Electrical conductivity of ammonium and phosphonium based deep eutectic solvents: Measurements and artificial intelligence-based prediction

Fluid Phase Equilibria 356 (2013) 30–37

Contents lists available at ScienceDirect

Fluid Phase Equilibria journal homepage: www.elsevier.com/locate/fluid

Electrical conductivity of ammonium and phosphonium based deep eutectic solvents: Measurements and artificial intelligence-based prediction F.S. Ghareh Bagh a , K. Shahbaz b , F.S. Mjalli c,∗ , I.M. AlNashef d , M.A. Hashim a a

Chemical Engineering Department, University of Malaya, Kuala Lumpur, Malaysia School of Engineering, Taylor’s University, Selangor, Malaysia c Petroleum and Chemical Engineering Department, Sultan Qaboos University, Muscat, Oman d Chemical Engineering Department, King Saud University, Riyadh, Saudi Arabia b

a r t i c l e

i n f o

Article history: Received 1 March 2013 Received in revised form 2 June 2013 Accepted 4 July 2013 Available online 17 July 2013 Keywords: Deep eutectic solvents Electrical conductivity Artificial neural network Phosphonium Ammonium

a b s t r a c t The evaluation of deep eutectic solvents (DESs) as a new generation of solvents for various practical application requires an insight of the main physical, chemical, and thermodynamic properties. In this study, the experimental measurements of the electrical conductivity of two classes of DESs based on ammonium and phosphonium salts at different compositions and temperatures were reported. The results revealed that electrical conductivity of DESs has temperature-dependency. In addition, molar conductivities of ammonium and phosphonium salts in DESs were obtained using DESs experimental values of electrical conductivities. The feasibility of using an artificial neural network (ANN) model to predict the electrical conductivity of ammonium and phosphonium based DESs at different temperatures and compositions was also examined. A feed-forward back propagation neural network with 8 hidden neurons was successfully developed and trained with the measured electrical conductivity data. The results indicated that among the different networks tested, the network with 8 hidden neurons had the best prediction performance and gave the smallest value of Normalized Mean Square Error (NMSE) (0.0010) and acceptable values of Index of Agreement (IA) (0.9999) and Regression Coefficient (R2 ) (0.9988). The comparison of the predicted electrical conductivity of DESs by the proposed model with those obtained by experiments confirmed the reliability of the ANN model with an average absolute relative deviation (AARD%) of 4.40%. © 2013 Elsevier B.V. All rights reserved.

1. Introduction Ionic liquids (ILs) are salts with weak ionic interaction which allows them to be liquid in ambient temperature (typically below 373.15 K). The scientific and significant importance of ILs have spanned a broad range of applications, owing to their tempting physicochemical properties, such as thermal and chemical stability, low melting point, negligible volatility, high ionic conductivity, moderate viscosity, high polarity, and solubility (affinity) with many compounds [1–5]. Their potential use in a variety of chemical and industrial applications as green solvents has been greatly explored [1,3]. Nevertheless, ILs are too expensive to be used in bulk applications since they cannot be well prepared at the laboratory with one step of synthesis. Due to the multi-stage purification processes required to purify the ILs after their synthesis, their production cost is quite high. Consequently, researchers prefer to buy

∗ Corresponding author. Tel.: +968 2414 2558; fax: +968 2414 1354. E-mail address: [email protected] (F.S. Mjalli). 0378-3812/$ – see front matter © 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.fluid.2013.07.012

them than to synthesize them locally. This imposes a constraint on using them as a viable and practical industrial chemical ingredient [6,7]. Fortunately, a low cost alternative for ILs is available. Deep eutectic solvents (DESs) belong to a class of ionic liquids which are mixtures of a quaternary salt with a metal halide (Lewis acid), a hydrated salt, or an ordinary hydrogen bond donor (HBD) such as alcohol, amide as well as carboxylic acid as complexing agent. This results in the formation of an eutectic mixture with a melting point that is considerably lower than its original precursors. For this reason, this mixture is called a DES. Moreover, DESs overcome some principal disadvantages from ILs, they are easy to prepare in pure state, non-reactive with water, fairly safe (when carefully designed from benign components) and biodegradable [8–12]. Recently, few research groups reported the synthesis and use of DESs in different applications. Abbott research group was the first to report the synthesis and use of ammonium-based DESs in different promising applications [9]. They described for the first time the electrodeposition of composite materials using DESs [13]. Kareem et al. [14] reported some important physical properties of

