus +
+ minus minus
minus + +
+ +
Response surface analysis process optimization and validation
Figures 3a and 3b illustrate the 3D surface plots for the individual and the interactive effects of independent variables on extraction efficiency In Figure 3a the concentration of extractant CTBP (x2) and its interaction with pH (x2 x4) show a positive effect on extraction efficiency (Table 3) whereas pH alone shows a negative effect A curvilinear ef-fect on extraction efficiency was observed due to the positive linear effect and negative quadratic ef-fect of CTBP This is attributed to the presence of the highly polar phosphoryl group (gtP=O) of TBP making it a strong Lewis base resulting in higher GATBP complex formation during the extraction process18 Hence the higher CTBP resulted in higher extraction The concentration of the undissociated acid always increases at lower pH due to the higher proton concentration TBP extracts only undissoci-ated form of acid31 hence extraction efficiency E () increases at lower pH values A similar effect of pH on the increase in efficiency was also observed in the extraction of other carboxylic acids32
Figure 3b illustrates that CTBP (x2) and T (x3) has a positive and negative effect respectively on ex-traction efficiency E () whereas the combined effect (x2 x3) shows a positive effect (Table 3) With an increase in temperature the formation of GATBP complex decreases The complex is formed via intermolecular hydrogen bonding through pro-ton transfer between acid and extractant The pro-cess of hydrogen bond formation is exothermic in nature which shifted the extraction equilibrium in backward direction at higher temperature thus re-ducing extraction efficiency Similar observations have been made by many researchers3334 Therefore it is concluded that lower temperature and pH with higher CTBP favour the extraction efficiency
Ta b l e 3 ndash Analysis of variance (ANOVA) for the experimen-tal results of the rotatable orthogonal central com-posite design
Source Coded terms
Coef- ficient
Sum of Squares F-value p-value Remarks
Model 3033 7544 00001 Significant
Constant 9737 00001
Linear
CGA0 x1 0017 000604 0021 0887
CTBP x2 346 25336 88228 00001 Significant
T x3 ndash051 558 1943 00007 Significant
pH0 x4 ndash051 554 1928 00007 Significant
Square
CGA0 CGA0 x12 ndash013 022 077 0395
CTBP CTBP x22 ndash151 3024 1053 00001 Significant
T T x32 ndash011 017 058 0461
pH0 pH0 x42 ndash0055 0041 014 0712
Interaction
CGA0 CTBP x1 x2 024 092 319 0097
CGA0 T x1 x3 ndash025 103 351 0081
CGA0 pH0 x1 x4 ndash0011 000181 000629 0940
CTBP T x2 x3 042 288 1003 0007 Significant
CTBP pH0 x2 x4 038 236 823 0013 Significant
T pH0 x3 x4 024 096 333 0091
Statistics
Adequate precision 3082
PRESS 2214
S 054
CV () 056
R2 098
R2(adj) 097
R2(pred) 092
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 39
Optimization of process variables CGA0 CTBP T and pH0 was performed using the numerical optimi-zation method to achieve optimal conditions to maximize the extraction The optimum variables were CGA0 = 234 g Lndash1 CTBP = 6565 (vv) T = 184 oC and pH0 = 18 with the predicted optimum extraction of 998 The experiment at optimum conditions was carried out in duplicate to validate the model predictions at the optimum conditions The experimental extraction efficiency obtained at the optimum conditions was found to be 9943 which is in close agreement with the model predict-
35
(a)
(b)
Fig 3 Three dimensional surface plot showing (a) the effect of CTBP and pH0 on the extraction
of GA at the centre level (T = 265 oC CGA0 = 14 g L-1) and (b) the effect of CTBP and T on
the extraction of GA at the centre level (pH0 = 19 CGA0 = 14 g L-1)
F i g 3 ndash Three dimensional surface plot showing (a) the ef-fect of CTBP and pH0 on the extraction of GA at the centre level (T = 265 oC CGA0 = 14 g Lndash1) and (b) the effect of CTBP and T on the extraction of GA at the centre level (pH0 = 19 CGA0 = 14 g Lndash1)
Ta b l e 4 ndash Additional experimental dataset used for the con-struction of ANN model
RunIndependent variables Response E ()
CGA0 (g Lndash1)
CTBP ( vv)
T (oC) pH0Experi- mental ANN Remark
29 05 65 11 196 9895 9868 Training
30 25 30 11 303 9488 9508 Training
31 14 30 11 302 9304 9325 Training
32 18 53 11 213 9728 9700 Training
33 25 75 11 305 9886 9851 Training
34 12 22 11 298 9013 9099 Validation
35 04 82 11 167 9723 9766 Training
36 028 15 11 225 8572 8704 Validation
37 10 30 15 308 9277 9243 Training
38 25 30 15 303 9473 9429 Validation
39 028 95 15 078 9730 9747 Training
40 14 30 16 301 9293 9305 Training
41 25 30 16 302 9449 9402 Training
42 25 30 18 303 9359 9347 Training
43 05 25 20 197 9076 9036 Training
44 14 30 20 302 9274 9212 Training
45 25 30 25 301 9124 9179 Training
46 028 15 25 078 8485 8482 Training
47 25 30 27 303 9263 9150 Validation
48 14 30 27 302 8768 8934 Training
49 04 85 27 139 9808 9796 Training
50 24 72 27 255 9866 9843 Training
51 12 24 27 22 8931 9044 Testing
52 10 62 28 172 9758 9826 Testing
53 15 72 28 165 9852 9831 Testing
54 25 30 28 303 9171 9136 Training
55 16 32 30 246 9224 9210 Training
56 18 78 30 261 9895 9852 Testing
57 18 27 30 261 9039 8930 Training
58 12 65 30 254 9806 9792 Training
59 25 95 30 078 9864 9879 Training
60 14 30 33 302 8798 8766 Training
61 25 30 35 303 9114 9062 Validation
62 25 30 35 3012 9038 9063 Training
63 14 30 37 302 8733 8729 Training
64 24 80 37 222 9859 9857 Training
65 08 60 37 218 9718 9694 Training
66 24 16 37 209 8515 8536 Training
67 16 32 37 178 9104 9102 Training
68 25 30 41 303 9099 9032 Training
69 10 30 41 150 8883 8890 Training
40 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
ed value In Table 1 Runs no 3 4 12 20 show the extraction efficiency above 992 at the cost of 80ndash95 TBP while the optimum condition gives the same efficiency with 65 TBP
ANN modelling of the GA extraction process
In ANN the three layered feed forward error back propagation (FF-EBP) with four neurons in the input layer (input parameters CGA0 CTBP T pH0) ten neurons in the hidden layer (optimum) one neuron in the output layer (extraction efficien-cy) was selected Figure 4 illustrates the ANN ar-chitecture applied for predicting the extraction effi-ciency of GA
A total of 69 experimental data-points (Table 1 and 4) were used to train the ANN model The input and the output data-points were normalized in the range 0 to 1 and 03 to 07 respectively This keeps the back propagation error within the limits and avoids over fitting of the data due to very large and very small values of weights and speeds up training time The data were normalized between the two points n1 and n0 using the following generalized equation (16)
minn 1 0 0
max min
( )(n n ) n( )
x xxx x
minus= minus + minus
where xn is normalized data-point and x xmin xmax are the actual minimum and maximum of inputoutput data The data were split into subsets of training (80 55 data-points) validation (10 7 data-points) and testing (10 7 data-points) Trainlm a training function that updates weights and bias values based on Levenberg-Marquardt backpropagation (LMB) algorithm was used for training the network LMB is an iterative technique which reduces the performance function in each it-eration and makes trainlm the fastest error back-propagation (EBP) algorithm for a moderate sized network35 The externally normalized input values were forwarded from the input layer to the hidden layer and then to the output layer to predict the re-sponse The error MSE was backpropagated from the output to the hidden layer and thereafter to the input layer to modify the weights (IW LW) and bi-ases (b1 b2) Figure 5 describes the general concept of FF-EBP algorithm used in ANN training to pre-dict extraction efficiency During the series of trials for EBP the training algorithm connection weights transfer function hidden layers and hidden neurons were varied and a network topology with an opti-mal network architecture (4101) was selected based on MSE value
The training was successfully terminated after 12 epochs (iterations) as the performance function
36
Fig 4 ANN architecture used for the prediction of GA extraction efficiency F i g 4 ndash ANN architecture used for the prediction of GA extraction efficiency
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 41
F i g 5 ndash Feed forward backpropagation algorithm used in the ANN training for the prediction of GA extraction efficiency
MSE reached a minimum value of 5middot10ndash4 which is lower than the set goal E0 = 1middot10ndash3 (Figure 6) The similar characteristic curve for the test and the vali-dation were observed suggesting no significant over-fitting The estimated MSE values for training (1middot10ndash4) testing (3middot10ndash4) and validation data (5middot10ndash4) were found to be lower than the set goal which is desirable Table 5 gives the optimal weights and bias Wherein IW = HtimesA b1 = Htimes1 LW =OtimesH ma-trix where H is hidden layer neurons (10) A is in-put layer neurons (4) and O is the output layer neu-ron (1) The bias b2 is the single value associated with a single neuron at the output
Figure 7 describes the ANN regression plot for training validation testing and the overall predic-tion set in the form of network output versus exper-imental extraction efficiency It can be observed that the network output values were close to the ex-
F i g 6 ndash MSE for training validation and test dataset com-puted using LMB (with performance of 5middot10ndash4 at 12 epochs and goal of 1middot10ndash3)
38
Fig 6 MSE for training validation and test dataset computed using LMB (with performance of
5middot10-4 at 12 epochs and goal of 1middot10-3)
42 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
perimental extraction efficiency in all the cases The correlation coefficients lsquoRrsquo for training validation and testing were 0995 0984 0989 respectively
whereas the overall prediction set was 0993 which confirms that the ANN model is satisfactory for in-terpolating the experimental data
Ta b l e 5 ndash Optimal values of weights and biases obtained from ANN training using Levenberg-Marquardt backpropagation algorithm
Input weight matrix IW (Weights to Hidden layer from inputs)
54485 74654 14374 2009934801 10038 71242 0667899763 46322 41336 7966216333 44955 0545 2141845359 42326 56151 16171
8043 27753 20607 0620637705 30939 60152 7216681829 118002 62271 3
minus minusminus minusminus minus
minus minus minusminus minus
minus minus 806645814 08091 62607 2537333117 47426 72301 08914
minusminus
Bias vector (Hidden layer) b1 102912 114372 94153 04321 26838 50968 73116 01728 4231 37916 Tminus minus minus minus
Layer weight vector LW (Weight to output layer from hidden layer)
01575 04295 00956 03633 02071 02006 01154 02645 02656 0313minus minus minus minus minus minus
Bias scalar (output layer) b2 06545
F i g 7 ndash ANN model showing the regression plots for training validation testing and overall prediction set
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 43
The ANN model was also validated using sta-tistical analysis ANOVA The degree of freedom due to regression (DFregression) and residuals (DFresidual) was calculated as follows20
regressionDF M 1= minus (17)
residualDF N M= minus (18)
M H(A O 1) O= + + + (19)
where M is the total number of network connec-tions including weights and biases N is the total number of experiments (N = 69) performed to con-struct and validate the ANN model Table 6 shows the ANOVA for the developed ANN model The F-value and the p-value were obtained from ANO-VA for the ANN model as 104 and 00004 respec-tively which confirms the significance of the ANN model Table 7 shows the values of R2 and R2 (adj) as 0986 and 088 respectively indicating the good fit of the ANN model to the experimental extraction efficiency Therefore the ANN model for predicting extraction efficiency E () for the extraction of GA can be written as follows
n n 1 2 (LW (IW ) )y purelin logsig x b b= times times + + (20)
Comparison of ANN and RSM models
The prediction and generalization capabilities of RSM and ANN models were evaluated based on its R2 RMSE AARE and SEP values Table 1 and 4 indicate that both the models predict well the ex-perimental extraction efficiency and their R2 values are closer to 1 Therefore both models can be con-sidered good for predicting the extraction efficiency of GA The generalization capabilities of both the models were compared only with unseen dataset Table 7 shows the unseen data set which were not used for the model development Figure 8 shows the comparative parity plot for RSM and ANN model predictability for the unseen data set and suggests that both the model predictions fit the ex-perimental data very well The comparative values of R2 RMSE AARE and SEP values for both mod-els using unseen data are shown in Table 8 Thus higher R2 and lower RMSE AARE SEP values for ANN indicate that the ANN model is superior over the RSM model for its generalization capability of the GA extraction process
Ta b l e 6 ndash Analysis of variance (ANOVA) analysis for ANN model
Source DFa SSb MSc F-value p-value R2 R2(adj)
Model or Regression 60 11421 1904 1041 00004 0986 088Residual Error 9 1646 183Total 69 11587aDegree of Freedom bSum of Squares cMean Square
Ta b l e 7 ndash Unseen experimental dataset used in the developed RSM and ANN models for the prediction of ex-traction efficiency of GA
RunIndependent variables Response E ()
CGA0 (g Lndash1)
CTBP ( vv)
T(oC) pH0Experi- mental RSM ANN
1 16 15 11 174 9057 9015 91402 25 30 40 30 9005 8836 90373 25 23 37 234 8888 8723 88534 25 37 27 211 9226 9323 92275 25 45 27 185 9525 9531 94786 25 23 37 234 8798 8723 88537 14 30 40 301 8533 8967 87368 18 20 32 221 8745 8805 86639 10 30 24 26 9256 9169 925710 18 36 32 21 9302 9285 938911 10 30 40 15 8883 9118 8908
Ta b l e 8 ndash Comparison of RSM and ANN models for its gen-eralization capability (using unseen data)
Parameters RSM ANN
R2 061 0915RMSE 173 080
AARE 143 066 SEP 191 088
40
Fig 8 Comparison of RSM and ANN model predictions for the unseen data (data not used in
the model development)
F i g 8 ndash Comparison of RSM and ANN model predictions for the unseen data (data not used in the model development)
44 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
Conclusion
The present work suggests that gallic acid can be efficiently removed from aqueous streams using TBP in hexanol The RSM and ANN models were developed to predict the extraction efficiency of GA by varying the parameters temperature concentra-tion of TBP in aqueous phase pH and initial con-centration of gallic acid in aqueous solution The importance of each parameter and its synergistic interactive effects on the extraction of GA was ex-plained with a minimum number of experiments us-ing the RSM model Both models were efficient in the prediction of GA extraction efficiency having close agreement with the experimental extraction efficiency However ANN showed superiority over RSM in terms of accuracy but the marginal errors between the models were very low The optimal conditions to achieve maximum extraction as pre-dicted by the RSM model were CGA0 = 234 g Lndash1 CTBP = 6565 (vv) T = 184 oC and pH0 = 18 at which an experimental extraction efficiency of 995 was observed The models also showed bet-ter performance with the use of unseen data which confirmed their generalization capabilities Thus the present investigation on experiments model-
ling and optimization of the GA extraction process could be of great significance for the design and evaluation of the performance of similar systems
S u p p l e m e n t a r y I n f o r m a t i o n
Supplementary information for confirmation of 2 h shaking time for reactive extraction of gallic acid and retention time of 39 min in HPLC analysis
The trial experiments were conducted for ob-taining an optimum shaking period Equal volumes (20 mL) of aqueous and organic phases were shak-en for different time intervals (5ndash240 min) at 250 rpm Equilibrium was achieved after half an hour (Figure S1) Hence further experiments were con-ducted with 2 h shaking time (sufficient to achieve equilibrium)
Calibration curve and chromatogram for HPLC analysis of gallic acid was obtained by diluting the mother stock of 5 g Lndash1 to prepare different concen-trations (5ndash300 mg Lndash1) of standard sample solu-tions
UV absorption spectra for gallic acid was ob-tained using UV Spectrophotometer Shimadzu UV-1800
F i g S 1 ndash Experimental extraction efficiency (E) of gallic acid versus time for the operating conditions CGA0 = 2 g Lndash1 CTBP = 40 (vv) T = 30 oC pH0 = 3
F i g S 2 ndash Calibration curve for gallic acid obtained us- ing HPLC
44
Fig S4b HPLC Chromatogram of gallic acid at 264 nm and 396 min
F i g S 3 ndash HPLC Chromatogram of gallic acid at 264 nm and 396 min
Ext
ract
ion
effic
ienc
y E
()
Peak
are
a (m
AU
s)
Time (min) Gallic acid concentration (mg Lndash1)
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 45
N o m e n c l a t u r e
AARE ndash absolute average relative errorCGAaq ndash equilibrium concentration of GA in aqueous
phase g Lndash1
CGAorg ndash equilibrium concentration of GA in organic phase g Lndash1
CGA0 ndash initial concentration of GA in aqueous phase g Lndash1
CTBP ndash TBP concentration in organic phase DF ndash degree of freedom in ANOVA analysisE ndash extraction efficiency response F-value ndash Fischer distribution value (ratio of variances)H A O ndash number of neurons in hidden layer input lay-
er and output layer respectivelyIW ndash input weight matrixk ndash number of input variables or factors in
RO-CCDlogsig ndash sigmoidal transfer function for neural net-
workLW ndash layer weight matrixM ndash number of network connections including
weights and biasesMS ndash mean square in ANOVA analysisMSE ndash mean squared errorN ndash total number of experiments to construct
ANN modeln0 ndash centre points in RO-CCDnc ndash corner points or factorial points in RO-CCDns ndash star or axial points in RO-CCDpH0 ndash initial pH of aqueous GA solutionPRESS ndash predicted residual sum of squarespurelin ndash linear transfer function for neural networkp-value ndash null-hypothesis testR ndash coefficient of correlation
R2 ndash coefficient of determinationR2 (adj) ndash adjusted R-squaredR2 (pred) ndash predicted R-squaredRMSE ndash root mean square errorSEP ndash standard error of predictionSS ndash sum of squares in ANOVA analysisT ndash extraction temperature oCw ndash connection weight between layersx1 x2 x3 x4 ndash coded levels of independent vaiables in
RSMxij yij bij ndash network input output and bias of jth neu-
ron to ith layer respectivelya ndash distance of each star point from centre in
RO-CCDbi bj bij ndash regression coefficients representing linear
quadratic and interactive effects in RSM
R e f e r e n c e s
1 Benitez F J Real F J Acero J L Leal A I Garcia C Gallic acid degradation in aqueous solutions by UVH2O2 treatment Fentonrsquos reagent and the photo-Fenton sys-tem J Hazard Mater 126 (2005) 31doi httpsdoiorg101016jjhazmat200504040
2 El-Gohary F A Badawy M I El-Khateeb M A El-Kalliny A S Integrated treatment of olive mill wastewater (OMW) by the combination of fentonrsquos reaction and anaer-obic treatment J Hazard Mater 162 (2009) 1536doi httpsdoiorg101016jjhazmat200806098
3 Herrero M Cifuentes A Ibanez E Sub- and supercriti-cal fluid extraction of functional ingredients from different natural sources plants food-by-products algae and mi-croalgae A Review Food Chem 98 (2006) 136doi httpsdoiorg101016jfoodchem200505058
4 Kougan G B Tabopda T Kuete V Verpoorte R Sim-ple phenols phenolic acids and related esters from the me-dicinal plants of africa in Kuet V (First Ed) Medicinal plant research in Africa Pharmacology and chemistry El-sevier Inc London 2013 225ndash249
45
UV absorption spectra for gallic acid was obtained using UV Spectrophotometer Shimadzu UV-1800
Fig S3 Absorption spectra of gallic acid showing max = 264 nm
--
F i g S 4 ndash Absorption spectra of gallic acid showing lmax = 264 nm
46 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
5 Kapellakis I E Tsagarakis K P Crowther J C Olive oil history production and by-product management Rev Environ Sci Biotechnol 7 (2008) 1doi httpsdoiorg101007s11157-007-9120-9
6 Calace N Nardi E Petronio B Pietroletti M Adsorp-tion of phenols by papermill sludges Environ Pollut 118 (2002) 315doi httpsdoiorg101016S0269-7491(01)00303-7
7 Fki I Allouche N Sayadi S The use of polyphenolic extract purified hydroxytyrosol and 34-dihydroxyphenyl acetic acid from olive mill wastewater for the stabilization of refined oils A potential alternative to synthetic antioxi-dants Food Chem 93 (2005) 197doi httpsdoiorg101016jfoodchem200409014
8 Ollis D F Pelizzetti E Serpone N Photocatalyzed de-struction of water contaminants Environ Sci Technol 25 (1991) 1522doi httpsdoiorg101021es00021a001
9 Masten S J Davies S H The use of ozonation to de-grade organic contaminants in wastewaters Environ Sci Technol 28 (1994) 180Adoi httpsdoiorg101021es00053a001
10 Spigno G Jauregi P Recovery of gallic acid with colloi-dal gas aphrons (CGA) Int J Food Eng 1 (2005) 1doi httpsdoiorg1022021556-37581038
11 Spigno G Dermiki M Pastori C Casanova F Jauregi P Recovery of gallic acid with colloidal gas aphrons gen-erated from a cationic surfactant Sep Purif Technol 71 (2010) 56doi httpsdoiorg101016jseppur200911002
12 Quici N Litter M I Heterogeneous photocatalytic degra-dation of gallic acid under different experimental condi-tions Photochem Photobiol Sci 8 (2009) 975doi httpsdoiorg101039b901904a