F.S.G. Bagh et al. / Fluid Phase Equilibria 356 (2013) 30–37

phosphonium-based DESs and used them for the potential use in the separation of aromatics from naphtha feed streams [15]. Shahbaz et al. [16] used phosphonium-based DESs to remove glycerol from palm oil-based biodiesel. They also revealed that DESs were successful for the removal of residual catalyst and water from palm oil-based biodiesel [17]. A recent review by Zhang et al. [8] gives a comprehensive overview of the recent applications of DESs. The evaluation of DESs as new generation of solvents for various practical applications requires enough knowledge of some of the main physical, chemical, and thermodynamic properties. Few research articles were recently added to the literature dealing with this topic [18–25], yet, the door is still wide-open for more research in this area. The electrical conductivity is one of fundamental physical properties that represent how well a material can conduct electrical current. It is also an indication for how much a material is resistive for the motion of electrons within its molecules (resistivity). In addition, it is a useful measurement in a variety of industries [14,26]. The electrical conductivity is also one of the important electrical properties of electrolytes, such as carrier concentration, carrier mobility, and the longevity of excess carriers. In engineering applications, it is crucial to measure the electrical conductivity of an electrolyte in order to design, control and optimize the electrolysis processes and the production of electrochemical power sources. For corrosion protection, electrical conductivity provides practical information for assessing the corrosivity of aqueous media and for cathodic protection system. It is also used to gain insight into the properties of electrolyte solutions and to evaluate the characteristic quantities such as dissociation constants [27–30]. Therefore, there is a significant need to measure the DESs’ electrical conductivity and develop a reliable technique to predict it in order to properly manage their future applications. It was found that using predictive tools such as artificial neural network (ANN) can be practical for the prediction of physical properties of DESs. The application of ANNs in predicting physical and chemical properties of materials is growing quickly. Several studies have been accomplished to predict physical properties of particular DESs and ILs using ANNs [22,31–35]. As the literature does not show any research article on the application ANN for the prediction of electrical conductivity of DESs, it is proposed here that such ANN is built for this purpose. In this work, we report experimental data of electrical conductivities of two classes of DESs based on ammonium and phosphonium salts at different conditions of composition and temperature. In addition, the applicability of artificial neural network model to predict the electrical conductivities of the above mentioned DESs at different temperatures is examined.

2. Methodology 2.1. Synthesis of DESs In this work, choline chloride (C5 H14 ClNO), N,N-diethyl ethanol ammonium chloride (C6 H16 ClNO) and methyl triphenyl phosphonium bromide (C19 H18 PBr) as salts and glycerol (C3 H8 O3 ) and ethylene glycol (C2 H6 O2 ) as hydrogen bond donors were selected for the synthesis of DESs. All the above-mentioned chemicals were supplied by Merck (Darmstadt, Germany) and were of high purity (>98%). The chemicals specifications are described in Table 1. The eutectic mixtures were formed by mixing the salt and HBD together at a specific temperature and atmospheric pressure in a jacketed vessel until a homogenous and colorless liquid formed. All the experimental work of this study was conducted inside a glove box whereby the humidity was less than 0.4 ppm. The synthesized DESs were placed in tight and humidity-safe screw-capped bottles and

31

Table 1 Material description. Chemical names

Source

Mass fraction purity

Choline chloride Methyl triphenyl phosphonium bromide N,N-diethyl ethanol ammonium chloride Glycerol Ethylene glycol

Merck Merck Merck Merck Merck

≥0.98 ≥0.98 ≥0.98 ≥0.99 ≥0.99

stored in a dehumidifier chamber to prevent any contamination with atmospheric water vapor. For each DES, the mole fractions of salt and HBD around the eutectic points were chosen. All of the DESs were easily synthesized at temperatures less than 363.15 K. The freezing point and water content of all synthesized DESs were measured using the methods described in our previous works [22]. In this work, the uncertainty in freezing point measurement was ±0.01 K. For the sake of simplicity, the DESs were given abbreviated names. The abbreviations together with the freezing temperatures are summarized in Table 2. 2.2. Electrical conductivity measurement The electrical conductivity of the synthesized DESs was measured by a multi parameter analyzer (DZS-708, Cheetah) with a resolution of 0.001 (␮S cm−1 ). The cell constant was calibrated by measuring the conductivities of aqueous solutions of KCl at different concentrations according to the IUPAC recommendation [36]. The accuracy of the cell constant was found to be 0.2%. The DESs’ electrical conductivity in this study was measured at a temperature range of 298.15-353.15 K at 5 K intervals. The variation of the temperature was achieved by using a water bath with temperature control. For each measurement three replicates were carried out and the uncertainties of the electrical conductivity and temperature values were within the range of ±0.003 mS cm−1 and ±0.1 K, respectively. 2.3. Neural network modeling An artificial neural network (ANN) consists of an interrelated set of artificial neurons, and it processes information using a connectionist approach to computation. ANNs are computing systems which can be trained to learn a complex relationship between two or more variables or datasets [37]. Amongst the available ANNs, the feed-forward neural network is one of the most important historical developments in neurocomputing [22]. In this study, a feed-forward back propagation neural network with hyperbolic tangent sigmoidal activation functions was developed for the prediction of DESs’ electrical conductivity. It consists of three layers namely, the input, hidden and the output. Normally, designing a neural network for a particular task involves the selection of the optimum network architecture and parameters. The designed network will be then trained with experimental input–output data. Subsequently, this is followed by a validation step of the network model. This is achieved by testing a dataset which was not considered in the training stage. This is needed to check the interpolative capability of the obtained optimum network. In order to train and validate the neural network built here, experimentally measured electrical conductivities of the studied DESs were used as testing data. The electrical conductivity datasets for the DESs of this study are shown in Tables 3, 4 and 5. In the first step, the input and output parameters of the model were defined. As can be seen from Fig. 1, the input parameters to the developed network comprise mole fractions of choline chloride (xC5 H14 ClNO ), N,N-diethyl ethanol ammonium chloride (xC6 H16 ClNO ) and methyl triphenyl phosphonium bromide (xC19 H18 PBr ) as sources of salt, mole fractions

32

F.S.G. Bagh et al. / Fluid Phase Equilibria 356 (2013) 30–37

Table 2 Synthesized DESs compositions, abbreviations, and freezing points. Salt

HDB

xsalt

Abbreviation

Freezing temp. (K)

Choline chloride

Ethylene glycol Ethylene glycol Ethylene glycol Glycerol Glycerol Glycerol

0.36 0.34 0.28 0.5 0.34 0.25

xHBD 0.64 0.66 0.72 0.50 0.66 0.75

DES1 DES2 DES3 DES4 DES5 DES6

239.83 207.14 277.3 281.18 237 240.5

N,N-diethyl ethanol ammonium chloride

Ethylene glycol Ethylene glycol Ethylene glycol Glycerol Glycerol Glycerol

0.28 0.25 0.2 0.34 0.25 0.2

0.72 0.75 0.8 0.66 0.75 0.8

DES7 DES8 DES9 DES10 DES11 DES12

240.03 250.9 251.48 271.82 274.85 275.12

Methyl triphenyl phosphonium bromide

Ethylene glycol Ethylene glycol Ethylene glycol Glycerol Glycerol Glycerol

0.25 0.2 0.16 0.34 0.25 0.2

0.75 0.8 0.84 0.66 0.75 0.8

DES13 DES14 DES15 DES16 DES17 DES18

226.9 223.8 224.6 276.9 267.6 288.9

Table 3 Electrical conductivity dataset for choline chloride-based DESs.a Temp. (K)

298.15 303.15 308.15 313.15 318.15 323.15 328.15 333.15 338.15 343.15 348.15 353.15 a

Conductivity (mS cm−1 ) DES1

DES2

DES3

DES4

DES5

DES6

6.801 9.138 10.857 12.935 14.794 16.335 18.393 20.102 21.773 23.474 24.75 26.043

7.332 10.191 11.407 13.553 15.895 17.185 20.227 22.991 24.2 25.599 27.065 28.072

8.317 10.665 12.07 14.133 16.558 17.977 21.257 24.247 25.152 26.275 27.799 28.69

1.929 2.191 3.161 3.68 4.603 5.864 6.668 7.805 8.98 9.863 11.548 12.954

1.749 1.951 2.549 3.004 3.991 5.12 6.046 7.187 8.16 8.955 10.629 12.191

1.463 1.553 2.035 2.57 3.112 3.816 4.811 5.757 6.717 7.805 9.286 10.8

Standard uncertainties u are u(T) = 0.1 K and the combined expanded uncertainties Uc are Uc(k) = 0.003 mS cm−1 .

of glycerol (xC3 H8 O3 ) and ethylene glycol (xC2 H6 O2 ) as hydrogen bond donors and temperature (K). The electrical conductivity of a DES was considered as the network output. In the next step, the main dataset was divided randomly into training (50%), validation (25%) and simulation (25%). The training, validation and simulation datasets were consisted of the experimental data at (298.15, 308.15, 318.15, 328.15, 338.18 and 348.15 K), (303.15, 323.15 and 343.15 K) and (313.15, 333.15 and 353.15 K), respectively. The optimum network architecture forward selection method was used for the selection of the neural network architecture. Initially, an ANN was generated with 1 neuron in hidden layer. The

network was trained and validated and its performance was tested in accordance with the experimental data. The number of hidden neurons was then increased and the process was repeated, which improved the overall results of the training and testing. For assessment of the designed networks with different number of neurons in the hidden layer, several statistical descriptors (explained comprehensively in previous work [22]) were calculated, and their values compared to assess the networks with different number of neurons in the hidden layer to know the extent to which they can be trained with superior prediction quality of the electrical conductivity of DESs.

Table 4 Electrical conductivity dataset for N,N-diethyl ethanol ammonium chloride-based DESs.a Temp. (K)

298.15 303.15 308.15 313.15 318.15 323.15 328.15 333.15 338.15 343.15 348.15 353.15 a

Conductivity (mS cm−1 ) DES7

DES8

DES9

DES10

DES11

DES12

5.12 6.627 7.94 9.283 10.955 12.539 14.33 16.161 17.779 19.417 21.347 23.097

5.429 6.878 8.305 9.486 11.339 13.147 14.769 16.664 18.395 19.9 21.924 23.667

5.661 6.994 8.245 9.699 11.579 13.408 15.086 17.137 18.755 20.286 22.262 24.053

0.75 1.177 1.635 2.067 2.716 3.381 3.903 4.878 5.754 6.521 7.754 9.109

0.602 0.958 1.041 1.562 2.112 2.637 3.426 4.086 4.916 5.748 6.474 7.1

0.487 0.78 1.099 1.387 1.878 2.357 2.716 3.246 3.962 4.646 5.335 6.095

Standard uncertainties u are u(T) = 0.1 K and the combined expanded uncertainties Uc are Uc(k) = 0.003 mS cm−1 .