13 Singh M Jha A Kumar A Hettiarachchy N Rai A K Sharma D Influence of the solvents on the extraction of major phenolic compounds (punicalagin ellagic acid and gallic acid) and their antioxidant activities in pomegranate aril J Food Sci Technol 51 (2014) 2070doi httpsdoiorg101007s13197-014-1267-0
14 Claudio A F M Ferreira A M Freire C S R Silves-tre A J D Freire M G Coutinho J A P Optimization of the gallic acid extraction using ionic-liquid-based aque-ous two-phase systems Sep Purif Technol 97 (2012) 142doi httpsdoiorg101016jseppur201202036
15 Li Y Wang Y Li Y Dai Y Extraction of glyoxylic acid glycolic acid acrylic acid and benzoic acid with trialkyl-phosphine oxide J Chem Eng Data 48 (2003) 621doi httpsdoiorg101021je020174r
16 Daneshfar A Ghaziaskar H S Homayoun N Solubility of gallic acid in methanol ethanol water and ethyl acetate J Chem Eng Data 53 (2008) 776doi httpsdoiorg101021je700633w
17 Wasewar K L Shende D Z Reactive extraction of caproic acid using tri- n -butyl phosphate in hexanol octa-nol and decanol J Chem Eng Data 56 (2011) 288doi httpsdoiorg101021je100974f
18 Keshav A Chand S Wasewar K L Reactive extraction of acrylic acid using tri- n -butyl phosphate in different di-luents J Chem Eng Data 54 (2009) 1782doi httpsdoiorg101021je800856e
19 Zhong K Wang Q Optimization of ultrasonic extraction of polysaccharides from dried longan pulp using response surface methodology Carbohydr Polym 80 (2010) 19doi httpsdoiorg101016jcarbpol200910066
20 Marchitan N Cojocaru C Mereuta A Duca G Crete-scu I Gonta M Modeling and optimization of tartaric acid reactive extraction from aqueous solutions A compar-ison between response surface methodology and artificial neural network Sep Purif Technol 75 (2010) 273doi httpsdoiorg101016jseppur201008016
21 Messikh N Samar M H Messikh L Neural network analysis of liquidndashliquid extraction of phenol from waste-water using tbp solvent Desalination 208 (2007) 42doi httpsdoiorg101016jdesal200604073
22 Montgomery D Design and analysis of experiments (Fifth Ed) John Wiley amp Sons Inc New Jersey 2001
23 Box G E P Hunter J S H W G Statistics for experi-menters design innovation and discovery (second ed) Vol 48 John Wiley ampSons New Jersey 2006
24 Dornier M Decloux M Trystram G Lebert A Dynam-ic modeling of crossflow microfiltration using neural net-works J Memb Sci 98 (1995) 263doi httpsdoiorg1010160376-7388(94)00195-5
25 Niemi H Bulsari A Palosaari S Simulation of mem-brane separation by neural networks J Memb Sci 102 (1995) 185doi httpsdoiorg1010160376-7388(94)00314-O
26 Bourquin J Schmidli H van Hoogevest P Leuenberger H Pitfalls of artificial neural networks (ANN) modelling technique for data sets containing outlier measurements us-ing a study on mixture properties of a direct compressed dosage form Eur J Pharm Sci 7 (1998) 17doi httpsdoiorg101016S0928-0987(97)10027-6
27 Cornell A I Khuri J A Response surfaces Designs and analyses (second ed) Marcel Dekker Inc New York 1996
28 Bezerra M A Santelli R E Oliveira E P Villar L S Escaleira L A Response surface methodology (RSM) as a tool for optimization in analytical chemistry Talanta 76 (2008) 965doi httpsdoiorg101016jtalanta200805019
29 Hornik K Stinchcombe M White H Multilayer feedfor-ward networks are universal approximators Neural Net-works 2 (1989) 359doi httpsdoiorg1010160893-6080(89)90020-8
30 Hagan M T Demuth H B Beale M H Neural network design PWS Publishing Company Boston 1995
31 Keshav A Wasewar K L Chand S Extraction of propi-onic acid using different extractants (tri-n-butylphosphate tri-n-octylamine and Aliquat 336) Ind Eng Chem Res 47 (2008) 6192doi httpsdoiorg101021ie800006r
32 Yang S T White S A Hsu S T Extraction of carboxyl-ic acids with tertiary and quaternary amines Effect of pH Ind Eng Chem Res 30 (1991) 1335doi httpsdoiorg101021ie00054a040
33 Keshav A Wasewar K L Chand S Extraction of Acryl-ic Propionic and butyric acid using Aliquat 336 in oleyl alcohol Equilibria and effect of temperature Ind Eng Chem Res 48 (2009) 888doi httpsdoiorg101021ie8010337
34 Uslu H Kırbaslar S I Effect of temperature and initial acid concentration on the reactive extraction of carboxylic acids J Chem Eng Data 58 (2013) 1822doi httpsdoiorg101021je4002202
35 Pham D T Sagiroglu S Training multilayered percep-trons for pattern recognition A comparative study of four training algorithms Int J Mach Tools Manuf 41 (2001) 419doi httpsdoiorg101016S0890-6955(00)00073-0
38 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
could well explain 98 of total variation in the response and confirms the goodness of fit for the RSM model The R2 (adj) and R2 (pred) values were 097 and 092 respectively which are closer to the R2 value of 0976 suggesting the significance and reasonable prediction ability of the model Ad-equate precision (signal to noise ratio) compares the predicted values with the average prediction error Adequate precision with a ratio greater than 4 is al-ways desirable The value of 3082 indicates an ad-equate signal The coefficient of variation (CV) was calculated to be 056 which indicates better pre-
cision and reliability of the experiments The Stan-dard deviation (S) with 054 indicates strong com-pliance with the predicted response The predicted residual sum of square (PRESS) is a measure of predictive power of the developed model The PRESS value of 2214 suggests that the regression model is significant to predict the response for a new experiment The coded terms of RSM model was then converted into actual variables and repre-sented as follows
(15)
02
0 02 2 2
0
0 0
0 0
0 0
9011 118( ) 029( )
011( ) 257(pH ) 026( )
00024( ) 00015( ) 011(pH )0014( )( ) 0043( )( )0022( )(pH ) 0002( )( )0022( )(pH ) 0041( )(pH )
GA TBP
GA
TBP
GA TBP GA
GA TBP
TBP
E C C
T C
C TC C C TC C TC T
= + + minus
minus minus minus minus
minus minus minus +
+ minus minus
minus + +
+ +
Response surface analysis process optimization and validation
Figures 3a and 3b illustrate the 3D surface plots for the individual and the interactive effects of independent variables on extraction efficiency In Figure 3a the concentration of extractant CTBP (x2) and its interaction with pH (x2 x4) show a positive effect on extraction efficiency (Table 3) whereas pH alone shows a negative effect A curvilinear ef-fect on extraction efficiency was observed due to the positive linear effect and negative quadratic ef-fect of CTBP This is attributed to the presence of the highly polar phosphoryl group (gtP=O) of TBP making it a strong Lewis base resulting in higher GATBP complex formation during the extraction process18 Hence the higher CTBP resulted in higher extraction The concentration of the undissociated acid always increases at lower pH due to the higher proton concentration TBP extracts only undissoci-ated form of acid31 hence extraction efficiency E () increases at lower pH values A similar effect of pH on the increase in efficiency was also observed in the extraction of other carboxylic acids32
Figure 3b illustrates that CTBP (x2) and T (x3) has a positive and negative effect respectively on ex-traction efficiency E () whereas the combined effect (x2 x3) shows a positive effect (Table 3) With an increase in temperature the formation of GATBP complex decreases The complex is formed via intermolecular hydrogen bonding through pro-ton transfer between acid and extractant The pro-cess of hydrogen bond formation is exothermic in nature which shifted the extraction equilibrium in backward direction at higher temperature thus re-ducing extraction efficiency Similar observations have been made by many researchers3334 Therefore it is concluded that lower temperature and pH with higher CTBP favour the extraction efficiency
Ta b l e 3 ndash Analysis of variance (ANOVA) for the experimen-tal results of the rotatable orthogonal central com-posite design
Source Coded terms
Coef- ficient
Sum of Squares F-value p-value Remarks
Model 3033 7544 00001 Significant
Constant 9737 00001
Linear
CGA0 x1 0017 000604 0021 0887
CTBP x2 346 25336 88228 00001 Significant
T x3 ndash051 558 1943 00007 Significant
pH0 x4 ndash051 554 1928 00007 Significant
Square
CGA0 CGA0 x12 ndash013 022 077 0395
CTBP CTBP x22 ndash151 3024 1053 00001 Significant
T T x32 ndash011 017 058 0461
pH0 pH0 x42 ndash0055 0041 014 0712
Interaction
CGA0 CTBP x1 x2 024 092 319 0097
CGA0 T x1 x3 ndash025 103 351 0081
CGA0 pH0 x1 x4 ndash0011 000181 000629 0940
CTBP T x2 x3 042 288 1003 0007 Significant
CTBP pH0 x2 x4 038 236 823 0013 Significant
T pH0 x3 x4 024 096 333 0091
Statistics
Adequate precision 3082
PRESS 2214
S 054
CV () 056
R2 098
R2(adj) 097
R2(pred) 092
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 39
Optimization of process variables CGA0 CTBP T and pH0 was performed using the numerical optimi-zation method to achieve optimal conditions to maximize the extraction The optimum variables were CGA0 = 234 g Lndash1 CTBP = 6565 (vv) T = 184 oC and pH0 = 18 with the predicted optimum extraction of 998 The experiment at optimum conditions was carried out in duplicate to validate the model predictions at the optimum conditions The experimental extraction efficiency obtained at the optimum conditions was found to be 9943 which is in close agreement with the model predict-
35
(a)
(b)
Fig 3 Three dimensional surface plot showing (a) the effect of CTBP and pH0 on the extraction
of GA at the centre level (T = 265 oC CGA0 = 14 g L-1) and (b) the effect of CTBP and T on
the extraction of GA at the centre level (pH0 = 19 CGA0 = 14 g L-1)
F i g 3 ndash Three dimensional surface plot showing (a) the ef-fect of CTBP and pH0 on the extraction of GA at the centre level (T = 265 oC CGA0 = 14 g Lndash1) and (b) the effect of CTBP and T on the extraction of GA at the centre level (pH0 = 19 CGA0 = 14 g Lndash1)
Ta b l e 4 ndash Additional experimental dataset used for the con-struction of ANN model
RunIndependent variables Response E ()
CGA0 (g Lndash1)
CTBP ( vv)
T (oC) pH0Experi- mental ANN Remark
29 05 65 11 196 9895 9868 Training
30 25 30 11 303 9488 9508 Training
31 14 30 11 302 9304 9325 Training
32 18 53 11 213 9728 9700 Training
33 25 75 11 305 9886 9851 Training
34 12 22 11 298 9013 9099 Validation
35 04 82 11 167 9723 9766 Training
36 028 15 11 225 8572 8704 Validation
37 10 30 15 308 9277 9243 Training
38 25 30 15 303 9473 9429 Validation
39 028 95 15 078 9730 9747 Training
40 14 30 16 301 9293 9305 Training
41 25 30 16 302 9449 9402 Training
42 25 30 18 303 9359 9347 Training
43 05 25 20 197 9076 9036 Training
44 14 30 20 302 9274 9212 Training
45 25 30 25 301 9124 9179 Training
46 028 15 25 078 8485 8482 Training
47 25 30 27 303 9263 9150 Validation
48 14 30 27 302 8768 8934 Training
49 04 85 27 139 9808 9796 Training
50 24 72 27 255 9866 9843 Training
51 12 24 27 22 8931 9044 Testing
52 10 62 28 172 9758 9826 Testing
53 15 72 28 165 9852 9831 Testing
54 25 30 28 303 9171 9136 Training
55 16 32 30 246 9224 9210 Training
56 18 78 30 261 9895 9852 Testing
57 18 27 30 261 9039 8930 Training
58 12 65 30 254 9806 9792 Training
59 25 95 30 078 9864 9879 Training
60 14 30 33 302 8798 8766 Training
61 25 30 35 303 9114 9062 Validation
62 25 30 35 3012 9038 9063 Training
63 14 30 37 302 8733 8729 Training
64 24 80 37 222 9859 9857 Training
65 08 60 37 218 9718 9694 Training
66 24 16 37 209 8515 8536 Training
67 16 32 37 178 9104 9102 Training
68 25 30 41 303 9099 9032 Training
69 10 30 41 150 8883 8890 Training
40 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
ed value In Table 1 Runs no 3 4 12 20 show the extraction efficiency above 992 at the cost of 80ndash95 TBP while the optimum condition gives the same efficiency with 65 TBP
ANN modelling of the GA extraction process
In ANN the three layered feed forward error back propagation (FF-EBP) with four neurons in the input layer (input parameters CGA0 CTBP T pH0) ten neurons in the hidden layer (optimum) one neuron in the output layer (extraction efficien-cy) was selected Figure 4 illustrates the ANN ar-chitecture applied for predicting the extraction effi-ciency of GA
A total of 69 experimental data-points (Table 1 and 4) were used to train the ANN model The input and the output data-points were normalized in the range 0 to 1 and 03 to 07 respectively This keeps the back propagation error within the limits and avoids over fitting of the data due to very large and very small values of weights and speeds up training time The data were normalized between the two points n1 and n0 using the following generalized equation (16)
minn 1 0 0
max min
( )(n n ) n( )
x xxx x
minus= minus + minus
where xn is normalized data-point and x xmin xmax are the actual minimum and maximum of inputoutput data The data were split into subsets of training (80 55 data-points) validation (10 7 data-points) and testing (10 7 data-points) Trainlm a training function that updates weights and bias values based on Levenberg-Marquardt backpropagation (LMB) algorithm was used for training the network LMB is an iterative technique which reduces the performance function in each it-eration and makes trainlm the fastest error back-propagation (EBP) algorithm for a moderate sized network35 The externally normalized input values were forwarded from the input layer to the hidden layer and then to the output layer to predict the re-sponse The error MSE was backpropagated from the output to the hidden layer and thereafter to the input layer to modify the weights (IW LW) and bi-ases (b1 b2) Figure 5 describes the general concept of FF-EBP algorithm used in ANN training to pre-dict extraction efficiency During the series of trials for EBP the training algorithm connection weights transfer function hidden layers and hidden neurons were varied and a network topology with an opti-mal network architecture (4101) was selected based on MSE value
The training was successfully terminated after 12 epochs (iterations) as the performance function
36
Fig 4 ANN architecture used for the prediction of GA extraction efficiency F i g 4 ndash ANN architecture used for the prediction of GA extraction efficiency
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 41
F i g 5 ndash Feed forward backpropagation algorithm used in the ANN training for the prediction of GA extraction efficiency
MSE reached a minimum value of 5middot10ndash4 which is lower than the set goal E0 = 1middot10ndash3 (Figure 6) The similar characteristic curve for the test and the vali-dation were observed suggesting no significant over-fitting The estimated MSE values for training (1middot10ndash4) testing (3middot10ndash4) and validation data (5middot10ndash4) were found to be lower than the set goal which is desirable Table 5 gives the optimal weights and bias Wherein IW = HtimesA b1 = Htimes1 LW =OtimesH ma-trix where H is hidden layer neurons (10) A is in-put layer neurons (4) and O is the output layer neu-ron (1) The bias b2 is the single value associated with a single neuron at the output
Figure 7 describes the ANN regression plot for training validation testing and the overall predic-tion set in the form of network output versus exper-imental extraction efficiency It can be observed that the network output values were close to the ex-
F i g 6 ndash MSE for training validation and test dataset com-puted using LMB (with performance of 5middot10ndash4 at 12 epochs and goal of 1middot10ndash3)
38
Fig 6 MSE for training validation and test dataset computed using LMB (with performance of
5middot10-4 at 12 epochs and goal of 1middot10-3)
42 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
perimental extraction efficiency in all the cases The correlation coefficients lsquoRrsquo for training validation and testing were 0995 0984 0989 respectively
whereas the overall prediction set was 0993 which confirms that the ANN model is satisfactory for in-terpolating the experimental data
Ta b l e 5 ndash Optimal values of weights and biases obtained from ANN training using Levenberg-Marquardt backpropagation algorithm
Input weight matrix IW (Weights to Hidden layer from inputs)
54485 74654 14374 2009934801 10038 71242 0667899763 46322 41336 7966216333 44955 0545 2141845359 42326 56151 16171
8043 27753 20607 0620637705 30939 60152 7216681829 118002 62271 3
minus minusminus minusminus minus
minus minus minusminus minus
minus minus 806645814 08091 62607 2537333117 47426 72301 08914
minusminus
Bias vector (Hidden layer) b1 102912 114372 94153 04321 26838 50968 73116 01728 4231 37916 Tminus minus minus minus
Layer weight vector LW (Weight to output layer from hidden layer)
01575 04295 00956 03633 02071 02006 01154 02645 02656 0313minus minus minus minus minus minus
Bias scalar (output layer) b2 06545
F i g 7 ndash ANN model showing the regression plots for training validation testing and overall prediction set
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 43
The ANN model was also validated using sta-tistical analysis ANOVA The degree of freedom due to regression (DFregression) and residuals (DFresidual) was calculated as follows20
regressionDF M 1= minus (17)
residualDF N M= minus (18)
M H(A O 1) O= + + + (19)
where M is the total number of network connec-tions including weights and biases N is the total number of experiments (N = 69) performed to con-struct and validate the ANN model Table 6 shows the ANOVA for the developed ANN model The F-value and the p-value were obtained from ANO-VA for the ANN model as 104 and 00004 respec-tively which confirms the significance of the ANN model Table 7 shows the values of R2 and R2 (adj) as 0986 and 088 respectively indicating the good fit of the ANN model to the experimental extraction efficiency Therefore the ANN model for predicting extraction efficiency E () for the extraction of GA can be written as follows
n n 1 2 (LW (IW ) )y purelin logsig x b b= times times + + (20)
Comparison of ANN and RSM models
The prediction and generalization capabilities of RSM and ANN models were evaluated based on its R2 RMSE AARE and SEP values Table 1 and 4 indicate that both the models predict well the ex-perimental extraction efficiency and their R2 values are closer to 1 Therefore both models can be con-sidered good for predicting the extraction efficiency of GA The generalization capabilities of both the models were compared only with unseen dataset Table 7 shows the unseen data set which were not used for the model development Figure 8 shows the comparative parity plot for RSM and ANN model predictability for the unseen data set and suggests that both the model predictions fit the ex-perimental data very well The comparative values of R2 RMSE AARE and SEP values for both mod-els using unseen data are shown in Table 8 Thus higher R2 and lower RMSE AARE SEP values for ANN indicate that the ANN model is superior over the RSM model for its generalization capability of the GA extraction process
Ta b l e 6 ndash Analysis of variance (ANOVA) analysis for ANN model
Source DFa SSb MSc F-value p-value R2 R2(adj)
Model or Regression 60 11421 1904 1041 00004 0986 088Residual Error 9 1646 183Total 69 11587aDegree of Freedom bSum of Squares cMean Square
Ta b l e 7 ndash Unseen experimental dataset used in the developed RSM and ANN models for the prediction of ex-traction efficiency of GA
RunIndependent variables Response E ()
CGA0 (g Lndash1)
CTBP ( vv)
T(oC) pH0Experi- mental RSM ANN
1 16 15 11 174 9057 9015 91402 25 30 40 30 9005 8836 90373 25 23 37 234 8888 8723 88534 25 37 27 211 9226 9323 92275 25 45 27 185 9525 9531 94786 25 23 37 234 8798 8723 88537 14 30 40 301 8533 8967 87368 18 20 32 221 8745 8805 86639 10 30 24 26 9256 9169 925710 18 36 32 21 9302 9285 938911 10 30 40 15 8883 9118 8908
Ta b l e 8 ndash Comparison of RSM and ANN models for its gen-eralization capability (using unseen data)
Parameters RSM ANN
R2 061 0915RMSE 173 080
AARE 143 066 SEP 191 088
40
Fig 8 Comparison of RSM and ANN model predictions for the unseen data (data not used in
the model development)
F i g 8 ndash Comparison of RSM and ANN model predictions for the unseen data (data not used in the model development)
44 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
Conclusion
The present work suggests that gallic acid can be efficiently removed from aqueous streams using TBP in hexanol The RSM and ANN models were developed to predict the extraction efficiency of GA by varying the parameters temperature concentra-tion of TBP in aqueous phase pH and initial con-centration of gallic acid in aqueous solution The importance of each parameter and its synergistic interactive effects on the extraction of GA was ex-plained with a minimum number of experiments us-ing the RSM model Both models were efficient in the prediction of GA extraction efficiency having close agreement with the experimental extraction efficiency However ANN showed superiority over RSM in terms of accuracy but the marginal errors between the models were very low The optimal conditions to achieve maximum extraction as pre-dicted by the RSM model were CGA0 = 234 g Lndash1 CTBP = 6565 (vv) T = 184 oC and pH0 = 18 at which an experimental extraction efficiency of 995 was observed The models also showed bet-ter performance with the use of unseen data which confirmed their generalization capabilities Thus the present investigation on experiments model-
ling and optimization of the GA extraction process could be of great significance for the design and evaluation of the performance of similar systems
S u p p l e m e n t a r y I n f o r m a t i o n
Supplementary information for confirmation of 2 h shaking time for reactive extraction of gallic acid and retention time of 39 min in HPLC analysis
The trial experiments were conducted for ob-taining an optimum shaking period Equal volumes (20 mL) of aqueous and organic phases were shak-en for different time intervals (5ndash240 min) at 250 rpm Equilibrium was achieved after half an hour (Figure S1) Hence further experiments were con-ducted with 2 h shaking time (sufficient to achieve equilibrium)
Calibration curve and chromatogram for HPLC analysis of gallic acid was obtained by diluting the mother stock of 5 g Lndash1 to prepare different concen-trations (5ndash300 mg Lndash1) of standard sample solu-tions
UV absorption spectra for gallic acid was ob-tained using UV Spectrophotometer Shimadzu UV-1800
F i g S 1 ndash Experimental extraction efficiency (E) of gallic acid versus time for the operating conditions CGA0 = 2 g Lndash1 CTBP = 40 (vv) T = 30 oC pH0 = 3
F i g S 2 ndash Calibration curve for gallic acid obtained us- ing HPLC
44
Fig S4b HPLC Chromatogram of gallic acid at 264 nm and 396 min
F i g S 3 ndash HPLC Chromatogram of gallic acid at 264 nm and 396 min
Ext
ract
ion
effic
ienc
y E
()
Peak
are
a (m
AU
s)
Time (min) Gallic acid concentration (mg Lndash1)
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 45
N o m e n c l a t u r e
AARE ndash absolute average relative errorCGAaq ndash equilibrium concentration of GA in aqueous
phase g Lndash1
CGAorg ndash equilibrium concentration of GA in organic phase g Lndash1
CGA0 ndash initial concentration of GA in aqueous phase g Lndash1
CTBP ndash TBP concentration in organic phase DF ndash degree of freedom in ANOVA analysisE ndash extraction efficiency response F-value ndash Fischer distribution value (ratio of variances)H A O ndash number of neurons in hidden layer input lay-
er and output layer respectivelyIW ndash input weight matrixk ndash number of input variables or factors in
RO-CCDlogsig ndash sigmoidal transfer function for neural net-
workLW ndash layer weight matrixM ndash number of network connections including
weights and biasesMS ndash mean square in ANOVA analysisMSE ndash mean squared errorN ndash total number of experiments to construct
ANN modeln0 ndash centre points in RO-CCDnc ndash corner points or factorial points in RO-CCDns ndash star or axial points in RO-CCDpH0 ndash initial pH of aqueous GA solutionPRESS ndash predicted residual sum of squarespurelin ndash linear transfer function for neural networkp-value ndash null-hypothesis testR ndash coefficient of correlation
R2 ndash coefficient of determinationR2 (adj) ndash adjusted R-squaredR2 (pred) ndash predicted R-squaredRMSE ndash root mean square errorSEP ndash standard error of predictionSS ndash sum of squares in ANOVA analysisT ndash extraction temperature oCw ndash connection weight between layersx1 x2 x3 x4 ndash coded levels of independent vaiables in
RSMxij yij bij ndash network input output and bias of jth neu-
ron to ith layer respectivelya ndash distance of each star point from centre in
RO-CCDbi bj bij ndash regression coefficients representing linear
quadratic and interactive effects in RSM
R e f e r e n c e s
1 Benitez F J Real F J Acero J L Leal A I Garcia C Gallic acid degradation in aqueous solutions by UVH2O2 treatment Fentonrsquos reagent and the photo-Fenton sys-tem J Hazard Mater 126 (2005) 31doi httpsdoiorg101016jjhazmat200504040
2 El-Gohary F A Badawy M I El-Khateeb M A El-Kalliny A S Integrated treatment of olive mill wastewater (OMW) by the combination of fentonrsquos reaction and anaer-obic treatment J Hazard Mater 162 (2009) 1536doi httpsdoiorg101016jjhazmat200806098
3 Herrero M Cifuentes A Ibanez E Sub- and supercriti-cal fluid extraction of functional ingredients from different natural sources plants food-by-products algae and mi-croalgae A Review Food Chem 98 (2006) 136doi httpsdoiorg101016jfoodchem200505058
4 Kougan G B Tabopda T Kuete V Verpoorte R Sim-ple phenols phenolic acids and related esters from the me-dicinal plants of africa in Kuet V (First Ed) Medicinal plant research in Africa Pharmacology and chemistry El-sevier Inc London 2013 225ndash249
45
UV absorption spectra for gallic acid was obtained using UV Spectrophotometer Shimadzu UV-1800
Fig S3 Absorption spectra of gallic acid showing max = 264 nm
--
F i g S 4 ndash Absorption spectra of gallic acid showing lmax = 264 nm
46 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
5 Kapellakis I E Tsagarakis K P Crowther J C Olive oil history production and by-product management Rev Environ Sci Biotechnol 7 (2008) 1doi httpsdoiorg101007s11157-007-9120-9
6 Calace N Nardi E Petronio B Pietroletti M Adsorp-tion of phenols by papermill sludges Environ Pollut 118 (2002) 315doi httpsdoiorg101016S0269-7491(01)00303-7
7 Fki I Allouche N Sayadi S The use of polyphenolic extract purified hydroxytyrosol and 34-dihydroxyphenyl acetic acid from olive mill wastewater for the stabilization of refined oils A potential alternative to synthetic antioxi-dants Food Chem 93 (2005) 197doi httpsdoiorg101016jfoodchem200409014
8 Ollis D F Pelizzetti E Serpone N Photocatalyzed de-struction of water contaminants Environ Sci Technol 25 (1991) 1522doi httpsdoiorg101021es00021a001
9 Masten S J Davies S H The use of ozonation to de-grade organic contaminants in wastewaters Environ Sci Technol 28 (1994) 180Adoi httpsdoiorg101021es00053a001
10 Spigno G Jauregi P Recovery of gallic acid with colloi-dal gas aphrons (CGA) Int J Food Eng 1 (2005) 1doi httpsdoiorg1022021556-37581038
11 Spigno G Dermiki M Pastori C Casanova F Jauregi P Recovery of gallic acid with colloidal gas aphrons gen-erated from a cationic surfactant Sep Purif Technol 71 (2010) 56doi httpsdoiorg101016jseppur200911002
12 Quici N Litter M I Heterogeneous photocatalytic degra-dation of gallic acid under different experimental condi-tions Photochem Photobiol Sci 8 (2009) 975doi httpsdoiorg101039b901904a
13 Singh M Jha A Kumar A Hettiarachchy N Rai A K Sharma D Influence of the solvents on the extraction of major phenolic compounds (punicalagin ellagic acid and gallic acid) and their antioxidant activities in pomegranate aril J Food Sci Technol 51 (2014) 2070doi httpsdoiorg101007s13197-014-1267-0
14 Claudio A F M Ferreira A M Freire C S R Silves-tre A J D Freire M G Coutinho J A P Optimization of the gallic acid extraction using ionic-liquid-based aque-ous two-phase systems Sep Purif Technol 97 (2012) 142doi httpsdoiorg101016jseppur201202036
15 Li Y Wang Y Li Y Dai Y Extraction of glyoxylic acid glycolic acid acrylic acid and benzoic acid with trialkyl-phosphine oxide J Chem Eng Data 48 (2003) 621doi httpsdoiorg101021je020174r
16 Daneshfar A Ghaziaskar H S Homayoun N Solubility of gallic acid in methanol ethanol water and ethyl acetate J Chem Eng Data 53 (2008) 776doi httpsdoiorg101021je700633w
17 Wasewar K L Shende D Z Reactive extraction of caproic acid using tri- n -butyl phosphate in hexanol octa-nol and decanol J Chem Eng Data 56 (2011) 288doi httpsdoiorg101021je100974f
18 Keshav A Chand S Wasewar K L Reactive extraction of acrylic acid using tri- n -butyl phosphate in different di-luents J Chem Eng Data 54 (2009) 1782doi httpsdoiorg101021je800856e
19 Zhong K Wang Q Optimization of ultrasonic extraction of polysaccharides from dried longan pulp using response surface methodology Carbohydr Polym 80 (2010) 19doi httpsdoiorg101016jcarbpol200910066
20 Marchitan N Cojocaru C Mereuta A Duca G Crete-scu I Gonta M Modeling and optimization of tartaric acid reactive extraction from aqueous solutions A compar-ison between response surface methodology and artificial neural network Sep Purif Technol 75 (2010) 273doi httpsdoiorg101016jseppur201008016
21 Messikh N Samar M H Messikh L Neural network analysis of liquidndashliquid extraction of phenol from waste-water using tbp solvent Desalination 208 (2007) 42doi httpsdoiorg101016jdesal200604073
22 Montgomery D Design and analysis of experiments (Fifth Ed) John Wiley amp Sons Inc New Jersey 2001
23 Box G E P Hunter J S H W G Statistics for experi-menters design innovation and discovery (second ed) Vol 48 John Wiley ampSons New Jersey 2006
24 Dornier M Decloux M Trystram G Lebert A Dynam-ic modeling of crossflow microfiltration using neural net-works J Memb Sci 98 (1995) 263doi httpsdoiorg1010160376-7388(94)00195-5
25 Niemi H Bulsari A Palosaari S Simulation of mem-brane separation by neural networks J Memb Sci 102 (1995) 185doi httpsdoiorg1010160376-7388(94)00314-O
26 Bourquin J Schmidli H van Hoogevest P Leuenberger H Pitfalls of artificial neural networks (ANN) modelling technique for data sets containing outlier measurements us-ing a study on mixture properties of a direct compressed dosage form Eur J Pharm Sci 7 (1998) 17doi httpsdoiorg101016S0928-0987(97)10027-6
27 Cornell A I Khuri J A Response surfaces Designs and analyses (second ed) Marcel Dekker Inc New York 1996
28 Bezerra M A Santelli R E Oliveira E P Villar L S Escaleira L A Response surface methodology (RSM) as a tool for optimization in analytical chemistry Talanta 76 (2008) 965doi httpsdoiorg101016jtalanta200805019
29 Hornik K Stinchcombe M White H Multilayer feedfor-ward networks are universal approximators Neural Net-works 2 (1989) 359doi httpsdoiorg1010160893-6080(89)90020-8
30 Hagan M T Demuth H B Beale M H Neural network design PWS Publishing Company Boston 1995
31 Keshav A Wasewar K L Chand S Extraction of propi-onic acid using different extractants (tri-n-butylphosphate tri-n-octylamine and Aliquat 336) Ind Eng Chem Res 47 (2008) 6192doi httpsdoiorg101021ie800006r
32 Yang S T White S A Hsu S T Extraction of carboxyl-ic acids with tertiary and quaternary amines Effect of pH Ind Eng Chem Res 30 (1991) 1335doi httpsdoiorg101021ie00054a040
33 Keshav A Wasewar K L Chand S Extraction of Acryl-ic Propionic and butyric acid using Aliquat 336 in oleyl alcohol Equilibria and effect of temperature Ind Eng Chem Res 48 (2009) 888doi httpsdoiorg101021ie8010337
34 Uslu H Kırbaslar S I Effect of temperature and initial acid concentration on the reactive extraction of carboxylic acids J Chem Eng Data 58 (2013) 1822doi httpsdoiorg101021je4002202
35 Pham D T Sagiroglu S Training multilayered percep-trons for pattern recognition A comparative study of four training algorithms Int J Mach Tools Manuf 41 (2001) 419doi httpsdoiorg101016S0890-6955(00)00073-0
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 39
Optimization of process variables CGA0 CTBP T and pH0 was performed using the numerical optimi-zation method to achieve optimal conditions to maximize the extraction The optimum variables were CGA0 = 234 g Lndash1 CTBP = 6565 (vv) T = 184 oC and pH0 = 18 with the predicted optimum extraction of 998 The experiment at optimum conditions was carried out in duplicate to validate the model predictions at the optimum conditions The experimental extraction efficiency obtained at the optimum conditions was found to be 9943 which is in close agreement with the model predict-
35
(a)
(b)
Fig 3 Three dimensional surface plot showing (a) the effect of CTBP and pH0 on the extraction
of GA at the centre level (T = 265 oC CGA0 = 14 g L-1) and (b) the effect of CTBP and T on
the extraction of GA at the centre level (pH0 = 19 CGA0 = 14 g L-1)
F i g 3 ndash Three dimensional surface plot showing (a) the ef-fect of CTBP and pH0 on the extraction of GA at the centre level (T = 265 oC CGA0 = 14 g Lndash1) and (b) the effect of CTBP and T on the extraction of GA at the centre level (pH0 = 19 CGA0 = 14 g Lndash1)
Ta b l e 4 ndash Additional experimental dataset used for the con-struction of ANN model
RunIndependent variables Response E ()
CGA0 (g Lndash1)
CTBP ( vv)
T (oC) pH0Experi- mental ANN Remark
29 05 65 11 196 9895 9868 Training
30 25 30 11 303 9488 9508 Training
31 14 30 11 302 9304 9325 Training
32 18 53 11 213 9728 9700 Training
33 25 75 11 305 9886 9851 Training
34 12 22 11 298 9013 9099 Validation
35 04 82 11 167 9723 9766 Training
36 028 15 11 225 8572 8704 Validation
37 10 30 15 308 9277 9243 Training
38 25 30 15 303 9473 9429 Validation
39 028 95 15 078 9730 9747 Training
40 14 30 16 301 9293 9305 Training
41 25 30 16 302 9449 9402 Training
42 25 30 18 303 9359 9347 Training
43 05 25 20 197 9076 9036 Training
44 14 30 20 302 9274 9212 Training
45 25 30 25 301 9124 9179 Training
46 028 15 25 078 8485 8482 Training
47 25 30 27 303 9263 9150 Validation
48 14 30 27 302 8768 8934 Training
49 04 85 27 139 9808 9796 Training
50 24 72 27 255 9866 9843 Training
51 12 24 27 22 8931 9044 Testing
52 10 62 28 172 9758 9826 Testing
53 15 72 28 165 9852 9831 Testing
54 25 30 28 303 9171 9136 Training
55 16 32 30 246 9224 9210 Training
56 18 78 30 261 9895 9852 Testing
57 18 27 30 261 9039 8930 Training
58 12 65 30 254 9806 9792 Training
59 25 95 30 078 9864 9879 Training
60 14 30 33 302 8798 8766 Training
61 25 30 35 303 9114 9062 Validation
62 25 30 35 3012 9038 9063 Training
63 14 30 37 302 8733 8729 Training
64 24 80 37 222 9859 9857 Training
65 08 60 37 218 9718 9694 Training
66 24 16 37 209 8515 8536 Training
67 16 32 37 178 9104 9102 Training
68 25 30 41 303 9099 9032 Training
69 10 30 41 150 8883 8890 Training
40 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
ed value In Table 1 Runs no 3 4 12 20 show the extraction efficiency above 992 at the cost of 80ndash95 TBP while the optimum condition gives the same efficiency with 65 TBP
ANN modelling of the GA extraction process
In ANN the three layered feed forward error back propagation (FF-EBP) with four neurons in the input layer (input parameters CGA0 CTBP T pH0) ten neurons in the hidden layer (optimum) one neuron in the output layer (extraction efficien-cy) was selected Figure 4 illustrates the ANN ar-chitecture applied for predicting the extraction effi-ciency of GA
A total of 69 experimental data-points (Table 1 and 4) were used to train the ANN model The input and the output data-points were normalized in the range 0 to 1 and 03 to 07 respectively This keeps the back propagation error within the limits and avoids over fitting of the data due to very large and very small values of weights and speeds up training time The data were normalized between the two points n1 and n0 using the following generalized equation (16)
minn 1 0 0
max min
( )(n n ) n( )
x xxx x
minus= minus + minus
where xn is normalized data-point and x xmin xmax are the actual minimum and maximum of inputoutput data The data were split into subsets of training (80 55 data-points) validation (10 7 data-points) and testing (10 7 data-points) Trainlm a training function that updates weights and bias values based on Levenberg-Marquardt backpropagation (LMB) algorithm was used for training the network LMB is an iterative technique which reduces the performance function in each it-eration and makes trainlm the fastest error back-propagation (EBP) algorithm for a moderate sized network35 The externally normalized input values were forwarded from the input layer to the hidden layer and then to the output layer to predict the re-sponse The error MSE was backpropagated from the output to the hidden layer and thereafter to the input layer to modify the weights (IW LW) and bi-ases (b1 b2) Figure 5 describes the general concept of FF-EBP algorithm used in ANN training to pre-dict extraction efficiency During the series of trials for EBP the training algorithm connection weights transfer function hidden layers and hidden neurons were varied and a network topology with an opti-mal network architecture (4101) was selected based on MSE value
The training was successfully terminated after 12 epochs (iterations) as the performance function
36
Fig 4 ANN architecture used for the prediction of GA extraction efficiency F i g 4 ndash ANN architecture used for the prediction of GA extraction efficiency
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 41
F i g 5 ndash Feed forward backpropagation algorithm used in the ANN training for the prediction of GA extraction efficiency
MSE reached a minimum value of 5middot10ndash4 which is lower than the set goal E0 = 1middot10ndash3 (Figure 6) The similar characteristic curve for the test and the vali-dation were observed suggesting no significant over-fitting The estimated MSE values for training (1middot10ndash4) testing (3middot10ndash4) and validation data (5middot10ndash4) were found to be lower than the set goal which is desirable Table 5 gives the optimal weights and bias Wherein IW = HtimesA b1 = Htimes1 LW =OtimesH ma-trix where H is hidden layer neurons (10) A is in-put layer neurons (4) and O is the output layer neu-ron (1) The bias b2 is the single value associated with a single neuron at the output
Figure 7 describes the ANN regression plot for training validation testing and the overall predic-tion set in the form of network output versus exper-imental extraction efficiency It can be observed that the network output values were close to the ex-
F i g 6 ndash MSE for training validation and test dataset com-puted using LMB (with performance of 5middot10ndash4 at 12 epochs and goal of 1middot10ndash3)
38
Fig 6 MSE for training validation and test dataset computed using LMB (with performance of
5middot10-4 at 12 epochs and goal of 1middot10-3)
42 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
perimental extraction efficiency in all the cases The correlation coefficients lsquoRrsquo for training validation and testing were 0995 0984 0989 respectively
whereas the overall prediction set was 0993 which confirms that the ANN model is satisfactory for in-terpolating the experimental data
Ta b l e 5 ndash Optimal values of weights and biases obtained from ANN training using Levenberg-Marquardt backpropagation algorithm
Input weight matrix IW (Weights to Hidden layer from inputs)
54485 74654 14374 2009934801 10038 71242 0667899763 46322 41336 7966216333 44955 0545 2141845359 42326 56151 16171
8043 27753 20607 0620637705 30939 60152 7216681829 118002 62271 3
minus minusminus minusminus minus
minus minus minusminus minus
minus minus 806645814 08091 62607 2537333117 47426 72301 08914
minusminus
Bias vector (Hidden layer) b1 102912 114372 94153 04321 26838 50968 73116 01728 4231 37916 Tminus minus minus minus
Layer weight vector LW (Weight to output layer from hidden layer)
01575 04295 00956 03633 02071 02006 01154 02645 02656 0313minus minus minus minus minus minus
Bias scalar (output layer) b2 06545
F i g 7 ndash ANN model showing the regression plots for training validation testing and overall prediction set
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 43
The ANN model was also validated using sta-tistical analysis ANOVA The degree of freedom due to regression (DFregression) and residuals (DFresidual) was calculated as follows20
regressionDF M 1= minus (17)
residualDF N M= minus (18)
M H(A O 1) O= + + + (19)
where M is the total number of network connec-tions including weights and biases N is the total number of experiments (N = 69) performed to con-struct and validate the ANN model Table 6 shows the ANOVA for the developed ANN model The F-value and the p-value were obtained from ANO-VA for the ANN model as 104 and 00004 respec-tively which confirms the significance of the ANN model Table 7 shows the values of R2 and R2 (adj) as 0986 and 088 respectively indicating the good fit of the ANN model to the experimental extraction efficiency Therefore the ANN model for predicting extraction efficiency E () for the extraction of GA can be written as follows
n n 1 2 (LW (IW ) )y purelin logsig x b b= times times + + (20)
Comparison of ANN and RSM models
The prediction and generalization capabilities of RSM and ANN models were evaluated based on its R2 RMSE AARE and SEP values Table 1 and 4 indicate that both the models predict well the ex-perimental extraction efficiency and their R2 values are closer to 1 Therefore both models can be con-sidered good for predicting the extraction efficiency of GA The generalization capabilities of both the models were compared only with unseen dataset Table 7 shows the unseen data set which were not used for the model development Figure 8 shows the comparative parity plot for RSM and ANN model predictability for the unseen data set and suggests that both the model predictions fit the ex-perimental data very well The comparative values of R2 RMSE AARE and SEP values for both mod-els using unseen data are shown in Table 8 Thus higher R2 and lower RMSE AARE SEP values for ANN indicate that the ANN model is superior over the RSM model for its generalization capability of the GA extraction process
Ta b l e 6 ndash Analysis of variance (ANOVA) analysis for ANN model
Source DFa SSb MSc F-value p-value R2 R2(adj)
Model or Regression 60 11421 1904 1041 00004 0986 088Residual Error 9 1646 183Total 69 11587aDegree of Freedom bSum of Squares cMean Square
Ta b l e 7 ndash Unseen experimental dataset used in the developed RSM and ANN models for the prediction of ex-traction efficiency of GA
RunIndependent variables Response E ()
CGA0 (g Lndash1)
CTBP ( vv)
T(oC) pH0Experi- mental RSM ANN
1 16 15 11 174 9057 9015 91402 25 30 40 30 9005 8836 90373 25 23 37 234 8888 8723 88534 25 37 27 211 9226 9323 92275 25 45 27 185 9525 9531 94786 25 23 37 234 8798 8723 88537 14 30 40 301 8533 8967 87368 18 20 32 221 8745 8805 86639 10 30 24 26 9256 9169 925710 18 36 32 21 9302 9285 938911 10 30 40 15 8883 9118 8908
Ta b l e 8 ndash Comparison of RSM and ANN models for its gen-eralization capability (using unseen data)
Parameters RSM ANN
R2 061 0915RMSE 173 080
AARE 143 066 SEP 191 088
40
Fig 8 Comparison of RSM and ANN model predictions for the unseen data (data not used in
the model development)
F i g 8 ndash Comparison of RSM and ANN model predictions for the unseen data (data not used in the model development)
44 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
Conclusion