F.S.G. Bagh et al. / Fluid Phase Equilibria 356 (2013) 30–37

33

Table 5 Electrical conductivity dataset for methyl triphenyl phosphonium bromides-based DESs.a Temp. (K)

Conductivity (mS cm−1 ) DES13

DES14

DES15

DES16

DES17

DES18

298.15 303.15 308.15 313.15 318.15 323.15 328.15 333.15 338.15 343.15 348.15 353.15

1.092 1.598 1.9136 2.502 2.964 3.265 4.307 5.129 5.797 6.723 7.372 8.114

1.557 2.193 2.649 3.246 3.858 4.405 5.395 6.221 7.423 8.192 9.074 10.027

1.942 2.57 3.103 3.845 4.437 5.072 6.279 7.11 8.169 9.496 10.31 11.196

0.062 0.124 0.186 0.277 0.405 0.496 0.701 0.858 1.16 1.493 1.811 2.154

0.103 0.172 0.319 0.394 0.549 0.719 0.927 1.124 1.487 1.778 2.196 2.599

0.116 0.198 0.37 0.41 0.607 0.816 0.965 1.233 1.608 1.971 2.874 3.594

a

Standard uncertainties u are u(T) = 0.1 K and the combined expanded uncertainties Uc are Uc(k) = 0.003 mS cm−1 .

3. Results and discussion 3.1. Electrical conductivity measurement In this study, eighteen DESs made from choline chloride (DESs 1–6), N,N-diethyl ethanol ammonium chloride (DESs 7–12) and methyl triphenyl phosphonium bromide (DESs 13–18) salts were synthesized at various salt:HBD mole ratios (Table 1). For the verification of the consistency of the physical behavior of DESs 1–18 with the general behavior of deep eutectic solvents, the freezing temperature and water content of all DESs were measured. The results revealed that freezing temperatures of all synthesized DESs were lower than the freezing temperatures of their original precursors. The water content of all DESs, on the other hand, was less than 0.1 wt%. Subsequently, electrical conductivities of these DESs were measured successfully over a wide temperature range (from 298.15 to 353.15 K at 5 K interval) at normal atmospheric pressure and with an uncertainty of ±0.003 mS cm−1 . Tables 3, 4 and 5 show the experimentally measured electrical conductivities of these DESs. It was found that the electrical conductivity of all synthesized DESs increased exponentially with raising temperature. This can be attributed to the decrease of viscosity of the DES exponentially with increasing the temperature. As expected the electrical conductivity of DESs increased with increasing the salt concentration in the mixture. This can be attributed to the increase of the number of charge-carriers in the solution. This behavior can be seen in Tables 3 and 4 for choline chloride:glycerol DESs (DESs 4–6) as well as N,N-diethyl ethanol

ammonium chloride:glycerol DESs (DESs 10–12). However, for other ammonium salts:ethylene glycol DESs (DESs 1–3 and DESs 7–9) and phosphonium-based DESs (DESs 13–18), at higher salt concentrations, the electrical conductivity decreases due to the strong influence of ion pairs, ion triplets, and higher ion aggregations, which reduces the overall mobility and number of the effective charge carriers. If a solvent is to be used for any particular electrochemical process its electrical conductivity should be higher than a certain limit. So if the DESs studied in this work will be used in electrochemical processes, their conductivity must be more than 10−1 mS cm−1 [38]. In this study, all measured electrical conductivities of DESs were found to be in the aforesaid range of electrical conductivity except for the DES made from methyl triphenyl phosphonium bromide and glycerol (DES16) at 298.15 K. Since the electrical conductivity is proportional to the concentration of free ions and the mobility of the ions, the molar conductivity (, S cm2 mol−1 ) of each salt in HBD can be calculated by Eq. (1) [39,40]: (T ) = k(T )

M (T )

(1)

where M, k and  are the molar mass (g mol−1 ), electrical conductivity (S cm−1 ) and density (g cm−3 ) of the DES, respectively. The molar mass of DES was calculated via multiplying of molar mass of salt and HBD by their mole fraction in DES. Using Eq. (1), the molar conductivity of each salt in HBD in this study was determined from the experimental values of the electrical conductivity and density of

Fig. 1. The schematic diagram of a 6-8-1 neural network for prediction of electrical conductivity.

34

F.S.G. Bagh et al. / Fluid Phase Equilibria 356 (2013) 30–37

Table 6 Molar mass, density,  (g cm−3 ) (upper row) [22] and molar conductivity,  (S cm2 mol−1 ) of studied DESs.a DES

M.wt. (g mol−1 )