The present work suggests that gallic acid can be efficiently removed from aqueous streams using TBP in hexanol The RSM and ANN models were developed to predict the extraction efficiency of GA by varying the parameters temperature concentra-tion of TBP in aqueous phase pH and initial con-centration of gallic acid in aqueous solution The importance of each parameter and its synergistic interactive effects on the extraction of GA was ex-plained with a minimum number of experiments us-ing the RSM model Both models were efficient in the prediction of GA extraction efficiency having close agreement with the experimental extraction efficiency However ANN showed superiority over RSM in terms of accuracy but the marginal errors between the models were very low The optimal conditions to achieve maximum extraction as pre-dicted by the RSM model were CGA0 = 234 g Lndash1 CTBP = 6565 (vv) T = 184 oC and pH0 = 18 at which an experimental extraction efficiency of 995 was observed The models also showed bet-ter performance with the use of unseen data which confirmed their generalization capabilities Thus the present investigation on experiments model-
ling and optimization of the GA extraction process could be of great significance for the design and evaluation of the performance of similar systems
S u p p l e m e n t a r y I n f o r m a t i o n
Supplementary information for confirmation of 2 h shaking time for reactive extraction of gallic acid and retention time of 39 min in HPLC analysis
The trial experiments were conducted for ob-taining an optimum shaking period Equal volumes (20 mL) of aqueous and organic phases were shak-en for different time intervals (5ndash240 min) at 250 rpm Equilibrium was achieved after half an hour (Figure S1) Hence further experiments were con-ducted with 2 h shaking time (sufficient to achieve equilibrium)
Calibration curve and chromatogram for HPLC analysis of gallic acid was obtained by diluting the mother stock of 5 g Lndash1 to prepare different concen-trations (5ndash300 mg Lndash1) of standard sample solu-tions
UV absorption spectra for gallic acid was ob-tained using UV Spectrophotometer Shimadzu UV-1800
F i g S 1 ndash Experimental extraction efficiency (E) of gallic acid versus time for the operating conditions CGA0 = 2 g Lndash1 CTBP = 40 (vv) T = 30 oC pH0 = 3
F i g S 2 ndash Calibration curve for gallic acid obtained us- ing HPLC
44
Fig S4b HPLC Chromatogram of gallic acid at 264 nm and 396 min
F i g S 3 ndash HPLC Chromatogram of gallic acid at 264 nm and 396 min
Ext
ract
ion
effic
ienc
y E
()
Peak
are
a (m
AU
s)
Time (min) Gallic acid concentration (mg Lndash1)
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 45
N o m e n c l a t u r e
AARE ndash absolute average relative errorCGAaq ndash equilibrium concentration of GA in aqueous
phase g Lndash1
CGAorg ndash equilibrium concentration of GA in organic phase g Lndash1
CGA0 ndash initial concentration of GA in aqueous phase g Lndash1
CTBP ndash TBP concentration in organic phase DF ndash degree of freedom in ANOVA analysisE ndash extraction efficiency response F-value ndash Fischer distribution value (ratio of variances)H A O ndash number of neurons in hidden layer input lay-
er and output layer respectivelyIW ndash input weight matrixk ndash number of input variables or factors in
RO-CCDlogsig ndash sigmoidal transfer function for neural net-
workLW ndash layer weight matrixM ndash number of network connections including
weights and biasesMS ndash mean square in ANOVA analysisMSE ndash mean squared errorN ndash total number of experiments to construct
ANN modeln0 ndash centre points in RO-CCDnc ndash corner points or factorial points in RO-CCDns ndash star or axial points in RO-CCDpH0 ndash initial pH of aqueous GA solutionPRESS ndash predicted residual sum of squarespurelin ndash linear transfer function for neural networkp-value ndash null-hypothesis testR ndash coefficient of correlation
R2 ndash coefficient of determinationR2 (adj) ndash adjusted R-squaredR2 (pred) ndash predicted R-squaredRMSE ndash root mean square errorSEP ndash standard error of predictionSS ndash sum of squares in ANOVA analysisT ndash extraction temperature oCw ndash connection weight between layersx1 x2 x3 x4 ndash coded levels of independent vaiables in
RSMxij yij bij ndash network input output and bias of jth neu-
ron to ith layer respectivelya ndash distance of each star point from centre in
RO-CCDbi bj bij ndash regression coefficients representing linear
quadratic and interactive effects in RSM
R e f e r e n c e s
1 Benitez F J Real F J Acero J L Leal A I Garcia C Gallic acid degradation in aqueous solutions by UVH2O2 treatment Fentonrsquos reagent and the photo-Fenton sys-tem J Hazard Mater 126 (2005) 31doi httpsdoiorg101016jjhazmat200504040
2 El-Gohary F A Badawy M I El-Khateeb M A El-Kalliny A S Integrated treatment of olive mill wastewater (OMW) by the combination of fentonrsquos reaction and anaer-obic treatment J Hazard Mater 162 (2009) 1536doi httpsdoiorg101016jjhazmat200806098
3 Herrero M Cifuentes A Ibanez E Sub- and supercriti-cal fluid extraction of functional ingredients from different natural sources plants food-by-products algae and mi-croalgae A Review Food Chem 98 (2006) 136doi httpsdoiorg101016jfoodchem200505058
4 Kougan G B Tabopda T Kuete V Verpoorte R Sim-ple phenols phenolic acids and related esters from the me-dicinal plants of africa in Kuet V (First Ed) Medicinal plant research in Africa Pharmacology and chemistry El-sevier Inc London 2013 225ndash249
45
UV absorption spectra for gallic acid was obtained using UV Spectrophotometer Shimadzu UV-1800
Fig S3 Absorption spectra of gallic acid showing max = 264 nm
--
F i g S 4 ndash Absorption spectra of gallic acid showing lmax = 264 nm
46 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
5 Kapellakis I E Tsagarakis K P Crowther J C Olive oil history production and by-product management Rev Environ Sci Biotechnol 7 (2008) 1doi httpsdoiorg101007s11157-007-9120-9
6 Calace N Nardi E Petronio B Pietroletti M Adsorp-tion of phenols by papermill sludges Environ Pollut 118 (2002) 315doi httpsdoiorg101016S0269-7491(01)00303-7
7 Fki I Allouche N Sayadi S The use of polyphenolic extract purified hydroxytyrosol and 34-dihydroxyphenyl acetic acid from olive mill wastewater for the stabilization of refined oils A potential alternative to synthetic antioxi-dants Food Chem 93 (2005) 197doi httpsdoiorg101016jfoodchem200409014
8 Ollis D F Pelizzetti E Serpone N Photocatalyzed de-struction of water contaminants Environ Sci Technol 25 (1991) 1522doi httpsdoiorg101021es00021a001
9 Masten S J Davies S H The use of ozonation to de-grade organic contaminants in wastewaters Environ Sci Technol 28 (1994) 180Adoi httpsdoiorg101021es00053a001
10 Spigno G Jauregi P Recovery of gallic acid with colloi-dal gas aphrons (CGA) Int J Food Eng 1 (2005) 1doi httpsdoiorg1022021556-37581038
11 Spigno G Dermiki M Pastori C Casanova F Jauregi P Recovery of gallic acid with colloidal gas aphrons gen-erated from a cationic surfactant Sep Purif Technol 71 (2010) 56doi httpsdoiorg101016jseppur200911002
12 Quici N Litter M I Heterogeneous photocatalytic degra-dation of gallic acid under different experimental condi-tions Photochem Photobiol Sci 8 (2009) 975doi httpsdoiorg101039b901904a
13 Singh M Jha A Kumar A Hettiarachchy N Rai A K Sharma D Influence of the solvents on the extraction of major phenolic compounds (punicalagin ellagic acid and gallic acid) and their antioxidant activities in pomegranate aril J Food Sci Technol 51 (2014) 2070doi httpsdoiorg101007s13197-014-1267-0
14 Claudio A F M Ferreira A M Freire C S R Silves-tre A J D Freire M G Coutinho J A P Optimization of the gallic acid extraction using ionic-liquid-based aque-ous two-phase systems Sep Purif Technol 97 (2012) 142doi httpsdoiorg101016jseppur201202036
15 Li Y Wang Y Li Y Dai Y Extraction of glyoxylic acid glycolic acid acrylic acid and benzoic acid with trialkyl-phosphine oxide J Chem Eng Data 48 (2003) 621doi httpsdoiorg101021je020174r
16 Daneshfar A Ghaziaskar H S Homayoun N Solubility of gallic acid in methanol ethanol water and ethyl acetate J Chem Eng Data 53 (2008) 776doi httpsdoiorg101021je700633w
17 Wasewar K L Shende D Z Reactive extraction of caproic acid using tri- n -butyl phosphate in hexanol octa-nol and decanol J Chem Eng Data 56 (2011) 288doi httpsdoiorg101021je100974f
18 Keshav A Chand S Wasewar K L Reactive extraction of acrylic acid using tri- n -butyl phosphate in different di-luents J Chem Eng Data 54 (2009) 1782doi httpsdoiorg101021je800856e
19 Zhong K Wang Q Optimization of ultrasonic extraction of polysaccharides from dried longan pulp using response surface methodology Carbohydr Polym 80 (2010) 19doi httpsdoiorg101016jcarbpol200910066
20 Marchitan N Cojocaru C Mereuta A Duca G Crete-scu I Gonta M Modeling and optimization of tartaric acid reactive extraction from aqueous solutions A compar-ison between response surface methodology and artificial neural network Sep Purif Technol 75 (2010) 273doi httpsdoiorg101016jseppur201008016
21 Messikh N Samar M H Messikh L Neural network analysis of liquidndashliquid extraction of phenol from waste-water using tbp solvent Desalination 208 (2007) 42doi httpsdoiorg101016jdesal200604073
22 Montgomery D Design and analysis of experiments (Fifth Ed) John Wiley amp Sons Inc New Jersey 2001
23 Box G E P Hunter J S H W G Statistics for experi-menters design innovation and discovery (second ed) Vol 48 John Wiley ampSons New Jersey 2006
24 Dornier M Decloux M Trystram G Lebert A Dynam-ic modeling of crossflow microfiltration using neural net-works J Memb Sci 98 (1995) 263doi httpsdoiorg1010160376-7388(94)00195-5
25 Niemi H Bulsari A Palosaari S Simulation of mem-brane separation by neural networks J Memb Sci 102 (1995) 185doi httpsdoiorg1010160376-7388(94)00314-O
26 Bourquin J Schmidli H van Hoogevest P Leuenberger H Pitfalls of artificial neural networks (ANN) modelling technique for data sets containing outlier measurements us-ing a study on mixture properties of a direct compressed dosage form Eur J Pharm Sci 7 (1998) 17doi httpsdoiorg101016S0928-0987(97)10027-6
27 Cornell A I Khuri J A Response surfaces Designs and analyses (second ed) Marcel Dekker Inc New York 1996
28 Bezerra M A Santelli R E Oliveira E P Villar L S Escaleira L A Response surface methodology (RSM) as a tool for optimization in analytical chemistry Talanta 76 (2008) 965doi httpsdoiorg101016jtalanta200805019
29 Hornik K Stinchcombe M White H Multilayer feedfor-ward networks are universal approximators Neural Net-works 2 (1989) 359doi httpsdoiorg1010160893-6080(89)90020-8
30 Hagan M T Demuth H B Beale M H Neural network design PWS Publishing Company Boston 1995
31 Keshav A Wasewar K L Chand S Extraction of propi-onic acid using different extractants (tri-n-butylphosphate tri-n-octylamine and Aliquat 336) Ind Eng Chem Res 47 (2008) 6192doi httpsdoiorg101021ie800006r
32 Yang S T White S A Hsu S T Extraction of carboxyl-ic acids with tertiary and quaternary amines Effect of pH Ind Eng Chem Res 30 (1991) 1335doi httpsdoiorg101021ie00054a040
33 Keshav A Wasewar K L Chand S Extraction of Acryl-ic Propionic and butyric acid using Aliquat 336 in oleyl alcohol Equilibria and effect of temperature Ind Eng Chem Res 48 (2009) 888doi httpsdoiorg101021ie8010337
34 Uslu H Kırbaslar S I Effect of temperature and initial acid concentration on the reactive extraction of carboxylic acids J Chem Eng Data 58 (2013) 1822doi httpsdoiorg101021je4002202
35 Pham D T Sagiroglu S Training multilayered percep-trons for pattern recognition A comparative study of four training algorithms Int J Mach Tools Manuf 41 (2001) 419doi httpsdoiorg101016S0890-6955(00)00073-0
40 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
ed value In Table 1 Runs no 3 4 12 20 show the extraction efficiency above 992 at the cost of 80ndash95 TBP while the optimum condition gives the same efficiency with 65 TBP
ANN modelling of the GA extraction process
In ANN the three layered feed forward error back propagation (FF-EBP) with four neurons in the input layer (input parameters CGA0 CTBP T pH0) ten neurons in the hidden layer (optimum) one neuron in the output layer (extraction efficien-cy) was selected Figure 4 illustrates the ANN ar-chitecture applied for predicting the extraction effi-ciency of GA
A total of 69 experimental data-points (Table 1 and 4) were used to train the ANN model The input and the output data-points were normalized in the range 0 to 1 and 03 to 07 respectively This keeps the back propagation error within the limits and avoids over fitting of the data due to very large and very small values of weights and speeds up training time The data were normalized between the two points n1 and n0 using the following generalized equation (16)
minn 1 0 0
max min
( )(n n ) n( )
x xxx x
minus= minus + minus
where xn is normalized data-point and x xmin xmax are the actual minimum and maximum of inputoutput data The data were split into subsets of training (80 55 data-points) validation (10 7 data-points) and testing (10 7 data-points) Trainlm a training function that updates weights and bias values based on Levenberg-Marquardt backpropagation (LMB) algorithm was used for training the network LMB is an iterative technique which reduces the performance function in each it-eration and makes trainlm the fastest error back-propagation (EBP) algorithm for a moderate sized network35 The externally normalized input values were forwarded from the input layer to the hidden layer and then to the output layer to predict the re-sponse The error MSE was backpropagated from the output to the hidden layer and thereafter to the input layer to modify the weights (IW LW) and bi-ases (b1 b2) Figure 5 describes the general concept of FF-EBP algorithm used in ANN training to pre-dict extraction efficiency During the series of trials for EBP the training algorithm connection weights transfer function hidden layers and hidden neurons were varied and a network topology with an opti-mal network architecture (4101) was selected based on MSE value
The training was successfully terminated after 12 epochs (iterations) as the performance function
36
Fig 4 ANN architecture used for the prediction of GA extraction efficiency F i g 4 ndash ANN architecture used for the prediction of GA extraction efficiency
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 41
F i g 5 ndash Feed forward backpropagation algorithm used in the ANN training for the prediction of GA extraction efficiency
MSE reached a minimum value of 5middot10ndash4 which is lower than the set goal E0 = 1middot10ndash3 (Figure 6) The similar characteristic curve for the test and the vali-dation were observed suggesting no significant over-fitting The estimated MSE values for training (1middot10ndash4) testing (3middot10ndash4) and validation data (5middot10ndash4) were found to be lower than the set goal which is desirable Table 5 gives the optimal weights and bias Wherein IW = HtimesA b1 = Htimes1 LW =OtimesH ma-trix where H is hidden layer neurons (10) A is in-put layer neurons (4) and O is the output layer neu-ron (1) The bias b2 is the single value associated with a single neuron at the output
Figure 7 describes the ANN regression plot for training validation testing and the overall predic-tion set in the form of network output versus exper-imental extraction efficiency It can be observed that the network output values were close to the ex-
F i g 6 ndash MSE for training validation and test dataset com-puted using LMB (with performance of 5middot10ndash4 at 12 epochs and goal of 1middot10ndash3)
38
Fig 6 MSE for training validation and test dataset computed using LMB (with performance of
5middot10-4 at 12 epochs and goal of 1middot10-3)
42 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
perimental extraction efficiency in all the cases The correlation coefficients lsquoRrsquo for training validation and testing were 0995 0984 0989 respectively
whereas the overall prediction set was 0993 which confirms that the ANN model is satisfactory for in-terpolating the experimental data
Ta b l e 5 ndash Optimal values of weights and biases obtained from ANN training using Levenberg-Marquardt backpropagation algorithm
Input weight matrix IW (Weights to Hidden layer from inputs)
54485 74654 14374 2009934801 10038 71242 0667899763 46322 41336 7966216333 44955 0545 2141845359 42326 56151 16171
8043 27753 20607 0620637705 30939 60152 7216681829 118002 62271 3
minus minusminus minusminus minus
minus minus minusminus minus
minus minus 806645814 08091 62607 2537333117 47426 72301 08914
minusminus
Bias vector (Hidden layer) b1 102912 114372 94153 04321 26838 50968 73116 01728 4231 37916 Tminus minus minus minus
Layer weight vector LW (Weight to output layer from hidden layer)
01575 04295 00956 03633 02071 02006 01154 02645 02656 0313minus minus minus minus minus minus
Bias scalar (output layer) b2 06545
F i g 7 ndash ANN model showing the regression plots for training validation testing and overall prediction set
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 43
The ANN model was also validated using sta-tistical analysis ANOVA The degree of freedom due to regression (DFregression) and residuals (DFresidual) was calculated as follows20
regressionDF M 1= minus (17)
residualDF N M= minus (18)
M H(A O 1) O= + + + (19)
where M is the total number of network connec-tions including weights and biases N is the total number of experiments (N = 69) performed to con-struct and validate the ANN model Table 6 shows the ANOVA for the developed ANN model The F-value and the p-value were obtained from ANO-VA for the ANN model as 104 and 00004 respec-tively which confirms the significance of the ANN model Table 7 shows the values of R2 and R2 (adj) as 0986 and 088 respectively indicating the good fit of the ANN model to the experimental extraction efficiency Therefore the ANN model for predicting extraction efficiency E () for the extraction of GA can be written as follows
n n 1 2 (LW (IW ) )y purelin logsig x b b= times times + + (20)
Comparison of ANN and RSM models
The prediction and generalization capabilities of RSM and ANN models were evaluated based on its R2 RMSE AARE and SEP values Table 1 and 4 indicate that both the models predict well the ex-perimental extraction efficiency and their R2 values are closer to 1 Therefore both models can be con-sidered good for predicting the extraction efficiency of GA The generalization capabilities of both the models were compared only with unseen dataset Table 7 shows the unseen data set which were not used for the model development Figure 8 shows the comparative parity plot for RSM and ANN model predictability for the unseen data set and suggests that both the model predictions fit the ex-perimental data very well The comparative values of R2 RMSE AARE and SEP values for both mod-els using unseen data are shown in Table 8 Thus higher R2 and lower RMSE AARE SEP values for ANN indicate that the ANN model is superior over the RSM model for its generalization capability of the GA extraction process
Ta b l e 6 ndash Analysis of variance (ANOVA) analysis for ANN model
Source DFa SSb MSc F-value p-value R2 R2(adj)
Model or Regression 60 11421 1904 1041 00004 0986 088Residual Error 9 1646 183Total 69 11587aDegree of Freedom bSum of Squares cMean Square
Ta b l e 7 ndash Unseen experimental dataset used in the developed RSM and ANN models for the prediction of ex-traction efficiency of GA
RunIndependent variables Response E ()
CGA0 (g Lndash1)
CTBP ( vv)
T(oC) pH0Experi- mental RSM ANN
1 16 15 11 174 9057 9015 91402 25 30 40 30 9005 8836 90373 25 23 37 234 8888 8723 88534 25 37 27 211 9226 9323 92275 25 45 27 185 9525 9531 94786 25 23 37 234 8798 8723 88537 14 30 40 301 8533 8967 87368 18 20 32 221 8745 8805 86639 10 30 24 26 9256 9169 925710 18 36 32 21 9302 9285 938911 10 30 40 15 8883 9118 8908
Ta b l e 8 ndash Comparison of RSM and ANN models for its gen-eralization capability (using unseen data)
Parameters RSM ANN
R2 061 0915RMSE 173 080
AARE 143 066 SEP 191 088
40
Fig 8 Comparison of RSM and ANN model predictions for the unseen data (data not used in
the model development)
F i g 8 ndash Comparison of RSM and ANN model predictions for the unseen data (data not used in the model development)
44 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
Conclusion
The present work suggests that gallic acid can be efficiently removed from aqueous streams using TBP in hexanol The RSM and ANN models were developed to predict the extraction efficiency of GA by varying the parameters temperature concentra-tion of TBP in aqueous phase pH and initial con-centration of gallic acid in aqueous solution The importance of each parameter and its synergistic interactive effects on the extraction of GA was ex-plained with a minimum number of experiments us-ing the RSM model Both models were efficient in the prediction of GA extraction efficiency having close agreement with the experimental extraction efficiency However ANN showed superiority over RSM in terms of accuracy but the marginal errors between the models were very low The optimal conditions to achieve maximum extraction as pre-dicted by the RSM model were CGA0 = 234 g Lndash1 CTBP = 6565 (vv) T = 184 oC and pH0 = 18 at which an experimental extraction efficiency of 995 was observed The models also showed bet-ter performance with the use of unseen data which confirmed their generalization capabilities Thus the present investigation on experiments model-
ling and optimization of the GA extraction process could be of great significance for the design and evaluation of the performance of similar systems
S u p p l e m e n t a r y I n f o r m a t i o n
Supplementary information for confirmation of 2 h shaking time for reactive extraction of gallic acid and retention time of 39 min in HPLC analysis
The trial experiments were conducted for ob-taining an optimum shaking period Equal volumes (20 mL) of aqueous and organic phases were shak-en for different time intervals (5ndash240 min) at 250 rpm Equilibrium was achieved after half an hour (Figure S1) Hence further experiments were con-ducted with 2 h shaking time (sufficient to achieve equilibrium)
Calibration curve and chromatogram for HPLC analysis of gallic acid was obtained by diluting the mother stock of 5 g Lndash1 to prepare different concen-trations (5ndash300 mg Lndash1) of standard sample solu-tions
UV absorption spectra for gallic acid was ob-tained using UV Spectrophotometer Shimadzu UV-1800
F i g S 1 ndash Experimental extraction efficiency (E) of gallic acid versus time for the operating conditions CGA0 = 2 g Lndash1 CTBP = 40 (vv) T = 30 oC pH0 = 3
F i g S 2 ndash Calibration curve for gallic acid obtained us- ing HPLC
44
Fig S4b HPLC Chromatogram of gallic acid at 264 nm and 396 min
F i g S 3 ndash HPLC Chromatogram of gallic acid at 264 nm and 396 min
Ext
ract
ion
effic
ienc
y E
()
Peak
are
a (m
AU
s)
Time (min) Gallic acid concentration (mg Lndash1)
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 45
N o m e n c l a t u r e
AARE ndash absolute average relative errorCGAaq ndash equilibrium concentration of GA in aqueous
phase g Lndash1
CGAorg ndash equilibrium concentration of GA in organic phase g Lndash1
CGA0 ndash initial concentration of GA in aqueous phase g Lndash1
CTBP ndash TBP concentration in organic phase DF ndash degree of freedom in ANOVA analysisE ndash extraction efficiency response F-value ndash Fischer distribution value (ratio of variances)H A O ndash number of neurons in hidden layer input lay-
er and output layer respectivelyIW ndash input weight matrixk ndash number of input variables or factors in
RO-CCDlogsig ndash sigmoidal transfer function for neural net-
workLW ndash layer weight matrixM ndash number of network connections including
weights and biasesMS ndash mean square in ANOVA analysisMSE ndash mean squared errorN ndash total number of experiments to construct
ANN modeln0 ndash centre points in RO-CCDnc ndash corner points or factorial points in RO-CCDns ndash star or axial points in RO-CCDpH0 ndash initial pH of aqueous GA solutionPRESS ndash predicted residual sum of squarespurelin ndash linear transfer function for neural networkp-value ndash null-hypothesis testR ndash coefficient of correlation
R2 ndash coefficient of determinationR2 (adj) ndash adjusted R-squaredR2 (pred) ndash predicted R-squaredRMSE ndash root mean square errorSEP ndash standard error of predictionSS ndash sum of squares in ANOVA analysisT ndash extraction temperature oCw ndash connection weight between layersx1 x2 x3 x4 ndash coded levels of independent vaiables in
RSMxij yij bij ndash network input output and bias of jth neu-
ron to ith layer respectivelya ndash distance of each star point from centre in
RO-CCDbi bj bij ndash regression coefficients representing linear
quadratic and interactive effects in RSM
R e f e r e n c e s
1 Benitez F J Real F J Acero J L Leal A I Garcia C Gallic acid degradation in aqueous solutions by UVH2O2 treatment Fentonrsquos reagent and the photo-Fenton sys-tem J Hazard Mater 126 (2005) 31doi httpsdoiorg101016jjhazmat200504040
2 El-Gohary F A Badawy M I El-Khateeb M A El-Kalliny A S Integrated treatment of olive mill wastewater (OMW) by the combination of fentonrsquos reaction and anaer-obic treatment J Hazard Mater 162 (2009) 1536doi httpsdoiorg101016jjhazmat200806098
3 Herrero M Cifuentes A Ibanez E Sub- and supercriti-cal fluid extraction of functional ingredients from different natural sources plants food-by-products algae and mi-croalgae A Review Food Chem 98 (2006) 136doi httpsdoiorg101016jfoodchem200505058
4 Kougan G B Tabopda T Kuete V Verpoorte R Sim-ple phenols phenolic acids and related esters from the me-dicinal plants of africa in Kuet V (First Ed) Medicinal plant research in Africa Pharmacology and chemistry El-sevier Inc London 2013 225ndash249
45
UV absorption spectra for gallic acid was obtained using UV Spectrophotometer Shimadzu UV-1800
Fig S3 Absorption spectra of gallic acid showing max = 264 nm
--
F i g S 4 ndash Absorption spectra of gallic acid showing lmax = 264 nm
46 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
5 Kapellakis I E Tsagarakis K P Crowther J C Olive oil history production and by-product management Rev Environ Sci Biotechnol 7 (2008) 1doi httpsdoiorg101007s11157-007-9120-9
6 Calace N Nardi E Petronio B Pietroletti M Adsorp-tion of phenols by papermill sludges Environ Pollut 118 (2002) 315doi httpsdoiorg101016S0269-7491(01)00303-7
7 Fki I Allouche N Sayadi S The use of polyphenolic extract purified hydroxytyrosol and 34-dihydroxyphenyl acetic acid from olive mill wastewater for the stabilization of refined oils A potential alternative to synthetic antioxi-dants Food Chem 93 (2005) 197doi httpsdoiorg101016jfoodchem200409014
8 Ollis D F Pelizzetti E Serpone N Photocatalyzed de-struction of water contaminants Environ Sci Technol 25 (1991) 1522doi httpsdoiorg101021es00021a001
9 Masten S J Davies S H The use of ozonation to de-grade organic contaminants in wastewaters Environ Sci Technol 28 (1994) 180Adoi httpsdoiorg101021es00053a001
10 Spigno G Jauregi P Recovery of gallic acid with colloi-dal gas aphrons (CGA) Int J Food Eng 1 (2005) 1doi httpsdoiorg1022021556-37581038
11 Spigno G Dermiki M Pastori C Casanova F Jauregi P Recovery of gallic acid with colloidal gas aphrons gen-erated from a cationic surfactant Sep Purif Technol 71 (2010) 56doi httpsdoiorg101016jseppur200911002
12 Quici N Litter M I Heterogeneous photocatalytic degra-dation of gallic acid under different experimental condi-tions Photochem Photobiol Sci 8 (2009) 975doi httpsdoiorg101039b901904a
13 Singh M Jha A Kumar A Hettiarachchy N Rai A K Sharma D Influence of the solvents on the extraction of major phenolic compounds (punicalagin ellagic acid and gallic acid) and their antioxidant activities in pomegranate aril J Food Sci Technol 51 (2014) 2070doi httpsdoiorg101007s13197-014-1267-0
14 Claudio A F M Ferreira A M Freire C S R Silves-tre A J D Freire M G Coutinho J A P Optimization of the gallic acid extraction using ionic-liquid-based aque-ous two-phase systems Sep Purif Technol 97 (2012) 142doi httpsdoiorg101016jseppur201202036
15 Li Y Wang Y Li Y Dai Y Extraction of glyoxylic acid glycolic acid acrylic acid and benzoic acid with trialkyl-phosphine oxide J Chem Eng Data 48 (2003) 621doi httpsdoiorg101021je020174r
16 Daneshfar A Ghaziaskar H S Homayoun N Solubility of gallic acid in methanol ethanol water and ethyl acetate J Chem Eng Data 53 (2008) 776doi httpsdoiorg101021je700633w
17 Wasewar K L Shende D Z Reactive extraction of caproic acid using tri- n -butyl phosphate in hexanol octa-nol and decanol J Chem Eng Data 56 (2011) 288doi httpsdoiorg101021je100974f
18 Keshav A Chand S Wasewar K L Reactive extraction of acrylic acid using tri- n -butyl phosphate in different di-luents J Chem Eng Data 54 (2009) 1782doi httpsdoiorg101021je800856e
19 Zhong K Wang Q Optimization of ultrasonic extraction of polysaccharides from dried longan pulp using response surface methodology Carbohydr Polym 80 (2010) 19doi httpsdoiorg101016jcarbpol200910066
20 Marchitan N Cojocaru C Mereuta A Duca G Crete-scu I Gonta M Modeling and optimization of tartaric acid reactive extraction from aqueous solutions A compar-ison between response surface methodology and artificial neural network Sep Purif Technol 75 (2010) 273doi httpsdoiorg101016jseppur201008016
21 Messikh N Samar M H Messikh L Neural network analysis of liquidndashliquid extraction of phenol from waste-water using tbp solvent Desalination 208 (2007) 42doi httpsdoiorg101016jdesal200604073
22 Montgomery D Design and analysis of experiments (Fifth Ed) John Wiley amp Sons Inc New Jersey 2001
23 Box G E P Hunter J S H W G Statistics for experi-menters design innovation and discovery (second ed) Vol 48 John Wiley ampSons New Jersey 2006
24 Dornier M Decloux M Trystram G Lebert A Dynam-ic modeling of crossflow microfiltration using neural net-works J Memb Sci 98 (1995) 263doi httpsdoiorg1010160376-7388(94)00195-5
25 Niemi H Bulsari A Palosaari S Simulation of mem-brane separation by neural networks J Memb Sci 102 (1995) 185doi httpsdoiorg1010160376-7388(94)00314-O
26 Bourquin J Schmidli H van Hoogevest P Leuenberger H Pitfalls of artificial neural networks (ANN) modelling technique for data sets containing outlier measurements us-ing a study on mixture properties of a direct compressed dosage form Eur J Pharm Sci 7 (1998) 17doi httpsdoiorg101016S0928-0987(97)10027-6
27 Cornell A I Khuri J A Response surfaces Designs and analyses (second ed) Marcel Dekker Inc New York 1996
28 Bezerra M A Santelli R E Oliveira E P Villar L S Escaleira L A Response surface methodology (RSM) as a tool for optimization in analytical chemistry Talanta 76 (2008) 965doi httpsdoiorg101016jtalanta200805019
29 Hornik K Stinchcombe M White H Multilayer feedfor-ward networks are universal approximators Neural Net-works 2 (1989) 359doi httpsdoiorg1010160893-6080(89)90020-8
30 Hagan M T Demuth H B Beale M H Neural network design PWS Publishing Company Boston 1995
31 Keshav A Wasewar K L Chand S Extraction of propi-onic acid using different extractants (tri-n-butylphosphate tri-n-octylamine and Aliquat 336) Ind Eng Chem Res 47 (2008) 6192doi httpsdoiorg101021ie800006r
32 Yang S T White S A Hsu S T Extraction of carboxyl-ic acids with tertiary and quaternary amines Effect of pH Ind Eng Chem Res 30 (1991) 1335doi httpsdoiorg101021ie00054a040
33 Keshav A Wasewar K L Chand S Extraction of Acryl-ic Propionic and butyric acid using Aliquat 336 in oleyl alcohol Equilibria and effect of temperature Ind Eng Chem Res 48 (2009) 888doi httpsdoiorg101021ie8010337
34 Uslu H Kırbaslar S I Effect of temperature and initial acid concentration on the reactive extraction of carboxylic acids J Chem Eng Data 58 (2013) 1822doi httpsdoiorg101021je4002202
35 Pham D T Sagiroglu S Training multilayered percep-trons for pattern recognition A comparative study of four training algorithms Int J Mach Tools Manuf 41 (2001) 419doi httpsdoiorg101016S0890-6955(00)00073-0
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 41
F i g 5 ndash Feed forward backpropagation algorithm used in the ANN training for the prediction of GA extraction efficiency
MSE reached a minimum value of 5middot10ndash4 which is lower than the set goal E0 = 1middot10ndash3 (Figure 6) The similar characteristic curve for the test and the vali-dation were observed suggesting no significant over-fitting The estimated MSE values for training (1middot10ndash4) testing (3middot10ndash4) and validation data (5middot10ndash4) were found to be lower than the set goal which is desirable Table 5 gives the optimal weights and bias Wherein IW = HtimesA b1 = Htimes1 LW =OtimesH ma-trix where H is hidden layer neurons (10) A is in-put layer neurons (4) and O is the output layer neu-ron (1) The bias b2 is the single value associated with a single neuron at the output
Figure 7 describes the ANN regression plot for training validation testing and the overall predic-tion set in the form of network output versus exper-imental extraction efficiency It can be observed that the network output values were close to the ex-
F i g 6 ndash MSE for training validation and test dataset com-puted using LMB (with performance of 5middot10ndash4 at 12 epochs and goal of 1middot10ndash3)
38
Fig 6 MSE for training validation and test dataset computed using LMB (with performance of
5middot10-4 at 12 epochs and goal of 1middot10-3)
42 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
perimental extraction efficiency in all the cases The correlation coefficients lsquoRrsquo for training validation and testing were 0995 0984 0989 respectively
whereas the overall prediction set was 0993 which confirms that the ANN model is satisfactory for in-terpolating the experimental data
Ta b l e 5 ndash Optimal values of weights and biases obtained from ANN training using Levenberg-Marquardt backpropagation algorithm
Input weight matrix IW (Weights to Hidden layer from inputs)
54485 74654 14374 2009934801 10038 71242 0667899763 46322 41336 7966216333 44955 0545 2141845359 42326 56151 16171
8043 27753 20607 0620637705 30939 60152 7216681829 118002 62271 3
minus minusminus minusminus minus
minus minus minusminus minus
minus minus 806645814 08091 62607 2537333117 47426 72301 08914
minusminus
Bias vector (Hidden layer) b1 102912 114372 94153 04321 26838 50968 73116 01728 4231 37916 Tminus minus minus minus
Layer weight vector LW (Weight to output layer from hidden layer)
01575 04295 00956 03633 02071 02006 01154 02645 02656 0313minus minus minus minus minus minus
Bias scalar (output layer) b2 06545
F i g 7 ndash ANN model showing the regression plots for training validation testing and overall prediction set
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 43
The ANN model was also validated using sta-tistical analysis ANOVA The degree of freedom due to regression (DFregression) and residuals (DFresidual) was calculated as follows20
regressionDF M 1= minus (17)
residualDF N M= minus (18)
M H(A O 1) O= + + + (19)
where M is the total number of network connec-tions including weights and biases N is the total number of experiments (N = 69) performed to con-struct and validate the ANN model Table 6 shows the ANOVA for the developed ANN model The F-value and the p-value were obtained from ANO-VA for the ANN model as 104 and 00004 respec-tively which confirms the significance of the ANN model Table 7 shows the values of R2 and R2 (adj) as 0986 and 088 respectively indicating the good fit of the ANN model to the experimental extraction efficiency Therefore the ANN model for predicting extraction efficiency E () for the extraction of GA can be written as follows
n n 1 2 (LW (IW ) )y purelin logsig x b b= times times + + (20)
Comparison of ANN and RSM models
The prediction and generalization capabilities of RSM and ANN models were evaluated based on its R2 RMSE AARE and SEP values Table 1 and 4 indicate that both the models predict well the ex-perimental extraction efficiency and their R2 values are closer to 1 Therefore both models can be con-sidered good for predicting the extraction efficiency of GA The generalization capabilities of both the models were compared only with unseen dataset Table 7 shows the unseen data set which were not used for the model development Figure 8 shows the comparative parity plot for RSM and ANN model predictability for the unseen data set and suggests that both the model predictions fit the ex-perimental data very well The comparative values of R2 RMSE AARE and SEP values for both mod-els using unseen data are shown in Table 8 Thus higher R2 and lower RMSE AARE SEP values for ANN indicate that the ANN model is superior over the RSM model for its generalization capability of the GA extraction process
Ta b l e 6 ndash Analysis of variance (ANOVA) analysis for ANN model
Source DFa SSb MSc F-value p-value R2 R2(adj)
Model or Regression 60 11421 1904 1041 00004 0986 088Residual Error 9 1646 183Total 69 11587aDegree of Freedom bSum of Squares cMean Square
Ta b l e 7 ndash Unseen experimental dataset used in the developed RSM and ANN models for the prediction of ex-traction efficiency of GA
RunIndependent variables Response E ()
CGA0 (g Lndash1)
CTBP ( vv)
T(oC) pH0Experi- mental RSM ANN
1 16 15 11 174 9057 9015 91402 25 30 40 30 9005 8836 90373 25 23 37 234 8888 8723 88534 25 37 27 211 9226 9323 92275 25 45 27 185 9525 9531 94786 25 23 37 234 8798 8723 88537 14 30 40 301 8533 8967 87368 18 20 32 221 8745 8805 86639 10 30 24 26 9256 9169 925710 18 36 32 21 9302 9285 938911 10 30 40 15 8883 9118 8908
Ta b l e 8 ndash Comparison of RSM and ANN models for its gen-eralization capability (using unseen data)
Parameters RSM ANN
R2 061 0915RMSE 173 080
AARE 143 066 SEP 191 088
40
Fig 8 Comparison of RSM and ANN model predictions for the unseen data (data not used in
the model development)
F i g 8 ndash Comparison of RSM and ANN model predictions for the unseen data (data not used in the model development)
44 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
Conclusion
The present work suggests that gallic acid can be efficiently removed from aqueous streams using TBP in hexanol The RSM and ANN models were developed to predict the extraction efficiency of GA by varying the parameters temperature concentra-tion of TBP in aqueous phase pH and initial con-centration of gallic acid in aqueous solution The importance of each parameter and its synergistic interactive effects on the extraction of GA was ex-plained with a minimum number of experiments us-ing the RSM model Both models were efficient in the prediction of GA extraction efficiency having close agreement with the experimental extraction efficiency However ANN showed superiority over RSM in terms of accuracy but the marginal errors between the models were very low The optimal conditions to achieve maximum extraction as pre-dicted by the RSM model were CGA0 = 234 g Lndash1 CTBP = 6565 (vv) T = 184 oC and pH0 = 18 at which an experimental extraction efficiency of 995 was observed The models also showed bet-ter performance with the use of unseen data which confirmed their generalization capabilities Thus the present investigation on experiments model-
ling and optimization of the GA extraction process could be of great significance for the design and evaluation of the performance of similar systems
S u p p l e m e n t a r y I n f o r m a t i o n
Supplementary information for confirmation of 2 h shaking time for reactive extraction of gallic acid and retention time of 39 min in HPLC analysis
The trial experiments were conducted for ob-taining an optimum shaking period Equal volumes (20 mL) of aqueous and organic phases were shak-en for different time intervals (5ndash240 min) at 250 rpm Equilibrium was achieved after half an hour (Figure S1) Hence further experiments were con-ducted with 2 h shaking time (sufficient to achieve equilibrium)
Calibration curve and chromatogram for HPLC analysis of gallic acid was obtained by diluting the mother stock of 5 g Lndash1 to prepare different concen-trations (5ndash300 mg Lndash1) of standard sample solu-tions
UV absorption spectra for gallic acid was ob-tained using UV Spectrophotometer Shimadzu UV-1800
F i g S 1 ndash Experimental extraction efficiency (E) of gallic acid versus time for the operating conditions CGA0 = 2 g Lndash1 CTBP = 40 (vv) T = 30 oC pH0 = 3
F i g S 2 ndash Calibration curve for gallic acid obtained us- ing HPLC
44
Fig S4b HPLC Chromatogram of gallic acid at 264 nm and 396 min
F i g S 3 ndash HPLC Chromatogram of gallic acid at 264 nm and 396 min
Ext
ract
ion
effic
ienc
y E
()
Peak
are
a (m
AU
s)
Time (min) Gallic acid concentration (mg Lndash1)
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 45
N o m e n c l a t u r e
AARE ndash absolute average relative errorCGAaq ndash equilibrium concentration of GA in aqueous
phase g Lndash1
CGAorg ndash equilibrium concentration of GA in organic phase g Lndash1
CGA0 ndash initial concentration of GA in aqueous phase g Lndash1
CTBP ndash TBP concentration in organic phase DF ndash degree of freedom in ANOVA analysisE ndash extraction efficiency response F-value ndash Fischer distribution value (ratio of variances)H A O ndash number of neurons in hidden layer input lay-
er and output layer respectivelyIW ndash input weight matrixk ndash number of input variables or factors in
RO-CCDlogsig ndash sigmoidal transfer function for neural net-
workLW ndash layer weight matrixM ndash number of network connections including
weights and biasesMS ndash mean square in ANOVA analysisMSE ndash mean squared errorN ndash total number of experiments to construct
ANN modeln0 ndash centre points in RO-CCDnc ndash corner points or factorial points in RO-CCDns ndash star or axial points in RO-CCDpH0 ndash initial pH of aqueous GA solutionPRESS ndash predicted residual sum of squarespurelin ndash linear transfer function for neural networkp-value ndash null-hypothesis testR ndash coefficient of correlation
R2 ndash coefficient of determinationR2 (adj) ndash adjusted R-squaredR2 (pred) ndash predicted R-squaredRMSE ndash root mean square errorSEP ndash standard error of predictionSS ndash sum of squares in ANOVA analysisT ndash extraction temperature oCw ndash connection weight between layersx1 x2 x3 x4 ndash coded levels of independent vaiables in
RSMxij yij bij ndash network input output and bias of jth neu-
ron to ith layer respectivelya ndash distance of each star point from centre in
RO-CCDbi bj bij ndash regression coefficients representing linear
quadratic and interactive effects in RSM
R e f e r e n c e s
1 Benitez F J Real F J Acero J L Leal A I Garcia C Gallic acid degradation in aqueous solutions by UVH2O2 treatment Fentonrsquos reagent and the photo-Fenton sys-tem J Hazard Mater 126 (2005) 31doi httpsdoiorg101016jjhazmat200504040
2 El-Gohary F A Badawy M I El-Khateeb M A El-Kalliny A S Integrated treatment of olive mill wastewater (OMW) by the combination of fentonrsquos reaction and anaer-obic treatment J Hazard Mater 162 (2009) 1536doi httpsdoiorg101016jjhazmat200806098