Temp. (K) 298.15

303.15

308.15

313.15

318.15

323.15

328.15

333.15

338.15

343.15

348.15

353.15

1.1156 0.7371

1.1122 0.8784

1.1097 1.0489

1.1064 1.2033

1.1037 1.3318

1.1006 1.5038

1.0968 1.6493

1.0948 1.7896

1.091 1.9362

1.089 2.0452

1.087 2.156

DES1

89.98

1.118 0.5474

DES2

88.43

1.1174 0.5803

1.1146 0.8086

1.1117 0.9074

1.1088 1.081

1.1059 1.2711

1.1036 1.3771

1.1001 1.6261

1.0974 1.8528

1.0944 1.9556

1.0916 2.0739

1.0885 2.1989

1.0854 2.2873

DES3

83.78

1.117 0.6239

1.1141 0.802

1.1111 0.9102

1.1082 1.0685

1.1052 1.2552

1.1023 1.3664

1.0993 1.6201

1.0962 1.8532

1.0934 1.9274

1.0904 2.0189

1.0875 2.1417

1.0847 2.2161

DES4

115.86

1.1558 0.1934

1.1535 0.2201

1.1506 0.3183

1.1483 0.3713

1.1454 0.4656

1.143 0.5944

1.1402 0.6776

1.1379 0.7947

1.1349 0.9167

1.1326 1.0089

1.1296 1.1845

1.1266 1.3322

DES5

108.19

1.192 0.1589

1.1895 0.1776

1.1867 0.2325

1.1838 0.2747

1.1814 0.3657

1.1776 0.4707

1.1761 0.5565

1.1741 0.6627

1.1708 0.7545

1.1674 0.8304

1.1655 0.9873

1.1635 1.1343

DES6

103.91

1.203 0.1265

1.2002 0.1346

1.1976 0.1767

1.1951 0.2236

1.1922 0.2714

1.1891 0.3337

1.1868 0.4215

1.1844 0.5054

1.1814 0.5912

1.1781 0.6889

1.176 0.821

1.1732 0.9572

DES7

87.71

1.1006 0.408

1.0976 0.5296

1.0944 0.6364

1.0915 0.746

1.0882 0.883

1.0851 1.0135

1.082 1.1617

1.0788 1.3141

1.0757 1.4497

1.0729 1.5874

1.0695 1.7507

1.0665 1.8996

DES8

84.96

1.1018 0.4187

1.0986 0.5319

1.0955 0.6441

1.0926 0.7377

1.0893 0.8845

1.0862 1.0284

1.083 1.1586

1.0797 1.3113

1.0767 1.4516

1.074 1.5743

1.0705 1.7401

1.0675 1.8837

DES9

80.38

1.1037 0.4123

1.1004 0.5109

1.0973 0.604

1.0942 0.7125

1.091 0.8532

1.0877 0.9909

1.0846 1.1181

1.0815 1.2738

1.0782 1.3983

1.075 1.5169

1.0718 1.6697

1.0683 1.8099

DES10

113.02

1.1731 0.0722

1.17 0.1137

1.1671 0.1583

1.1638 0.2008

1.1613 0.2643

1.1585 0.3299

1.1554 0.3819

1.1531 0.4782

1.1492 0.5659

1.1458 0.6432

1.1431 0.7667

1.1403 0.9029

DES11

107.48

1.2051 0.0537

1.202 0.0857

1.199 0.0933

1.1961 0.1404

1.1929 0.1903

1.1884 0.2385

1.1868 0.3103

1.184 0.371

1.1807 0.4476

1.1777 0.5246

1.1746 0.5925

1.1719 0.6512

DES12

104.41

1.2201 0.0417

1.2173 0.0669

1.2141 0.0945

1.2114 0.1196

1.2081 0.1623

1.2048 0.2043

1.2021 0.2359

1.1989 0.2827

1.1961 0.3458

1.1936 0.4064

1.1901 0.468

1.1873 0.536

DES13

135.86

1.2501 0.1186

1.2424 0.1747

1.2432 0.2091

1.2398 0.2742

1.2363 0.3257

1.2324 0.3599

1.2293 0.476

1.226 0.5684

1.2223 0.6444

1.2188 0.7495

1.2153 0.8242

1.2119 0.9097

DES14

121.1

1.2326 0.153

1.229 0.2161

1.2256 0.2617

1.222 0.3217

1.2185 0.3834

1.2151 0.439

1.2114 0.5394

1.2079 0.6237

1.2043 0.7465

1.2006 0.8263

1.1972 0.9179

1.1936 1.0173

DES15

109.29

1.2195 0.174

1.2163 0.2309

1.2124 0.2797

1.2086 0.3477

1.2052 0.4024

1.202 0.4611

1.198 0.5729

1.1947 0.6504

1.1907 0.7499

1.187 0.8743

1.1834 0.9522

1.1792 1.0377

DES16

182.24

1.3064 0.0087

1.303 0.0174

1.299 0.0262

1.2952 0.039

1.292 0.0571

1.2886 0.0701

1.2851 0.0993

1.2817 0.122

1.2782 0.1654

1.2749 0.2135

1.2715 0.2595

1.2682 0.3096

DES17

158.38

1.2976 0.0126

1.2945 0.0211

1.2906 0.0391

1.287 0.0485

1.2836 0.0677

1.2809 0.0889

1.2766 0.1149

1.2736 0.1398

1.2696 0.1855

1.266 0.2225

1.2626 0.2755

1.2593 0.3268

DES18

145.12

1.2889 0.0131

1.2855 0.0224

1.2817 0.0418

1.278 0.0465

1.2745 0.0691

1.271 0.0932

1.2673 0.1106

1.2634 0.1416

1.2601 0.1852

1.2568 0.2276

1.2529 0.3329

1.2495 0.4174

a

Standard uncertainties u are u(T) = 0.1 K and the combined expanded uncertainties Uc are Uc() = 0.0003 g cm−3 and Uc() = 0.003 mS cm2 mol−1 .