3 Herrero M Cifuentes A Ibanez E Sub- and supercriti-cal fluid extraction of functional ingredients from different natural sources plants food-by-products algae and mi-croalgae A Review Food Chem 98 (2006) 136doi httpsdoiorg101016jfoodchem200505058
4 Kougan G B Tabopda T Kuete V Verpoorte R Sim-ple phenols phenolic acids and related esters from the me-dicinal plants of africa in Kuet V (First Ed) Medicinal plant research in Africa Pharmacology and chemistry El-sevier Inc London 2013 225ndash249
45
UV absorption spectra for gallic acid was obtained using UV Spectrophotometer Shimadzu UV-1800
Fig S3 Absorption spectra of gallic acid showing max = 264 nm
--
F i g S 4 ndash Absorption spectra of gallic acid showing lmax = 264 nm
46 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
5 Kapellakis I E Tsagarakis K P Crowther J C Olive oil history production and by-product management Rev Environ Sci Biotechnol 7 (2008) 1doi httpsdoiorg101007s11157-007-9120-9
6 Calace N Nardi E Petronio B Pietroletti M Adsorp-tion of phenols by papermill sludges Environ Pollut 118 (2002) 315doi httpsdoiorg101016S0269-7491(01)00303-7
7 Fki I Allouche N Sayadi S The use of polyphenolic extract purified hydroxytyrosol and 34-dihydroxyphenyl acetic acid from olive mill wastewater for the stabilization of refined oils A potential alternative to synthetic antioxi-dants Food Chem 93 (2005) 197doi httpsdoiorg101016jfoodchem200409014
8 Ollis D F Pelizzetti E Serpone N Photocatalyzed de-struction of water contaminants Environ Sci Technol 25 (1991) 1522doi httpsdoiorg101021es00021a001
9 Masten S J Davies S H The use of ozonation to de-grade organic contaminants in wastewaters Environ Sci Technol 28 (1994) 180Adoi httpsdoiorg101021es00053a001
10 Spigno G Jauregi P Recovery of gallic acid with colloi-dal gas aphrons (CGA) Int J Food Eng 1 (2005) 1doi httpsdoiorg1022021556-37581038
11 Spigno G Dermiki M Pastori C Casanova F Jauregi P Recovery of gallic acid with colloidal gas aphrons gen-erated from a cationic surfactant Sep Purif Technol 71 (2010) 56doi httpsdoiorg101016jseppur200911002
12 Quici N Litter M I Heterogeneous photocatalytic degra-dation of gallic acid under different experimental condi-tions Photochem Photobiol Sci 8 (2009) 975doi httpsdoiorg101039b901904a
13 Singh M Jha A Kumar A Hettiarachchy N Rai A K Sharma D Influence of the solvents on the extraction of major phenolic compounds (punicalagin ellagic acid and gallic acid) and their antioxidant activities in pomegranate aril J Food Sci Technol 51 (2014) 2070doi httpsdoiorg101007s13197-014-1267-0
14 Claudio A F M Ferreira A M Freire C S R Silves-tre A J D Freire M G Coutinho J A P Optimization of the gallic acid extraction using ionic-liquid-based aque-ous two-phase systems Sep Purif Technol 97 (2012) 142doi httpsdoiorg101016jseppur201202036
15 Li Y Wang Y Li Y Dai Y Extraction of glyoxylic acid glycolic acid acrylic acid and benzoic acid with trialkyl-phosphine oxide J Chem Eng Data 48 (2003) 621doi httpsdoiorg101021je020174r
16 Daneshfar A Ghaziaskar H S Homayoun N Solubility of gallic acid in methanol ethanol water and ethyl acetate J Chem Eng Data 53 (2008) 776doi httpsdoiorg101021je700633w
17 Wasewar K L Shende D Z Reactive extraction of caproic acid using tri- n -butyl phosphate in hexanol octa-nol and decanol J Chem Eng Data 56 (2011) 288doi httpsdoiorg101021je100974f
18 Keshav A Chand S Wasewar K L Reactive extraction of acrylic acid using tri- n -butyl phosphate in different di-luents J Chem Eng Data 54 (2009) 1782doi httpsdoiorg101021je800856e
19 Zhong K Wang Q Optimization of ultrasonic extraction of polysaccharides from dried longan pulp using response surface methodology Carbohydr Polym 80 (2010) 19doi httpsdoiorg101016jcarbpol200910066
20 Marchitan N Cojocaru C Mereuta A Duca G Crete-scu I Gonta M Modeling and optimization of tartaric acid reactive extraction from aqueous solutions A compar-ison between response surface methodology and artificial neural network Sep Purif Technol 75 (2010) 273doi httpsdoiorg101016jseppur201008016
21 Messikh N Samar M H Messikh L Neural network analysis of liquidndashliquid extraction of phenol from waste-water using tbp solvent Desalination 208 (2007) 42doi httpsdoiorg101016jdesal200604073
22 Montgomery D Design and analysis of experiments (Fifth Ed) John Wiley amp Sons Inc New Jersey 2001
23 Box G E P Hunter J S H W G Statistics for experi-menters design innovation and discovery (second ed) Vol 48 John Wiley ampSons New Jersey 2006
24 Dornier M Decloux M Trystram G Lebert A Dynam-ic modeling of crossflow microfiltration using neural net-works J Memb Sci 98 (1995) 263doi httpsdoiorg1010160376-7388(94)00195-5
25 Niemi H Bulsari A Palosaari S Simulation of mem-brane separation by neural networks J Memb Sci 102 (1995) 185doi httpsdoiorg1010160376-7388(94)00314-O
26 Bourquin J Schmidli H van Hoogevest P Leuenberger H Pitfalls of artificial neural networks (ANN) modelling technique for data sets containing outlier measurements us-ing a study on mixture properties of a direct compressed dosage form Eur J Pharm Sci 7 (1998) 17doi httpsdoiorg101016S0928-0987(97)10027-6
27 Cornell A I Khuri J A Response surfaces Designs and analyses (second ed) Marcel Dekker Inc New York 1996
28 Bezerra M A Santelli R E Oliveira E P Villar L S Escaleira L A Response surface methodology (RSM) as a tool for optimization in analytical chemistry Talanta 76 (2008) 965doi httpsdoiorg101016jtalanta200805019
29 Hornik K Stinchcombe M White H Multilayer feedfor-ward networks are universal approximators Neural Net-works 2 (1989) 359doi httpsdoiorg1010160893-6080(89)90020-8
30 Hagan M T Demuth H B Beale M H Neural network design PWS Publishing Company Boston 1995
31 Keshav A Wasewar K L Chand S Extraction of propi-onic acid using different extractants (tri-n-butylphosphate tri-n-octylamine and Aliquat 336) Ind Eng Chem Res 47 (2008) 6192doi httpsdoiorg101021ie800006r
32 Yang S T White S A Hsu S T Extraction of carboxyl-ic acids with tertiary and quaternary amines Effect of pH Ind Eng Chem Res 30 (1991) 1335doi httpsdoiorg101021ie00054a040
33 Keshav A Wasewar K L Chand S Extraction of Acryl-ic Propionic and butyric acid using Aliquat 336 in oleyl alcohol Equilibria and effect of temperature Ind Eng Chem Res 48 (2009) 888doi httpsdoiorg101021ie8010337
34 Uslu H Kırbaslar S I Effect of temperature and initial acid concentration on the reactive extraction of carboxylic acids J Chem Eng Data 58 (2013) 1822doi httpsdoiorg101021je4002202
35 Pham D T Sagiroglu S Training multilayered percep-trons for pattern recognition A comparative study of four training algorithms Int J Mach Tools Manuf 41 (2001) 419doi httpsdoiorg101016S0890-6955(00)00073-0
42 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
perimental extraction efficiency in all the cases The correlation coefficients lsquoRrsquo for training validation and testing were 0995 0984 0989 respectively
whereas the overall prediction set was 0993 which confirms that the ANN model is satisfactory for in-terpolating the experimental data
Ta b l e 5 ndash Optimal values of weights and biases obtained from ANN training using Levenberg-Marquardt backpropagation algorithm
Input weight matrix IW (Weights to Hidden layer from inputs)
54485 74654 14374 2009934801 10038 71242 0667899763 46322 41336 7966216333 44955 0545 2141845359 42326 56151 16171
8043 27753 20607 0620637705 30939 60152 7216681829 118002 62271 3
minus minusminus minusminus minus
minus minus minusminus minus
minus minus 806645814 08091 62607 2537333117 47426 72301 08914
minusminus
Bias vector (Hidden layer) b1 102912 114372 94153 04321 26838 50968 73116 01728 4231 37916 Tminus minus minus minus
Layer weight vector LW (Weight to output layer from hidden layer)
01575 04295 00956 03633 02071 02006 01154 02645 02656 0313minus minus minus minus minus minus
Bias scalar (output layer) b2 06545
F i g 7 ndash ANN model showing the regression plots for training validation testing and overall prediction set
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 43
The ANN model was also validated using sta-tistical analysis ANOVA The degree of freedom due to regression (DFregression) and residuals (DFresidual) was calculated as follows20
regressionDF M 1= minus (17)
residualDF N M= minus (18)
M H(A O 1) O= + + + (19)
where M is the total number of network connec-tions including weights and biases N is the total number of experiments (N = 69) performed to con-struct and validate the ANN model Table 6 shows the ANOVA for the developed ANN model The F-value and the p-value were obtained from ANO-VA for the ANN model as 104 and 00004 respec-tively which confirms the significance of the ANN model Table 7 shows the values of R2 and R2 (adj) as 0986 and 088 respectively indicating the good fit of the ANN model to the experimental extraction efficiency Therefore the ANN model for predicting extraction efficiency E () for the extraction of GA can be written as follows
n n 1 2 (LW (IW ) )y purelin logsig x b b= times times + + (20)
Comparison of ANN and RSM models
The prediction and generalization capabilities of RSM and ANN models were evaluated based on its R2 RMSE AARE and SEP values Table 1 and 4 indicate that both the models predict well the ex-perimental extraction efficiency and their R2 values are closer to 1 Therefore both models can be con-sidered good for predicting the extraction efficiency of GA The generalization capabilities of both the models were compared only with unseen dataset Table 7 shows the unseen data set which were not used for the model development Figure 8 shows the comparative parity plot for RSM and ANN model predictability for the unseen data set and suggests that both the model predictions fit the ex-perimental data very well The comparative values of R2 RMSE AARE and SEP values for both mod-els using unseen data are shown in Table 8 Thus higher R2 and lower RMSE AARE SEP values for ANN indicate that the ANN model is superior over the RSM model for its generalization capability of the GA extraction process
Ta b l e 6 ndash Analysis of variance (ANOVA) analysis for ANN model
Source DFa SSb MSc F-value p-value R2 R2(adj)
Model or Regression 60 11421 1904 1041 00004 0986 088Residual Error 9 1646 183Total 69 11587aDegree of Freedom bSum of Squares cMean Square
Ta b l e 7 ndash Unseen experimental dataset used in the developed RSM and ANN models for the prediction of ex-traction efficiency of GA
RunIndependent variables Response E ()
CGA0 (g Lndash1)
CTBP ( vv)
T(oC) pH0Experi- mental RSM ANN
1 16 15 11 174 9057 9015 91402 25 30 40 30 9005 8836 90373 25 23 37 234 8888 8723 88534 25 37 27 211 9226 9323 92275 25 45 27 185 9525 9531 94786 25 23 37 234 8798 8723 88537 14 30 40 301 8533 8967 87368 18 20 32 221 8745 8805 86639 10 30 24 26 9256 9169 925710 18 36 32 21 9302 9285 938911 10 30 40 15 8883 9118 8908
Ta b l e 8 ndash Comparison of RSM and ANN models for its gen-eralization capability (using unseen data)
Parameters RSM ANN
R2 061 0915RMSE 173 080
AARE 143 066 SEP 191 088
40
Fig 8 Comparison of RSM and ANN model predictions for the unseen data (data not used in
the model development)
F i g 8 ndash Comparison of RSM and ANN model predictions for the unseen data (data not used in the model development)
44 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
Conclusion
The present work suggests that gallic acid can be efficiently removed from aqueous streams using TBP in hexanol The RSM and ANN models were developed to predict the extraction efficiency of GA by varying the parameters temperature concentra-tion of TBP in aqueous phase pH and initial con-centration of gallic acid in aqueous solution The importance of each parameter and its synergistic interactive effects on the extraction of GA was ex-plained with a minimum number of experiments us-ing the RSM model Both models were efficient in the prediction of GA extraction efficiency having close agreement with the experimental extraction efficiency However ANN showed superiority over RSM in terms of accuracy but the marginal errors between the models were very low The optimal conditions to achieve maximum extraction as pre-dicted by the RSM model were CGA0 = 234 g Lndash1 CTBP = 6565 (vv) T = 184 oC and pH0 = 18 at which an experimental extraction efficiency of 995 was observed The models also showed bet-ter performance with the use of unseen data which confirmed their generalization capabilities Thus the present investigation on experiments model-
ling and optimization of the GA extraction process could be of great significance for the design and evaluation of the performance of similar systems
S u p p l e m e n t a r y I n f o r m a t i o n
Supplementary information for confirmation of 2 h shaking time for reactive extraction of gallic acid and retention time of 39 min in HPLC analysis
The trial experiments were conducted for ob-taining an optimum shaking period Equal volumes (20 mL) of aqueous and organic phases were shak-en for different time intervals (5ndash240 min) at 250 rpm Equilibrium was achieved after half an hour (Figure S1) Hence further experiments were con-ducted with 2 h shaking time (sufficient to achieve equilibrium)
Calibration curve and chromatogram for HPLC analysis of gallic acid was obtained by diluting the mother stock of 5 g Lndash1 to prepare different concen-trations (5ndash300 mg Lndash1) of standard sample solu-tions
UV absorption spectra for gallic acid was ob-tained using UV Spectrophotometer Shimadzu UV-1800
F i g S 1 ndash Experimental extraction efficiency (E) of gallic acid versus time for the operating conditions CGA0 = 2 g Lndash1 CTBP = 40 (vv) T = 30 oC pH0 = 3
F i g S 2 ndash Calibration curve for gallic acid obtained us- ing HPLC
44
Fig S4b HPLC Chromatogram of gallic acid at 264 nm and 396 min
F i g S 3 ndash HPLC Chromatogram of gallic acid at 264 nm and 396 min
Ext
ract
ion
effic
ienc
y E
()
Peak
are
a (m
AU
s)
Time (min) Gallic acid concentration (mg Lndash1)
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 45
N o m e n c l a t u r e
AARE ndash absolute average relative errorCGAaq ndash equilibrium concentration of GA in aqueous
phase g Lndash1
CGAorg ndash equilibrium concentration of GA in organic phase g Lndash1
CGA0 ndash initial concentration of GA in aqueous phase g Lndash1
CTBP ndash TBP concentration in organic phase DF ndash degree of freedom in ANOVA analysisE ndash extraction efficiency response F-value ndash Fischer distribution value (ratio of variances)H A O ndash number of neurons in hidden layer input lay-
er and output layer respectivelyIW ndash input weight matrixk ndash number of input variables or factors in
RO-CCDlogsig ndash sigmoidal transfer function for neural net-
workLW ndash layer weight matrixM ndash number of network connections including
weights and biasesMS ndash mean square in ANOVA analysisMSE ndash mean squared errorN ndash total number of experiments to construct
ANN modeln0 ndash centre points in RO-CCDnc ndash corner points or factorial points in RO-CCDns ndash star or axial points in RO-CCDpH0 ndash initial pH of aqueous GA solutionPRESS ndash predicted residual sum of squarespurelin ndash linear transfer function for neural networkp-value ndash null-hypothesis testR ndash coefficient of correlation
R2 ndash coefficient of determinationR2 (adj) ndash adjusted R-squaredR2 (pred) ndash predicted R-squaredRMSE ndash root mean square errorSEP ndash standard error of predictionSS ndash sum of squares in ANOVA analysisT ndash extraction temperature oCw ndash connection weight between layersx1 x2 x3 x4 ndash coded levels of independent vaiables in
RSMxij yij bij ndash network input output and bias of jth neu-
ron to ith layer respectivelya ndash distance of each star point from centre in
RO-CCDbi bj bij ndash regression coefficients representing linear
quadratic and interactive effects in RSM
R e f e r e n c e s
1 Benitez F J Real F J Acero J L Leal A I Garcia C Gallic acid degradation in aqueous solutions by UVH2O2 treatment Fentonrsquos reagent and the photo-Fenton sys-tem J Hazard Mater 126 (2005) 31doi httpsdoiorg101016jjhazmat200504040
2 El-Gohary F A Badawy M I El-Khateeb M A El-Kalliny A S Integrated treatment of olive mill wastewater (OMW) by the combination of fentonrsquos reaction and anaer-obic treatment J Hazard Mater 162 (2009) 1536doi httpsdoiorg101016jjhazmat200806098
3 Herrero M Cifuentes A Ibanez E Sub- and supercriti-cal fluid extraction of functional ingredients from different natural sources plants food-by-products algae and mi-croalgae A Review Food Chem 98 (2006) 136doi httpsdoiorg101016jfoodchem200505058
4 Kougan G B Tabopda T Kuete V Verpoorte R Sim-ple phenols phenolic acids and related esters from the me-dicinal plants of africa in Kuet V (First Ed) Medicinal plant research in Africa Pharmacology and chemistry El-sevier Inc London 2013 225ndash249
45
UV absorption spectra for gallic acid was obtained using UV Spectrophotometer Shimadzu UV-1800
Fig S3 Absorption spectra of gallic acid showing max = 264 nm
--
F i g S 4 ndash Absorption spectra of gallic acid showing lmax = 264 nm
46 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
5 Kapellakis I E Tsagarakis K P Crowther J C Olive oil history production and by-product management Rev Environ Sci Biotechnol 7 (2008) 1doi httpsdoiorg101007s11157-007-9120-9
6 Calace N Nardi E Petronio B Pietroletti M Adsorp-tion of phenols by papermill sludges Environ Pollut 118 (2002) 315doi httpsdoiorg101016S0269-7491(01)00303-7
7 Fki I Allouche N Sayadi S The use of polyphenolic extract purified hydroxytyrosol and 34-dihydroxyphenyl acetic acid from olive mill wastewater for the stabilization of refined oils A potential alternative to synthetic antioxi-dants Food Chem 93 (2005) 197doi httpsdoiorg101016jfoodchem200409014
8 Ollis D F Pelizzetti E Serpone N Photocatalyzed de-struction of water contaminants Environ Sci Technol 25 (1991) 1522doi httpsdoiorg101021es00021a001
9 Masten S J Davies S H The use of ozonation to de-grade organic contaminants in wastewaters Environ Sci Technol 28 (1994) 180Adoi httpsdoiorg101021es00053a001
10 Spigno G Jauregi P Recovery of gallic acid with colloi-dal gas aphrons (CGA) Int J Food Eng 1 (2005) 1doi httpsdoiorg1022021556-37581038
11 Spigno G Dermiki M Pastori C Casanova F Jauregi P Recovery of gallic acid with colloidal gas aphrons gen-erated from a cationic surfactant Sep Purif Technol 71 (2010) 56doi httpsdoiorg101016jseppur200911002
12 Quici N Litter M I Heterogeneous photocatalytic degra-dation of gallic acid under different experimental condi-tions Photochem Photobiol Sci 8 (2009) 975doi httpsdoiorg101039b901904a
13 Singh M Jha A Kumar A Hettiarachchy N Rai A K Sharma D Influence of the solvents on the extraction of major phenolic compounds (punicalagin ellagic acid and gallic acid) and their antioxidant activities in pomegranate aril J Food Sci Technol 51 (2014) 2070doi httpsdoiorg101007s13197-014-1267-0
14 Claudio A F M Ferreira A M Freire C S R Silves-tre A J D Freire M G Coutinho J A P Optimization of the gallic acid extraction using ionic-liquid-based aque-ous two-phase systems Sep Purif Technol 97 (2012) 142doi httpsdoiorg101016jseppur201202036
15 Li Y Wang Y Li Y Dai Y Extraction of glyoxylic acid glycolic acid acrylic acid and benzoic acid with trialkyl-phosphine oxide J Chem Eng Data 48 (2003) 621doi httpsdoiorg101021je020174r
16 Daneshfar A Ghaziaskar H S Homayoun N Solubility of gallic acid in methanol ethanol water and ethyl acetate J Chem Eng Data 53 (2008) 776doi httpsdoiorg101021je700633w
17 Wasewar K L Shende D Z Reactive extraction of caproic acid using tri- n -butyl phosphate in hexanol octa-nol and decanol J Chem Eng Data 56 (2011) 288doi httpsdoiorg101021je100974f
18 Keshav A Chand S Wasewar K L Reactive extraction of acrylic acid using tri- n -butyl phosphate in different di-luents J Chem Eng Data 54 (2009) 1782doi httpsdoiorg101021je800856e
19 Zhong K Wang Q Optimization of ultrasonic extraction of polysaccharides from dried longan pulp using response surface methodology Carbohydr Polym 80 (2010) 19doi httpsdoiorg101016jcarbpol200910066
20 Marchitan N Cojocaru C Mereuta A Duca G Crete-scu I Gonta M Modeling and optimization of tartaric acid reactive extraction from aqueous solutions A compar-ison between response surface methodology and artificial neural network Sep Purif Technol 75 (2010) 273doi httpsdoiorg101016jseppur201008016
21 Messikh N Samar M H Messikh L Neural network analysis of liquidndashliquid extraction of phenol from waste-water using tbp solvent Desalination 208 (2007) 42doi httpsdoiorg101016jdesal200604073
22 Montgomery D Design and analysis of experiments (Fifth Ed) John Wiley amp Sons Inc New Jersey 2001
23 Box G E P Hunter J S H W G Statistics for experi-menters design innovation and discovery (second ed) Vol 48 John Wiley ampSons New Jersey 2006
24 Dornier M Decloux M Trystram G Lebert A Dynam-ic modeling of crossflow microfiltration using neural net-works J Memb Sci 98 (1995) 263doi httpsdoiorg1010160376-7388(94)00195-5
25 Niemi H Bulsari A Palosaari S Simulation of mem-brane separation by neural networks J Memb Sci 102 (1995) 185doi httpsdoiorg1010160376-7388(94)00314-O
26 Bourquin J Schmidli H van Hoogevest P Leuenberger H Pitfalls of artificial neural networks (ANN) modelling technique for data sets containing outlier measurements us-ing a study on mixture properties of a direct compressed dosage form Eur J Pharm Sci 7 (1998) 17doi httpsdoiorg101016S0928-0987(97)10027-6
27 Cornell A I Khuri J A Response surfaces Designs and analyses (second ed) Marcel Dekker Inc New York 1996
28 Bezerra M A Santelli R E Oliveira E P Villar L S Escaleira L A Response surface methodology (RSM) as a tool for optimization in analytical chemistry Talanta 76 (2008) 965doi httpsdoiorg101016jtalanta200805019
29 Hornik K Stinchcombe M White H Multilayer feedfor-ward networks are universal approximators Neural Net-works 2 (1989) 359doi httpsdoiorg1010160893-6080(89)90020-8
30 Hagan M T Demuth H B Beale M H Neural network design PWS Publishing Company Boston 1995
31 Keshav A Wasewar K L Chand S Extraction of propi-onic acid using different extractants (tri-n-butylphosphate tri-n-octylamine and Aliquat 336) Ind Eng Chem Res 47 (2008) 6192doi httpsdoiorg101021ie800006r
32 Yang S T White S A Hsu S T Extraction of carboxyl-ic acids with tertiary and quaternary amines Effect of pH Ind Eng Chem Res 30 (1991) 1335doi httpsdoiorg101021ie00054a040
33 Keshav A Wasewar K L Chand S Extraction of Acryl-ic Propionic and butyric acid using Aliquat 336 in oleyl alcohol Equilibria and effect of temperature Ind Eng Chem Res 48 (2009) 888doi httpsdoiorg101021ie8010337
34 Uslu H Kırbaslar S I Effect of temperature and initial acid concentration on the reactive extraction of carboxylic acids J Chem Eng Data 58 (2013) 1822doi httpsdoiorg101021je4002202
35 Pham D T Sagiroglu S Training multilayered percep-trons for pattern recognition A comparative study of four training algorithms Int J Mach Tools Manuf 41 (2001) 419doi httpsdoiorg101016S0890-6955(00)00073-0
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 43
The ANN model was also validated using sta-tistical analysis ANOVA The degree of freedom due to regression (DFregression) and residuals (DFresidual) was calculated as follows20
regressionDF M 1= minus (17)
residualDF N M= minus (18)
M H(A O 1) O= + + + (19)
where M is the total number of network connec-tions including weights and biases N is the total number of experiments (N = 69) performed to con-struct and validate the ANN model Table 6 shows the ANOVA for the developed ANN model The F-value and the p-value were obtained from ANO-VA for the ANN model as 104 and 00004 respec-tively which confirms the significance of the ANN model Table 7 shows the values of R2 and R2 (adj) as 0986 and 088 respectively indicating the good fit of the ANN model to the experimental extraction efficiency Therefore the ANN model for predicting extraction efficiency E () for the extraction of GA can be written as follows
n n 1 2 (LW (IW ) )y purelin logsig x b b= times times + + (20)
Comparison of ANN and RSM models
The prediction and generalization capabilities of RSM and ANN models were evaluated based on its R2 RMSE AARE and SEP values Table 1 and 4 indicate that both the models predict well the ex-perimental extraction efficiency and their R2 values are closer to 1 Therefore both models can be con-sidered good for predicting the extraction efficiency of GA The generalization capabilities of both the models were compared only with unseen dataset Table 7 shows the unseen data set which were not used for the model development Figure 8 shows the comparative parity plot for RSM and ANN model predictability for the unseen data set and suggests that both the model predictions fit the ex-perimental data very well The comparative values of R2 RMSE AARE and SEP values for both mod-els using unseen data are shown in Table 8 Thus higher R2 and lower RMSE AARE SEP values for ANN indicate that the ANN model is superior over the RSM model for its generalization capability of the GA extraction process
Ta b l e 6 ndash Analysis of variance (ANOVA) analysis for ANN model
Source DFa SSb MSc F-value p-value R2 R2(adj)
Model or Regression 60 11421 1904 1041 00004 0986 088Residual Error 9 1646 183Total 69 11587aDegree of Freedom bSum of Squares cMean Square
Ta b l e 7 ndash Unseen experimental dataset used in the developed RSM and ANN models for the prediction of ex-traction efficiency of GA
RunIndependent variables Response E ()
CGA0 (g Lndash1)
CTBP ( vv)
T(oC) pH0Experi- mental RSM ANN
1 16 15 11 174 9057 9015 91402 25 30 40 30 9005 8836 90373 25 23 37 234 8888 8723 88534 25 37 27 211 9226 9323 92275 25 45 27 185 9525 9531 94786 25 23 37 234 8798 8723 88537 14 30 40 301 8533 8967 87368 18 20 32 221 8745 8805 86639 10 30 24 26 9256 9169 925710 18 36 32 21 9302 9285 938911 10 30 40 15 8883 9118 8908
Ta b l e 8 ndash Comparison of RSM and ANN models for its gen-eralization capability (using unseen data)
Parameters RSM ANN
R2 061 0915RMSE 173 080
AARE 143 066 SEP 191 088
40
Fig 8 Comparison of RSM and ANN model predictions for the unseen data (data not used in
the model development)
F i g 8 ndash Comparison of RSM and ANN model predictions for the unseen data (data not used in the model development)
44 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
Conclusion
The present work suggests that gallic acid can be efficiently removed from aqueous streams using TBP in hexanol The RSM and ANN models were developed to predict the extraction efficiency of GA by varying the parameters temperature concentra-tion of TBP in aqueous phase pH and initial con-centration of gallic acid in aqueous solution The importance of each parameter and its synergistic interactive effects on the extraction of GA was ex-plained with a minimum number of experiments us-ing the RSM model Both models were efficient in the prediction of GA extraction efficiency having close agreement with the experimental extraction efficiency However ANN showed superiority over RSM in terms of accuracy but the marginal errors between the models were very low The optimal conditions to achieve maximum extraction as pre-dicted by the RSM model were CGA0 = 234 g Lndash1 CTBP = 6565 (vv) T = 184 oC and pH0 = 18 at which an experimental extraction efficiency of 995 was observed The models also showed bet-ter performance with the use of unseen data which confirmed their generalization capabilities Thus the present investigation on experiments model-
ling and optimization of the GA extraction process could be of great significance for the design and evaluation of the performance of similar systems
S u p p l e m e n t a r y I n f o r m a t i o n
Supplementary information for confirmation of 2 h shaking time for reactive extraction of gallic acid and retention time of 39 min in HPLC analysis
The trial experiments were conducted for ob-taining an optimum shaking period Equal volumes (20 mL) of aqueous and organic phases were shak-en for different time intervals (5ndash240 min) at 250 rpm Equilibrium was achieved after half an hour (Figure S1) Hence further experiments were con-ducted with 2 h shaking time (sufficient to achieve equilibrium)
Calibration curve and chromatogram for HPLC analysis of gallic acid was obtained by diluting the mother stock of 5 g Lndash1 to prepare different concen-trations (5ndash300 mg Lndash1) of standard sample solu-tions
UV absorption spectra for gallic acid was ob-tained using UV Spectrophotometer Shimadzu UV-1800
F i g S 1 ndash Experimental extraction efficiency (E) of gallic acid versus time for the operating conditions CGA0 = 2 g Lndash1 CTBP = 40 (vv) T = 30 oC pH0 = 3
F i g S 2 ndash Calibration curve for gallic acid obtained us- ing HPLC
44
Fig S4b HPLC Chromatogram of gallic acid at 264 nm and 396 min
F i g S 3 ndash HPLC Chromatogram of gallic acid at 264 nm and 396 min
Ext
ract
ion
effic
ienc
y E
()
Peak
are
a (m
AU
s)
Time (min) Gallic acid concentration (mg Lndash1)
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 45
N o m e n c l a t u r e
AARE ndash absolute average relative errorCGAaq ndash equilibrium concentration of GA in aqueous
phase g Lndash1
CGAorg ndash equilibrium concentration of GA in organic phase g Lndash1
CGA0 ndash initial concentration of GA in aqueous phase g Lndash1
CTBP ndash TBP concentration in organic phase DF ndash degree of freedom in ANOVA analysisE ndash extraction efficiency response F-value ndash Fischer distribution value (ratio of variances)H A O ndash number of neurons in hidden layer input lay-
er and output layer respectivelyIW ndash input weight matrixk ndash number of input variables or factors in
RO-CCDlogsig ndash sigmoidal transfer function for neural net-
workLW ndash layer weight matrixM ndash number of network connections including
weights and biasesMS ndash mean square in ANOVA analysisMSE ndash mean squared errorN ndash total number of experiments to construct
ANN modeln0 ndash centre points in RO-CCDnc ndash corner points or factorial points in RO-CCDns ndash star or axial points in RO-CCDpH0 ndash initial pH of aqueous GA solutionPRESS ndash predicted residual sum of squarespurelin ndash linear transfer function for neural networkp-value ndash null-hypothesis testR ndash coefficient of correlation
R2 ndash coefficient of determinationR2 (adj) ndash adjusted R-squaredR2 (pred) ndash predicted R-squaredRMSE ndash root mean square errorSEP ndash standard error of predictionSS ndash sum of squares in ANOVA analysisT ndash extraction temperature oCw ndash connection weight between layersx1 x2 x3 x4 ndash coded levels of independent vaiables in
RSMxij yij bij ndash network input output and bias of jth neu-
ron to ith layer respectivelya ndash distance of each star point from centre in
RO-CCDbi bj bij ndash regression coefficients representing linear
quadratic and interactive effects in RSM
R e f e r e n c e s
1 Benitez F J Real F J Acero J L Leal A I Garcia C Gallic acid degradation in aqueous solutions by UVH2O2 treatment Fentonrsquos reagent and the photo-Fenton sys-tem J Hazard Mater 126 (2005) 31doi httpsdoiorg101016jjhazmat200504040
2 El-Gohary F A Badawy M I El-Khateeb M A El-Kalliny A S Integrated treatment of olive mill wastewater (OMW) by the combination of fentonrsquos reaction and anaer-obic treatment J Hazard Mater 162 (2009) 1536doi httpsdoiorg101016jjhazmat200806098
3 Herrero M Cifuentes A Ibanez E Sub- and supercriti-cal fluid extraction of functional ingredients from different natural sources plants food-by-products algae and mi-croalgae A Review Food Chem 98 (2006) 136doi httpsdoiorg101016jfoodchem200505058
4 Kougan G B Tabopda T Kuete V Verpoorte R Sim-ple phenols phenolic acids and related esters from the me-dicinal plants of africa in Kuet V (First Ed) Medicinal plant research in Africa Pharmacology and chemistry El-sevier Inc London 2013 225ndash249
45
UV absorption spectra for gallic acid was obtained using UV Spectrophotometer Shimadzu UV-1800
Fig S3 Absorption spectra of gallic acid showing max = 264 nm
--
F i g S 4 ndash Absorption spectra of gallic acid showing lmax = 264 nm
46 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
5 Kapellakis I E Tsagarakis K P Crowther J C Olive oil history production and by-product management Rev Environ Sci Biotechnol 7 (2008) 1doi httpsdoiorg101007s11157-007-9120-9
6 Calace N Nardi E Petronio B Pietroletti M Adsorp-tion of phenols by papermill sludges Environ Pollut 118 (2002) 315doi httpsdoiorg101016S0269-7491(01)00303-7
7 Fki I Allouche N Sayadi S The use of polyphenolic extract purified hydroxytyrosol and 34-dihydroxyphenyl acetic acid from olive mill wastewater for the stabilization of refined oils A potential alternative to synthetic antioxi-dants Food Chem 93 (2005) 197doi httpsdoiorg101016jfoodchem200409014
8 Ollis D F Pelizzetti E Serpone N Photocatalyzed de-struction of water contaminants Environ Sci Technol 25 (1991) 1522doi httpsdoiorg101021es00021a001
9 Masten S J Davies S H The use of ozonation to de-grade organic contaminants in wastewaters Environ Sci Technol 28 (1994) 180Adoi httpsdoiorg101021es00053a001
10 Spigno G Jauregi P Recovery of gallic acid with colloi-dal gas aphrons (CGA) Int J Food Eng 1 (2005) 1doi httpsdoiorg1022021556-37581038
11 Spigno G Dermiki M Pastori C Casanova F Jauregi P Recovery of gallic acid with colloidal gas aphrons gen-erated from a cationic surfactant Sep Purif Technol 71 (2010) 56doi httpsdoiorg101016jseppur200911002
12 Quici N Litter M I Heterogeneous photocatalytic degra-dation of gallic acid under different experimental condi-tions Photochem Photobiol Sci 8 (2009) 975doi httpsdoiorg101039b901904a
13 Singh M Jha A Kumar A Hettiarachchy N Rai A K Sharma D Influence of the solvents on the extraction of major phenolic compounds (punicalagin ellagic acid and gallic acid) and their antioxidant activities in pomegranate aril J Food Sci Technol 51 (2014) 2070doi httpsdoiorg101007s13197-014-1267-0
14 Claudio A F M Ferreira A M Freire C S R Silves-tre A J D Freire M G Coutinho J A P Optimization of the gallic acid extraction using ionic-liquid-based aque-ous two-phase systems Sep Purif Technol 97 (2012) 142doi httpsdoiorg101016jseppur201202036
15 Li Y Wang Y Li Y Dai Y Extraction of glyoxylic acid glycolic acid acrylic acid and benzoic acid with trialkyl-phosphine oxide J Chem Eng Data 48 (2003) 621doi httpsdoiorg101021je020174r
16 Daneshfar A Ghaziaskar H S Homayoun N Solubility of gallic acid in methanol ethanol water and ethyl acetate J Chem Eng Data 53 (2008) 776doi httpsdoiorg101021je700633w
17 Wasewar K L Shende D Z Reactive extraction of caproic acid using tri- n -butyl phosphate in hexanol octa-nol and decanol J Chem Eng Data 56 (2011) 288doi httpsdoiorg101021je100974f
18 Keshav A Chand S Wasewar K L Reactive extraction of acrylic acid using tri- n -butyl phosphate in different di-luents J Chem Eng Data 54 (2009) 1782doi httpsdoiorg101021je800856e
19 Zhong K Wang Q Optimization of ultrasonic extraction of polysaccharides from dried longan pulp using response surface methodology Carbohydr Polym 80 (2010) 19doi httpsdoiorg101016jcarbpol200910066
20 Marchitan N Cojocaru C Mereuta A Duca G Crete-scu I Gonta M Modeling and optimization of tartaric acid reactive extraction from aqueous solutions A compar-ison between response surface methodology and artificial neural network Sep Purif Technol 75 (2010) 273doi httpsdoiorg101016jseppur201008016
21 Messikh N Samar M H Messikh L Neural network analysis of liquidndashliquid extraction of phenol from waste-water using tbp solvent Desalination 208 (2007) 42doi httpsdoiorg101016jdesal200604073
22 Montgomery D Design and analysis of experiments (Fifth Ed) John Wiley amp Sons Inc New Jersey 2001
23 Box G E P Hunter J S H W G Statistics for experi-menters design innovation and discovery (second ed) Vol 48 John Wiley ampSons New Jersey 2006
24 Dornier M Decloux M Trystram G Lebert A Dynam-ic modeling of crossflow microfiltration using neural net-works J Memb Sci 98 (1995) 263doi httpsdoiorg1010160376-7388(94)00195-5
25 Niemi H Bulsari A Palosaari S Simulation of mem-brane separation by neural networks J Memb Sci 102 (1995) 185doi httpsdoiorg1010160376-7388(94)00314-O
26 Bourquin J Schmidli H van Hoogevest P Leuenberger H Pitfalls of artificial neural networks (ANN) modelling technique for data sets containing outlier measurements us-ing a study on mixture properties of a direct compressed dosage form Eur J Pharm Sci 7 (1998) 17doi httpsdoiorg101016S0928-0987(97)10027-6
27 Cornell A I Khuri J A Response surfaces Designs and analyses (second ed) Marcel Dekker Inc New York 1996
28 Bezerra M A Santelli R E Oliveira E P Villar L S Escaleira L A Response surface methodology (RSM) as a tool for optimization in analytical chemistry Talanta 76 (2008) 965doi httpsdoiorg101016jtalanta200805019
29 Hornik K Stinchcombe M White H Multilayer feedfor-ward networks are universal approximators Neural Net-works 2 (1989) 359doi httpsdoiorg1010160893-6080(89)90020-8
30 Hagan M T Demuth H B Beale M H Neural network design PWS Publishing Company Boston 1995
31 Keshav A Wasewar K L Chand S Extraction of propi-onic acid using different extractants (tri-n-butylphosphate tri-n-octylamine and Aliquat 336) Ind Eng Chem Res 47 (2008) 6192doi httpsdoiorg101021ie800006r
32 Yang S T White S A Hsu S T Extraction of carboxyl-ic acids with tertiary and quaternary amines Effect of pH Ind Eng Chem Res 30 (1991) 1335doi httpsdoiorg101021ie00054a040
33 Keshav A Wasewar K L Chand S Extraction of Acryl-ic Propionic and butyric acid using Aliquat 336 in oleyl alcohol Equilibria and effect of temperature Ind Eng Chem Res 48 (2009) 888doi httpsdoiorg101021ie8010337
34 Uslu H Kırbaslar S I Effect of temperature and initial acid concentration on the reactive extraction of carboxylic acids J Chem Eng Data 58 (2013) 1822doi httpsdoiorg101021je4002202
35 Pham D T Sagiroglu S Training multilayered percep-trons for pattern recognition A comparative study of four training algorithms Int J Mach Tools Manuf 41 (2001) 419doi httpsdoiorg101016S0890-6955(00)00073-0
44 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
Conclusion
The present work suggests that gallic acid can be efficiently removed from aqueous streams using TBP in hexanol The RSM and ANN models were developed to predict the extraction efficiency of GA by varying the parameters temperature concentra-tion of TBP in aqueous phase pH and initial con-centration of gallic acid in aqueous solution The importance of each parameter and its synergistic interactive effects on the extraction of GA was ex-plained with a minimum number of experiments us-ing the RSM model Both models were efficient in the prediction of GA extraction efficiency having close agreement with the experimental extraction efficiency However ANN showed superiority over RSM in terms of accuracy but the marginal errors between the models were very low The optimal conditions to achieve maximum extraction as pre-dicted by the RSM model were CGA0 = 234 g Lndash1 CTBP = 6565 (vv) T = 184 oC and pH0 = 18 at which an experimental extraction efficiency of 995 was observed The models also showed bet-ter performance with the use of unseen data which confirmed their generalization capabilities Thus the present investigation on experiments model-
ling and optimization of the GA extraction process could be of great significance for the design and evaluation of the performance of similar systems
S u p p l e m e n t a r y I n f o r m a t i o n
Supplementary information for confirmation of 2 h shaking time for reactive extraction of gallic acid and retention time of 39 min in HPLC analysis
The trial experiments were conducted for ob-taining an optimum shaking period Equal volumes (20 mL) of aqueous and organic phases were shak-en for different time intervals (5ndash240 min) at 250 rpm Equilibrium was achieved after half an hour (Figure S1) Hence further experiments were con-ducted with 2 h shaking time (sufficient to achieve equilibrium)
Calibration curve and chromatogram for HPLC analysis of gallic acid was obtained by diluting the mother stock of 5 g Lndash1 to prepare different concen-trations (5ndash300 mg Lndash1) of standard sample solu-tions
UV absorption spectra for gallic acid was ob-tained using UV Spectrophotometer Shimadzu UV-1800
F i g S 1 ndash Experimental extraction efficiency (E) of gallic acid versus time for the operating conditions CGA0 = 2 g Lndash1 CTBP = 40 (vv) T = 30 oC pH0 = 3
F i g S 2 ndash Calibration curve for gallic acid obtained us- ing HPLC
44
Fig S4b HPLC Chromatogram of gallic acid at 264 nm and 396 min
F i g S 3 ndash HPLC Chromatogram of gallic acid at 264 nm and 396 min
Ext
ract
ion
effic
ienc
y E
()
Peak
are
a (m
AU
s)
Time (min) Gallic acid concentration (mg Lndash1)
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 45
N o m e n c l a t u r e
AARE ndash absolute average relative errorCGAaq ndash equilibrium concentration of GA in aqueous
phase g Lndash1
CGAorg ndash equilibrium concentration of GA in organic phase g Lndash1
CGA0 ndash initial concentration of GA in aqueous phase g Lndash1
CTBP ndash TBP concentration in organic phase DF ndash degree of freedom in ANOVA analysisE ndash extraction efficiency response F-value ndash Fischer distribution value (ratio of variances)H A O ndash number of neurons in hidden layer input lay-
er and output layer respectivelyIW ndash input weight matrixk ndash number of input variables or factors in
RO-CCDlogsig ndash sigmoidal transfer function for neural net-
workLW ndash layer weight matrixM ndash number of network connections including
weights and biasesMS ndash mean square in ANOVA analysisMSE ndash mean squared errorN ndash total number of experiments to construct
ANN modeln0 ndash centre points in RO-CCDnc ndash corner points or factorial points in RO-CCDns ndash star or axial points in RO-CCDpH0 ndash initial pH of aqueous GA solutionPRESS ndash predicted residual sum of squarespurelin ndash linear transfer function for neural networkp-value ndash null-hypothesis testR ndash coefficient of correlation
R2 ndash coefficient of determinationR2 (adj) ndash adjusted R-squaredR2 (pred) ndash predicted R-squaredRMSE ndash root mean square errorSEP ndash standard error of predictionSS ndash sum of squares in ANOVA analysisT ndash extraction temperature oCw ndash connection weight between layersx1 x2 x3 x4 ndash coded levels of independent vaiables in
RSMxij yij bij ndash network input output and bias of jth neu-
ron to ith layer respectivelya ndash distance of each star point from centre in
RO-CCDbi bj bij ndash regression coefficients representing linear
quadratic and interactive effects in RSM
R e f e r e n c e s
1 Benitez F J Real F J Acero J L Leal A I Garcia C Gallic acid degradation in aqueous solutions by UVH2O2 treatment Fentonrsquos reagent and the photo-Fenton sys-tem J Hazard Mater 126 (2005) 31doi httpsdoiorg101016jjhazmat200504040
2 El-Gohary F A Badawy M I El-Khateeb M A El-Kalliny A S Integrated treatment of olive mill wastewater (OMW) by the combination of fentonrsquos reaction and anaer-obic treatment J Hazard Mater 162 (2009) 1536doi httpsdoiorg101016jjhazmat200806098
3 Herrero M Cifuentes A Ibanez E Sub- and supercriti-cal fluid extraction of functional ingredients from different natural sources plants food-by-products algae and mi-croalgae A Review Food Chem 98 (2006) 136doi httpsdoiorg101016jfoodchem200505058