each DES. The required densities of DESs used to calculate the molar conductivities were taken from our previous work [22]. Table 6 shows the molar mass and densities of all synthesized DESs as well as calculated molar conductivities of all salts at different temperatures. It can be seen from Table 6 that the molar conductivity of each salt in HBD increases with increasing temperature. 3.2. Neural network modeling A feed-forward back propagation neural network, with three layers (input, hidden and output), was used in this study to estimate the DESs’ electrical conductivity. The methodology of this approach was carried out using the Matlab neural networks toolbox implementation. All of the 216 electrical conductivity data points were collected from experimental data of DESs’ electrical conductivity. In this regards, 50% of the collected data were used to train, 25% for validation and the rest were used for testing the accuracy of the obtained optimum architecture after training. Different numbers of hidden neurons were tested. Table 7 gives the results of statistical assessments to find the optimal ANN. The bold numbers in this table refer to the best number of neurons in hidden layer. These results

show that among the different networks tested, the networks with 8 hidden neurons had the best prediction performance and gave the smallest overall value of Normalized Mean Square Error (NMSE) (0.0010) and acceptable values of Index of Agreement (IA) (0.9999) and Regression Coefficient (R2 ) (0.9988). Furthermore, the results of statistical assessment indicated that the selected inputs appeared to have a pronounced contribution to predict the DESs’ electrical conductivity in the network with 8 hidden neurons. Fig. 1 shows a schematic layout of a 6-8-1 neural network considered for the prediction of electrical conductivity. The results obtained in the training and validation stages using 8 hidden neurons are given in Figs. 2 and 3, respectively. As shown in these figures, the trained data and validated data are in a very good agreement with the experimental data of the DESs’ electrical conductivities. These results indicate that the network has been well-trained and can be used to simulate the electrical conductivities of the DESs of this study within a wide range of input conditions. The simulation results are shown in Fig. 4, whereby a good agreement was observed between the experimental data and the ANN predicted electrical conductivities data. In addition to this, an absolute average relative deviation (AARD%) value of 4.40% and

F.S.G. Bagh et al. / Fluid Phase Equilibria 356 (2013) 30–37

35

Table 7 Statistical assessment for the potential of hidden layer neurons. Hidden neurons number

1 2 3 4 5 6 7 8 9 10 12 14 16 18 20 30 a

R2

IA

FB

NMSE

MG

VG

(1.000)a

(1.000)a

(0.000)a

(0.000)a

(1.000)a

(1.000)a

0.9837 0.9843 0.9777 0.9926 0.9906 0.9912 0.9963 0.9988 0.9958 0.9956 0.9916 0.9928 0.9962 0.9941 0.9952 0.9916

0.9991 0.9991 0.9988 0.9996 0.9995 0.9986 0.9998 0.9999 0.9998 0.9998 0.9995 0.9996 0.9998 0.9994 0.9996 0.9995

0.0096 0.0052 0.0292 0.0108 −0.004 −0.0021 0.0103 0.0106 −0.0001 0.0066 0.0193 0.0148 −0.0017 0.0506 0.0135 0.0112

0.0074 0.007 0.0101 0.0031 0.0039 0.0038 0.0017 0.001 0.0018 0.002 0.0038 0.0034 0.0016 0.0047 0.0031 0.0037

1.0544 1.0299 1.0589 0.992 1.0046 1.0869 1.0215 0.9954 1.0194 0.9897 1.0484 1.0444 1.0061 1.0936 0.9722 0.9724

1.06 1.0645 1.164 1.0128 1.0152 1.4442 1.0604 1.0046 1.0186 1.0122 1.0274 1.0227 1.035 1.0304 1.0108 1.0167

Figures in brackets are ideal values.

-1

Predicted conductivity (m S.cm )

30

-1

Trained conductivity (m S.cm )

30

25

Y= 0.9990x+0.0045 2

R = 0.9995 20

15

25

Y= 0.9734x+0.1537 2

R = 0.9983

20

15

10

5

10 0

5

0

5

10

15

20

25

30

Experimental conductivity (m S.cm -1) 0

0

5

10

15

20

25

30

-1

Fig. 4. Simulated and measured electrical conductivities using 8 neurons in the hidden layer.

Experimental conductivity (m S.cm ) Fig. 2. Training results for the optimum neural network with 8 hidden neurons. 30 28 30

-1

Predicted conductivity (m S.cm )

-1

Conductivity (m S.cm )

26

Y= 0.9983x+0.0897

25

2

R = 0.9988 20

15

24 22 20 18 16 14

10

12 10 309 312 315 318 321 324 327 330 333 336 339 342 345 348 351 354 357

5

0

Temperature (K) 0

5

10

15

20

25

30

Experimental conductivity (m S.cm -1) Fig. 3. Validation results for the optimum neural network with 8 hidden neurons.

Fig. 5. Experimental and predicted electrical conductivities of choline chlorefer to experimental and predicted data of ride:ethylene glycol DESs. (䊉 and DES1,  and – – – refer to experimental and predicted data of DES2, and  and –· ·– refer to experimental and predicted data of DES3).

36

F.S.G. Bagh et al. / Fluid Phase Equilibria 356 (2013) 30–37 12

14 10 -1

Conductivity ( m S.cm )

-1

Conductivity (m S.cm )

12 10 8 6 4

8

6

4

2

2

0 309 312 315 318 321 324 327 330 333 336 339 342 345 348 351 354 357

0 309 312 315 318 321 324 327 330 333 336 339 342 345 348 351 354 357

Temperature (K) Fig. 6. Experimental and predicted electrical conductivities of choline chlorefer to experimental and predicted data of ride:ethylene glycol DESs. (䊉 and DES4,  and – – – refer to experimental and predicted data of DES5, and  and –· ·– refer to experimental and predicted data of DES6).

Temperature (K) Fig. 9. Experimental and predicted electrical conductivities of choline chlorefer to experimental and predicted data of ride:ethylene glycol DESs. (䊉 and DES13,  and – – – refer to experimental and predicted data of DES14, and  and –· ·– refer to experimental and predicted data of DES15).

24

-1

Conductivity (m S.cm )

22 20 18 16 14 12 10 8 309 312 315 318 321 324 327 330 333 336 339 342 345 348 351 354 357

Temperature (K)

a regression coefficient (R2 ) value of 0.9983 both confirm a good ability of the proposed network to predict the electrical conductivities of the DESs of this study. Because of the adequate amount of training data, the network is believed to be well-trained and the predictions to be of high quality. The predicted results from the proposed neural network were also compared graphically to the DESs’ electrical conductivities experimental data. Figs. 5–10 show comparisons between the experimental and predicted electrical conductivities of (DESs 1–3), (DESs 4–6), (DESs 7–9), (DESs 10–12), (DESs 13–15) and (DESs 16–18), respectively. The figures indicate acceptable predicted trends of the DESs electrical conductivities using the proposed ANN model. In addition, in terms of the effects of temperature and composition on the electrical conductivities predicted by the proposed ANN, all predicted values had identical trends with the experimental data as described earlier.

Fig. 7. Experimental and predicted electrical conductivities of choline chloride:ethylene glycol DESs. (䊉 and refer to experimental and predicted data of DES7,  and – – – refer to experimental and predicted data of DES8, and  and –· ·– refer to experimental and predicted data of DES9).

4.0

10

8

-1

Conductivity (m S.cm )

-1

Conductivity (m S.cm )

3.5

6

4

3.0 2.5 2.0 1.5 1.0

2 0.5

0 309 312 315 318 321 324 327 330 333 336 339 342 345 348 351 354 357

Temperature (K) Fig. 8. Experimental and predicted electrical conductivities of choline chlorefer to experimental and predicted data of ride:ethylene glycol DESs. (䊉 and DES10,  and – – – refer to experimental and predicted data of DES11, and  and –· ·– refer to experimental and predicted data of DES12).

0.0 309 312 315 318 321 324 327 330 333 336 339 342 345 348 351 354 357

Temperature (K) Fig. 10. Experimental and predicted electrical conductivities of choline chlorefer to experimental and predicted data of ride:ethylene glycol DESs. (䊉 and DES16,  and – – – refer to experimental and predicted data of DES17, and  and –· ·– refer to experimental and predicted data of DES18).

F.S.G. Bagh et al. / Fluid Phase Equilibria 356 (2013) 30–37

4. Conclusion In this study, we reported the electrical conductivities of eighteen DESs made from choline chloride, N,N-diethyl ethanol ammonium chloride and methyl triphenyl phosphonium bromide salts over a wide range of temperature (298.15–353.15 K at 5 K interval) at atmospheric pressure. The results revealed that the electrical conductivities of all synthesized DESs demonstrate a strong temperature-dependent behavior. The electrical conductivities of all DESs increased exponentially with the increase in temperature. In addition, the molar conductivities of ammonium and phosphonium salts in DESs were calculated from the experimental values of electrical conductivities and densities of the DESs. The temperature-dependent behavior of molar conductivity of salts was also observed. The applicability of an ANN model to predict the electrical conductivities of ammonium and phosphonium based DESs at different temperatures and compositions was investigated. In this regards, experimental data of electrical conductivity of the DESs were employed to verify the proposed neural network. The experimental data were divided into training, validation and simulation datasets. The best neural network was selected based on statistical assessments. These results revealed that among the different networks structures tested, the network with 8 hidden neurons had the best prediction performance and gave the smallest overall value of NMSE (0.0010) and acceptable values of IA (0.9999) and R2 (0.9988). The comparison of the predicted electrical conductivities of DESs with those obtained experimentally confirmed that the proposed ANN model was able to well-predict the DESs’ electrical conductivity with an average absolute relative deviation of 4.40%. Acknowledgments This research was funded by University of Malaya research grant number HIR-MOHE D000003-16001 and by the Deanship of Scientific Research at King Saud University through group project no. RGP-VPP-108 in collaboration with the petroleum and chemical engineering department, engineering faculty in Sultan Qaboos University, Oman. References [1] J.G. Huddleston, A.E. Visser, W.M. Reichert, H.D. Willauer, G.A. Broker, R.D. Rogers, Green Chem. 3 (2001) 156–164. [2] P. Bonhte, A.P. Dias, N. Papageorgiou, K. Kalyanasundaram, M. Grtzel, Inorg. Chem. 35 (1996) 1168–1178. [3] J.D. Holbrey, K.R. Seddon, J. Chem. Soc., Dalton Trans. (1999) 2133–2140. [4] B.D. Fitchett, T.N. Knepp, J.C. Conboy, J. Electrochem. Soc. 151 (2004) E219–E225. [5] H. Tokuda, K. Hayamizu, K. Ishii, M.A.B.H. Susan, M. Watanabe, J. Phys. Chem. B 108 (2004) 16593–16600.