4 Kougan G B Tabopda T Kuete V Verpoorte R Sim-ple phenols phenolic acids and related esters from the me-dicinal plants of africa in Kuet V (First Ed) Medicinal plant research in Africa Pharmacology and chemistry El-sevier Inc London 2013 225ndash249
45
UV absorption spectra for gallic acid was obtained using UV Spectrophotometer Shimadzu UV-1800
Fig S3 Absorption spectra of gallic acid showing max = 264 nm
--
F i g S 4 ndash Absorption spectra of gallic acid showing lmax = 264 nm
46 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
5 Kapellakis I E Tsagarakis K P Crowther J C Olive oil history production and by-product management Rev Environ Sci Biotechnol 7 (2008) 1doi httpsdoiorg101007s11157-007-9120-9
6 Calace N Nardi E Petronio B Pietroletti M Adsorp-tion of phenols by papermill sludges Environ Pollut 118 (2002) 315doi httpsdoiorg101016S0269-7491(01)00303-7
7 Fki I Allouche N Sayadi S The use of polyphenolic extract purified hydroxytyrosol and 34-dihydroxyphenyl acetic acid from olive mill wastewater for the stabilization of refined oils A potential alternative to synthetic antioxi-dants Food Chem 93 (2005) 197doi httpsdoiorg101016jfoodchem200409014
8 Ollis D F Pelizzetti E Serpone N Photocatalyzed de-struction of water contaminants Environ Sci Technol 25 (1991) 1522doi httpsdoiorg101021es00021a001
9 Masten S J Davies S H The use of ozonation to de-grade organic contaminants in wastewaters Environ Sci Technol 28 (1994) 180Adoi httpsdoiorg101021es00053a001
10 Spigno G Jauregi P Recovery of gallic acid with colloi-dal gas aphrons (CGA) Int J Food Eng 1 (2005) 1doi httpsdoiorg1022021556-37581038
11 Spigno G Dermiki M Pastori C Casanova F Jauregi P Recovery of gallic acid with colloidal gas aphrons gen-erated from a cationic surfactant Sep Purif Technol 71 (2010) 56doi httpsdoiorg101016jseppur200911002
12 Quici N Litter M I Heterogeneous photocatalytic degra-dation of gallic acid under different experimental condi-tions Photochem Photobiol Sci 8 (2009) 975doi httpsdoiorg101039b901904a
13 Singh M Jha A Kumar A Hettiarachchy N Rai A K Sharma D Influence of the solvents on the extraction of major phenolic compounds (punicalagin ellagic acid and gallic acid) and their antioxidant activities in pomegranate aril J Food Sci Technol 51 (2014) 2070doi httpsdoiorg101007s13197-014-1267-0
14 Claudio A F M Ferreira A M Freire C S R Silves-tre A J D Freire M G Coutinho J A P Optimization of the gallic acid extraction using ionic-liquid-based aque-ous two-phase systems Sep Purif Technol 97 (2012) 142doi httpsdoiorg101016jseppur201202036
15 Li Y Wang Y Li Y Dai Y Extraction of glyoxylic acid glycolic acid acrylic acid and benzoic acid with trialkyl-phosphine oxide J Chem Eng Data 48 (2003) 621doi httpsdoiorg101021je020174r
16 Daneshfar A Ghaziaskar H S Homayoun N Solubility of gallic acid in methanol ethanol water and ethyl acetate J Chem Eng Data 53 (2008) 776doi httpsdoiorg101021je700633w
17 Wasewar K L Shende D Z Reactive extraction of caproic acid using tri- n -butyl phosphate in hexanol octa-nol and decanol J Chem Eng Data 56 (2011) 288doi httpsdoiorg101021je100974f
18 Keshav A Chand S Wasewar K L Reactive extraction of acrylic acid using tri- n -butyl phosphate in different di-luents J Chem Eng Data 54 (2009) 1782doi httpsdoiorg101021je800856e
19 Zhong K Wang Q Optimization of ultrasonic extraction of polysaccharides from dried longan pulp using response surface methodology Carbohydr Polym 80 (2010) 19doi httpsdoiorg101016jcarbpol200910066
20 Marchitan N Cojocaru C Mereuta A Duca G Crete-scu I Gonta M Modeling and optimization of tartaric acid reactive extraction from aqueous solutions A compar-ison between response surface methodology and artificial neural network Sep Purif Technol 75 (2010) 273doi httpsdoiorg101016jseppur201008016
21 Messikh N Samar M H Messikh L Neural network analysis of liquidndashliquid extraction of phenol from waste-water using tbp solvent Desalination 208 (2007) 42doi httpsdoiorg101016jdesal200604073
22 Montgomery D Design and analysis of experiments (Fifth Ed) John Wiley amp Sons Inc New Jersey 2001
23 Box G E P Hunter J S H W G Statistics for experi-menters design innovation and discovery (second ed) Vol 48 John Wiley ampSons New Jersey 2006
24 Dornier M Decloux M Trystram G Lebert A Dynam-ic modeling of crossflow microfiltration using neural net-works J Memb Sci 98 (1995) 263doi httpsdoiorg1010160376-7388(94)00195-5
25 Niemi H Bulsari A Palosaari S Simulation of mem-brane separation by neural networks J Memb Sci 102 (1995) 185doi httpsdoiorg1010160376-7388(94)00314-O
26 Bourquin J Schmidli H van Hoogevest P Leuenberger H Pitfalls of artificial neural networks (ANN) modelling technique for data sets containing outlier measurements us-ing a study on mixture properties of a direct compressed dosage form Eur J Pharm Sci 7 (1998) 17doi httpsdoiorg101016S0928-0987(97)10027-6
27 Cornell A I Khuri J A Response surfaces Designs and analyses (second ed) Marcel Dekker Inc New York 1996
28 Bezerra M A Santelli R E Oliveira E P Villar L S Escaleira L A Response surface methodology (RSM) as a tool for optimization in analytical chemistry Talanta 76 (2008) 965doi httpsdoiorg101016jtalanta200805019
29 Hornik K Stinchcombe M White H Multilayer feedfor-ward networks are universal approximators Neural Net-works 2 (1989) 359doi httpsdoiorg1010160893-6080(89)90020-8
30 Hagan M T Demuth H B Beale M H Neural network design PWS Publishing Company Boston 1995
31 Keshav A Wasewar K L Chand S Extraction of propi-onic acid using different extractants (tri-n-butylphosphate tri-n-octylamine and Aliquat 336) Ind Eng Chem Res 47 (2008) 6192doi httpsdoiorg101021ie800006r
32 Yang S T White S A Hsu S T Extraction of carboxyl-ic acids with tertiary and quaternary amines Effect of pH Ind Eng Chem Res 30 (1991) 1335doi httpsdoiorg101021ie00054a040
33 Keshav A Wasewar K L Chand S Extraction of Acryl-ic Propionic and butyric acid using Aliquat 336 in oleyl alcohol Equilibria and effect of temperature Ind Eng Chem Res 48 (2009) 888doi httpsdoiorg101021ie8010337
34 Uslu H Kırbaslar S I Effect of temperature and initial acid concentration on the reactive extraction of carboxylic acids J Chem Eng Data 58 (2013) 1822doi httpsdoiorg101021je4002202
35 Pham D T Sagiroglu S Training multilayered percep-trons for pattern recognition A comparative study of four training algorithms Int J Mach Tools Manuf 41 (2001) 419doi httpsdoiorg101016S0890-6955(00)00073-0
K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017) 45
N o m e n c l a t u r e
AARE ndash absolute average relative errorCGAaq ndash equilibrium concentration of GA in aqueous
phase g Lndash1
CGAorg ndash equilibrium concentration of GA in organic phase g Lndash1
CGA0 ndash initial concentration of GA in aqueous phase g Lndash1
CTBP ndash TBP concentration in organic phase DF ndash degree of freedom in ANOVA analysisE ndash extraction efficiency response F-value ndash Fischer distribution value (ratio of variances)H A O ndash number of neurons in hidden layer input lay-
er and output layer respectivelyIW ndash input weight matrixk ndash number of input variables or factors in
RO-CCDlogsig ndash sigmoidal transfer function for neural net-
workLW ndash layer weight matrixM ndash number of network connections including
weights and biasesMS ndash mean square in ANOVA analysisMSE ndash mean squared errorN ndash total number of experiments to construct
ANN modeln0 ndash centre points in RO-CCDnc ndash corner points or factorial points in RO-CCDns ndash star or axial points in RO-CCDpH0 ndash initial pH of aqueous GA solutionPRESS ndash predicted residual sum of squarespurelin ndash linear transfer function for neural networkp-value ndash null-hypothesis testR ndash coefficient of correlation
R2 ndash coefficient of determinationR2 (adj) ndash adjusted R-squaredR2 (pred) ndash predicted R-squaredRMSE ndash root mean square errorSEP ndash standard error of predictionSS ndash sum of squares in ANOVA analysisT ndash extraction temperature oCw ndash connection weight between layersx1 x2 x3 x4 ndash coded levels of independent vaiables in
RSMxij yij bij ndash network input output and bias of jth neu-
ron to ith layer respectivelya ndash distance of each star point from centre in
RO-CCDbi bj bij ndash regression coefficients representing linear
quadratic and interactive effects in RSM
R e f e r e n c e s
1 Benitez F J Real F J Acero J L Leal A I Garcia C Gallic acid degradation in aqueous solutions by UVH2O2 treatment Fentonrsquos reagent and the photo-Fenton sys-tem J Hazard Mater 126 (2005) 31doi httpsdoiorg101016jjhazmat200504040
2 El-Gohary F A Badawy M I El-Khateeb M A El-Kalliny A S Integrated treatment of olive mill wastewater (OMW) by the combination of fentonrsquos reaction and anaer-obic treatment J Hazard Mater 162 (2009) 1536doi httpsdoiorg101016jjhazmat200806098
3 Herrero M Cifuentes A Ibanez E Sub- and supercriti-cal fluid extraction of functional ingredients from different natural sources plants food-by-products algae and mi-croalgae A Review Food Chem 98 (2006) 136doi httpsdoiorg101016jfoodchem200505058
4 Kougan G B Tabopda T Kuete V Verpoorte R Sim-ple phenols phenolic acids and related esters from the me-dicinal plants of africa in Kuet V (First Ed) Medicinal plant research in Africa Pharmacology and chemistry El-sevier Inc London 2013 225ndash249
45
UV absorption spectra for gallic acid was obtained using UV Spectrophotometer Shimadzu UV-1800
Fig S3 Absorption spectra of gallic acid showing max = 264 nm
--
F i g S 4 ndash Absorption spectra of gallic acid showing lmax = 264 nm
46 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
5 Kapellakis I E Tsagarakis K P Crowther J C Olive oil history production and by-product management Rev Environ Sci Biotechnol 7 (2008) 1doi httpsdoiorg101007s11157-007-9120-9
6 Calace N Nardi E Petronio B Pietroletti M Adsorp-tion of phenols by papermill sludges Environ Pollut 118 (2002) 315doi httpsdoiorg101016S0269-7491(01)00303-7
7 Fki I Allouche N Sayadi S The use of polyphenolic extract purified hydroxytyrosol and 34-dihydroxyphenyl acetic acid from olive mill wastewater for the stabilization of refined oils A potential alternative to synthetic antioxi-dants Food Chem 93 (2005) 197doi httpsdoiorg101016jfoodchem200409014
8 Ollis D F Pelizzetti E Serpone N Photocatalyzed de-struction of water contaminants Environ Sci Technol 25 (1991) 1522doi httpsdoiorg101021es00021a001
9 Masten S J Davies S H The use of ozonation to de-grade organic contaminants in wastewaters Environ Sci Technol 28 (1994) 180Adoi httpsdoiorg101021es00053a001
10 Spigno G Jauregi P Recovery of gallic acid with colloi-dal gas aphrons (CGA) Int J Food Eng 1 (2005) 1doi httpsdoiorg1022021556-37581038
11 Spigno G Dermiki M Pastori C Casanova F Jauregi P Recovery of gallic acid with colloidal gas aphrons gen-erated from a cationic surfactant Sep Purif Technol 71 (2010) 56doi httpsdoiorg101016jseppur200911002
12 Quici N Litter M I Heterogeneous photocatalytic degra-dation of gallic acid under different experimental condi-tions Photochem Photobiol Sci 8 (2009) 975doi httpsdoiorg101039b901904a
13 Singh M Jha A Kumar A Hettiarachchy N Rai A K Sharma D Influence of the solvents on the extraction of major phenolic compounds (punicalagin ellagic acid and gallic acid) and their antioxidant activities in pomegranate aril J Food Sci Technol 51 (2014) 2070doi httpsdoiorg101007s13197-014-1267-0
14 Claudio A F M Ferreira A M Freire C S R Silves-tre A J D Freire M G Coutinho J A P Optimization of the gallic acid extraction using ionic-liquid-based aque-ous two-phase systems Sep Purif Technol 97 (2012) 142doi httpsdoiorg101016jseppur201202036
15 Li Y Wang Y Li Y Dai Y Extraction of glyoxylic acid glycolic acid acrylic acid and benzoic acid with trialkyl-phosphine oxide J Chem Eng Data 48 (2003) 621doi httpsdoiorg101021je020174r
16 Daneshfar A Ghaziaskar H S Homayoun N Solubility of gallic acid in methanol ethanol water and ethyl acetate J Chem Eng Data 53 (2008) 776doi httpsdoiorg101021je700633w
17 Wasewar K L Shende D Z Reactive extraction of caproic acid using tri- n -butyl phosphate in hexanol octa-nol and decanol J Chem Eng Data 56 (2011) 288doi httpsdoiorg101021je100974f
18 Keshav A Chand S Wasewar K L Reactive extraction of acrylic acid using tri- n -butyl phosphate in different di-luents J Chem Eng Data 54 (2009) 1782doi httpsdoiorg101021je800856e
19 Zhong K Wang Q Optimization of ultrasonic extraction of polysaccharides from dried longan pulp using response surface methodology Carbohydr Polym 80 (2010) 19doi httpsdoiorg101016jcarbpol200910066
20 Marchitan N Cojocaru C Mereuta A Duca G Crete-scu I Gonta M Modeling and optimization of tartaric acid reactive extraction from aqueous solutions A compar-ison between response surface methodology and artificial neural network Sep Purif Technol 75 (2010) 273doi httpsdoiorg101016jseppur201008016
21 Messikh N Samar M H Messikh L Neural network analysis of liquidndashliquid extraction of phenol from waste-water using tbp solvent Desalination 208 (2007) 42doi httpsdoiorg101016jdesal200604073
22 Montgomery D Design and analysis of experiments (Fifth Ed) John Wiley amp Sons Inc New Jersey 2001
23 Box G E P Hunter J S H W G Statistics for experi-menters design innovation and discovery (second ed) Vol 48 John Wiley ampSons New Jersey 2006
24 Dornier M Decloux M Trystram G Lebert A Dynam-ic modeling of crossflow microfiltration using neural net-works J Memb Sci 98 (1995) 263doi httpsdoiorg1010160376-7388(94)00195-5
25 Niemi H Bulsari A Palosaari S Simulation of mem-brane separation by neural networks J Memb Sci 102 (1995) 185doi httpsdoiorg1010160376-7388(94)00314-O
26 Bourquin J Schmidli H van Hoogevest P Leuenberger H Pitfalls of artificial neural networks (ANN) modelling technique for data sets containing outlier measurements us-ing a study on mixture properties of a direct compressed dosage form Eur J Pharm Sci 7 (1998) 17doi httpsdoiorg101016S0928-0987(97)10027-6
27 Cornell A I Khuri J A Response surfaces Designs and analyses (second ed) Marcel Dekker Inc New York 1996
28 Bezerra M A Santelli R E Oliveira E P Villar L S Escaleira L A Response surface methodology (RSM) as a tool for optimization in analytical chemistry Talanta 76 (2008) 965doi httpsdoiorg101016jtalanta200805019
29 Hornik K Stinchcombe M White H Multilayer feedfor-ward networks are universal approximators Neural Net-works 2 (1989) 359doi httpsdoiorg1010160893-6080(89)90020-8
30 Hagan M T Demuth H B Beale M H Neural network design PWS Publishing Company Boston 1995
31 Keshav A Wasewar K L Chand S Extraction of propi-onic acid using different extractants (tri-n-butylphosphate tri-n-octylamine and Aliquat 336) Ind Eng Chem Res 47 (2008) 6192doi httpsdoiorg101021ie800006r
32 Yang S T White S A Hsu S T Extraction of carboxyl-ic acids with tertiary and quaternary amines Effect of pH Ind Eng Chem Res 30 (1991) 1335doi httpsdoiorg101021ie00054a040
33 Keshav A Wasewar K L Chand S Extraction of Acryl-ic Propionic and butyric acid using Aliquat 336 in oleyl alcohol Equilibria and effect of temperature Ind Eng Chem Res 48 (2009) 888doi httpsdoiorg101021ie8010337
34 Uslu H Kırbaslar S I Effect of temperature and initial acid concentration on the reactive extraction of carboxylic acids J Chem Eng Data 58 (2013) 1822doi httpsdoiorg101021je4002202
35 Pham D T Sagiroglu S Training multilayered percep-trons for pattern recognition A comparative study of four training algorithms Int J Mach Tools Manuf 41 (2001) 419doi httpsdoiorg101016S0890-6955(00)00073-0
46 K Rewatkar et al Reactive Separation of Gallic Acid Experimentation and Optimizationhellip Chem Biochem Eng Q 31 (1) 33ndash46 (2017)
5 Kapellakis I E Tsagarakis K P Crowther J C Olive oil history production and by-product management Rev Environ Sci Biotechnol 7 (2008) 1doi httpsdoiorg101007s11157-007-9120-9
6 Calace N Nardi E Petronio B Pietroletti M Adsorp-tion of phenols by papermill sludges Environ Pollut 118 (2002) 315doi httpsdoiorg101016S0269-7491(01)00303-7
7 Fki I Allouche N Sayadi S The use of polyphenolic extract purified hydroxytyrosol and 34-dihydroxyphenyl acetic acid from olive mill wastewater for the stabilization of refined oils A potential alternative to synthetic antioxi-dants Food Chem 93 (2005) 197doi httpsdoiorg101016jfoodchem200409014
8 Ollis D F Pelizzetti E Serpone N Photocatalyzed de-struction of water contaminants Environ Sci Technol 25 (1991) 1522doi httpsdoiorg101021es00021a001
9 Masten S J Davies S H The use of ozonation to de-grade organic contaminants in wastewaters Environ Sci Technol 28 (1994) 180Adoi httpsdoiorg101021es00053a001
10 Spigno G Jauregi P Recovery of gallic acid with colloi-dal gas aphrons (CGA) Int J Food Eng 1 (2005) 1doi httpsdoiorg1022021556-37581038
11 Spigno G Dermiki M Pastori C Casanova F Jauregi P Recovery of gallic acid with colloidal gas aphrons gen-erated from a cationic surfactant Sep Purif Technol 71 (2010) 56doi httpsdoiorg101016jseppur200911002
12 Quici N Litter M I Heterogeneous photocatalytic degra-dation of gallic acid under different experimental condi-tions Photochem Photobiol Sci 8 (2009) 975doi httpsdoiorg101039b901904a
13 Singh M Jha A Kumar A Hettiarachchy N Rai A K Sharma D Influence of the solvents on the extraction of major phenolic compounds (punicalagin ellagic acid and gallic acid) and their antioxidant activities in pomegranate aril J Food Sci Technol 51 (2014) 2070doi httpsdoiorg101007s13197-014-1267-0
14 Claudio A F M Ferreira A M Freire C S R Silves-tre A J D Freire M G Coutinho J A P Optimization of the gallic acid extraction using ionic-liquid-based aque-ous two-phase systems Sep Purif Technol 97 (2012) 142doi httpsdoiorg101016jseppur201202036
15 Li Y Wang Y Li Y Dai Y Extraction of glyoxylic acid glycolic acid acrylic acid and benzoic acid with trialkyl-phosphine oxide J Chem Eng Data 48 (2003) 621doi httpsdoiorg101021je020174r
16 Daneshfar A Ghaziaskar H S Homayoun N Solubility of gallic acid in methanol ethanol water and ethyl acetate J Chem Eng Data 53 (2008) 776doi httpsdoiorg101021je700633w
17 Wasewar K L Shende D Z Reactive extraction of caproic acid using tri- n -butyl phosphate in hexanol octa-nol and decanol J Chem Eng Data 56 (2011) 288doi httpsdoiorg101021je100974f
18 Keshav A Chand S Wasewar K L Reactive extraction of acrylic acid using tri- n -butyl phosphate in different di-luents J Chem Eng Data 54 (2009) 1782doi httpsdoiorg101021je800856e
19 Zhong K Wang Q Optimization of ultrasonic extraction of polysaccharides from dried longan pulp using response surface methodology Carbohydr Polym 80 (2010) 19doi httpsdoiorg101016jcarbpol200910066
20 Marchitan N Cojocaru C Mereuta A Duca G Crete-scu I Gonta M Modeling and optimization of tartaric acid reactive extraction from aqueous solutions A compar-ison between response surface methodology and artificial neural network Sep Purif Technol 75 (2010) 273doi httpsdoiorg101016jseppur201008016
21 Messikh N Samar M H Messikh L Neural network analysis of liquidndashliquid extraction of phenol from waste-water using tbp solvent Desalination 208 (2007) 42doi httpsdoiorg101016jdesal200604073
22 Montgomery D Design and analysis of experiments (Fifth Ed) John Wiley amp Sons Inc New Jersey 2001
23 Box G E P Hunter J S H W G Statistics for experi-menters design innovation and discovery (second ed) Vol 48 John Wiley ampSons New Jersey 2006
24 Dornier M Decloux M Trystram G Lebert A Dynam-ic modeling of crossflow microfiltration using neural net-works J Memb Sci 98 (1995) 263doi httpsdoiorg1010160376-7388(94)00195-5
25 Niemi H Bulsari A Palosaari S Simulation of mem-brane separation by neural networks J Memb Sci 102 (1995) 185doi httpsdoiorg1010160376-7388(94)00314-O
26 Bourquin J Schmidli H van Hoogevest P Leuenberger H Pitfalls of artificial neural networks (ANN) modelling technique for data sets containing outlier measurements us-ing a study on mixture properties of a direct compressed dosage form Eur J Pharm Sci 7 (1998) 17doi httpsdoiorg101016S0928-0987(97)10027-6
27 Cornell A I Khuri J A Response surfaces Designs and analyses (second ed) Marcel Dekker Inc New York 1996
28 Bezerra M A Santelli R E Oliveira E P Villar L S Escaleira L A Response surface methodology (RSM) as a tool for optimization in analytical chemistry Talanta 76 (2008) 965doi httpsdoiorg101016jtalanta200805019
29 Hornik K Stinchcombe M White H Multilayer feedfor-ward networks are universal approximators Neural Net-works 2 (1989) 359doi httpsdoiorg1010160893-6080(89)90020-8
30 Hagan M T Demuth H B Beale M H Neural network design PWS Publishing Company Boston 1995
31 Keshav A Wasewar K L Chand S Extraction of propi-onic acid using different extractants (tri-n-butylphosphate tri-n-octylamine and Aliquat 336) Ind Eng Chem Res 47 (2008) 6192doi httpsdoiorg101021ie800006r
32 Yang S T White S A Hsu S T Extraction of carboxyl-ic acids with tertiary and quaternary amines Effect of pH Ind Eng Chem Res 30 (1991) 1335doi httpsdoiorg101021ie00054a040
33 Keshav A Wasewar K L Chand S Extraction of Acryl-ic Propionic and butyric acid using Aliquat 336 in oleyl alcohol Equilibria and effect of temperature Ind Eng Chem Res 48 (2009) 888doi httpsdoiorg101021ie8010337
34 Uslu H Kırbaslar S I Effect of temperature and initial acid concentration on the reactive extraction of carboxylic acids J Chem Eng Data 58 (2013) 1822doi httpsdoiorg101021je4002202
35 Pham D T Sagiroglu S Training multilayered percep-trons for pattern recognition A comparative study of four training algorithms Int J Mach Tools Manuf 41 (2001) 419doi httpsdoiorg101016S0890-6955(00)00073-0