37

[6] D.J. Couling, R.J. Bernot, K.M. Docherty, J.K. Dixon, E.J. Maginn, Green Chem. 8 (2006) 82–90. [7] C. Pretti, C. Chiappe, I. Baldetti, S. Brunini, G. Monni, L. Intorre, Ecotoxicol. Environ. Saf. 72 (2009) 1170–1176. [8] Q. Zhang, K. De Oliveira Vigier, S. Royer, F. Jerome, Chem. Soc. Rev. 41 (2012) 7108–7146. [9] A.P. Abbott, G. Capper, D.L. Davies, R.K. Rasheed, V. Tambyrajah, Chem. Commun. (2003) 70–71. [10] E. Durand, J. Lecomte, B. Bara, G. Piombo, E. Dubreucq, P. Villeneuve, Process. Biochem. 47 (2012) 2081–2089. [11] H.G. Morrison, C.C. Sun, S. Neervannan, Int. J. Pharm. 378 (2009) 136–139. [12] H.R. Jhong, D.S.H. Wong, C.C. Wan, Y.Y. Wang, T.C. Wei, Electrochem. Commun. 11 (2009) 209–211. [13] A.P. Abbott, K. El Ttaib, G. Frisch, K.J. McKenzie, K.S. Ryder, Phys. Chem. Chem. Phys. 11 (2009) 4269–4277. [14] M.A. Kareem, F.S. Mjalli, M.A. Hashim, I.M. Alnashef, J. Chem. Eng. Data 55 (2010) 4632–4637. [15] M.A. Kareem, F.S. Mjalli, M.A. Hashim, M.K.O. Hadj-Kali, F.S.G. Bagh, I.M. Alnashef, Fluid Phase Equilib. 333 (2012) 47–54. [16] K. Shahbaz, F.S. Mjalli, M.A. Hashim, I.M. AlNashef, Energy Fuels 25 (2011) 2671–2678. [17] K. Shahbaz, F.S. Mjalli, M.A. Hashim, I.M. AlNashef, Sep. Purif. Technol. 81 (2011) 216–222. [18] R.B. Leron, A.N. Soriano, M.-H. Li, J. Taiwan Inst. Chem. Eng. 43 (2012) 551–557. [19] R.B. Leron, M.-H. Li, Thermochim. Acta 530 (2012) 52–57. [20] R.B. Leron, M.-H. Li, J. Chem. Thermodyn. 54 (2012) 293–301. [21] R.B. Leron, D.S.H. Wong, M.-H. Li, Fluid Phase Equilib. 335 (2012) 32–38. [22] K. Shahbaz, S. Baroutian, F.S. Mjalli, M.A. Hashim, I.M. Alnashef, Thermochim. Acta 527 (2012) 59–66. [23] K. Shahbaz, F.S. Mjalli, M.A. Hashim, I.M. AlNashef, Thermochim. Acta 515 (2011) 67–72. [24] K. Shahbaz, F.S. Mjalli, M.A. Hashim, I.M. AlNashef, Fluid Phase Equilib. 319 (2012) 48–54. [25] S.-H. Wu, A.R. Caparanga, R.B. Leron, M.-H. Li, Thermochim. Acta 544 (2012) 1–5. [26] Y. Ju, K. Inoue, M. Saka, H. Ab, Appl. Phys. Lett. 81 (2002) 3585–3587. [27] J. Vila, C. Franjo, J.M. Pico, L.M. Varela, O. Cabeza, Port. Electrochim. Acta 25 (2007) 163. [28] A. Anderko, M.M. Lencka, Ind. Eng. Chem. Res. 36 (1997) 1932–1943. [29] P.N. Tshibangu, S.N. Ndwandwe, E.D. Dikio, J. Electrochem. Sci. 6 (2011) 2201–2213. [30] J. Leys, M. Wbbenhorst, P. Menon Chirukandath, R. Rajesh, J. Thoen, C. Glorieux, P. Nockemann, B. Thijs, K. Binnemans, S. Longuemart, J. Chem. Phys. 128 (2008) 1–7. [31] J.O. Valderrama, A. Reategui, R.E. Rojas, Ind. Eng. Chem. Res. 48 (2009) 3254–3259. [32] M. Lashkarbolooki, A.Z. Hezave, S. Ayatollahi, Fluid Phase Equilib. 324 (2012) 102–107. [33] A.Z. Hezave, M. Lashkarbolooki, S. Raeissi, Fluid Phase Equilib. 314 (2012) 128–133. [34] M. Lashkarblooki, A.Z. Hezave, A.M. Al-Ajmi, S. Ayatollahi, Fluid Phase Equilib. 326 (2012) 15–20. [35] J.S. Torrecilla, F. Rodriguez, J.L. Bravo, G. Rothenberg, K.R. Seddon, I. LopezMartin, Phys. Chem. Chem. Phys. 10 (2008) 5826–5831. [36] E.J.a.K.N. Marsh, Pure Appl. Chem. 53 (1981) 1841–1845. [37] C. Lau, Neural Networks: Theoretical Foundation and Analysis, IEEE Press, Piscataway, NJ, USA, 1992. [38] M.R.G. Farnoush Faridbod, P. Norouzi, S. Riahi, H. Rashedi, Application of room temperature ionic liquids, in: A. Kokorin (Ed.), Electrochemical Sensors and Biosensors, Ionic Liquids: Applications and Perspectives, 2011. [39] L.H. Tzi-Yi Wu, C.-W. Kuo, Y.-C. Lin, S.-G. Su, P.-L. Kuo, I.-W. Sun, Int. J. Electrochem. Sci. 7 (2012) 2047–2064. [40] A.P. Abbott, R.C. Harris, K.S. Ryder, J. Phys. Chem. B 111 (2007) 4910–4913.