University of Groningen
Computer-aided Ionic Liquids Design for Separation Processes Peng, Daili
DOI:
10.33612/diss.168550903
IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from it. Please check the document version below.
Document Version
Publisher's PDF, also known as Version of record
Publication date: 2021
Link to publication in University of Groningen/UMCG research database
Citation for published version (APA):
Peng, D. (2021). Computer-aided Ionic Liquids Design for Separation Processes. University of Groningen. https://doi.org/10.33612/diss.168550903
Copyright
Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons).
Take-down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.
Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons the number of authors shown on this cover page is limited to 10 maximum.
1
Chapter 1
Introduction
2
1. Ionic liquids
Ionic liquids (ILs) are a subset of molten salts that typically comprise organic cations and (organic or) inorganic anions (Fig. 1). The melting points of ILs are often equal or lower than 100 °C; if their melting point is below room temperature, they are called room-temperature ionic liquids (RTILs)1. The first reported IL, ethylammonium
nitrate, by Walden2 in 1914, was synthesized by neutralizing ethylamine with
concentrated nitric acid. Ethylammonium nitrate is a clear, colorless, odorless liquid, and as opposed to water, it has a much higher viscosity. More importantly, the electric conductivity of ethylammonium nitrate is in line with a composition of pure anions and cations2-4. In 1951, Hurley and Wier5 synthesized another kind of IL by mixing
alkylpyridinium chlorides with AlCl3. However, those ILs are not stable in a humid
environment, and it is also difficult to regulate their acidity/basicity6. In 1992, Wilkes
and Zaworotko7 reported an air- and moisture-resistant IL, which resulted in a dramatic
increase in research related to ILs as shown, among others (Fig. 2), by the number of publications in the last two decades. According to Web of Science, almost 9000 papers related to ILs are published in 2019, which is 14 times the number of publications in 2000.
3
Fig. 1. Structures of typical cations and anions.
4
Because of their unique physicochemical properties, such as negligible vapor pressure, excellent chemical, thermal, and electrochemical stability, broad liquid-phase range, solvent polarity, low flammability, and wide electrochemical window, ILs have been applied in diverse research fields (Fig. 3). These include electrified interfaces8-12,
polymer additives13-18, biomedicine19-23, and reaction catalysts or solvents24-29.
Especially in the fields of extraction30-33 and gas absorption34-38, ILs have received
significant attention in recent years. Compared to conventional solvents, as non-volatile solvents, ILs are less likely to evaporate to the environment, which can prevent pollution and solvent loss. Another feature of ILs, which makes them particularly promising in separation processes, is their capability of dissolving different compounds selectively39. Moreover, because of the low vapor pressure and high thermal stability
of ILs, the solvent regeneration step in the separation process requires less energy40-43.
Thus, ILs are widely regarded as promising alternatives to volatile organic compounds (VOCs) in different separation processes, such as extraction of azeotropes33,44-48, CO
2
capture49-53, water purification32,54-57, and desulfurization30,31,58.
5
ILs are also referred to as designer solvents and task-specific solvents because of their diversity in possible structures. The physicochemical properties of ILs can be fine-tuned by judicious selection of cation and anion “cores” or by modifying the length and the type of functional groups on both charged moieties. Theoretically, there are around 1018 ILs that can be synthesized from the different ion combinations59. The huge
number of ILs makes it is very challenging to find a proper IL for a specific separation problem60. Traditionally, the screening of an IL solvent is guided by trial-and-error
approaches; however, the time and labor requirements for searching through the large design space to find the suitable ILs are prohibitive. Moreover, such screened ILs are not optimal as the experimental study tends to focus on a small class of ILs, thus neglecting other possible combinations of cations and anions. Therefore, in order to keep pace with the growing demand for new ILs and explore the full chemical design space more adequately, it is necessary to develop an efficient method for IL solvents design.
2. Computer-aided ionic liquids design
Although the general use of computers in chemistry surfaced a few decades ago, computer-aided molecular design (CAMD) is quite recent as it has been developed in the 1980s. It basically deals with the problem of designing an optimal molecular structure for a certain application. CAMD combines molecular modeling techniques, thermodynamics, and numerical optimization to design the needed molecular structures, many of them often completely novel. Due to the rapid growth in computer science in the last few decades, the hardware and algorithms for calculation have been significantly improved in the same period. The modeling methods involving large-scale and complex calculations have become readily accessible for researchers, basically at the level of desktop PCs. As a consequence, CAMD has become very popular in developing new materials. When using CAMD to design ILs, a new name has been proposed: CAILD (computer-aided ionic liquids design).
6
As can be seen from Fig. 4, in general, CAILD methods comprise three parts: objective function, constraints, and solver. The objective function represents the goal of the CAILD, which is varied along with the requirement of different applications. For a separation problem, the objective function is always related to the thermodynamic properties, e.g., the activity coefficient (𝛾). The constraints confine the structure and physical properties of the designed ILs. The structural constraint can ensure the CAILD problem to produce structures that do not violate any inherent laws of chemical bonding, e.g. octet rule. The fundamental physical properties, especially the melting point and the viscosity, are important criteria for selecting suitable solvents for separation processes, and thus should also be taken into consideration during CAILD. Of particular note, the computational approaches used in the field of CAILD leverages the simplicity of the semi-empirical quantitative structure-property relationship (QSPR) in combination with fast and efficient numerical optimization algorithms. The various types of QSPR can quickly and often accurately estimate physical properties from the structure of ILs. The application of QSPR techniques in the prediction of properties from structures is known as the “forward problem”. The CAILD can broadly be considered as the “reverse problem”, i.e. predicting structures from properties, and can be described as follows:
max. 𝑜𝑟 min. 𝑂𝐹 = 𝑓(𝑣) (1)
s.t.
ℎ(𝑣) ≤ 0 (2)
𝑔(𝑣) ≤ 0 (3)
where variable 𝑣 is a vector containing the frequency of the candidate groups; 𝑓(𝑣) is the objective function (𝑂𝐹 ); s.t. means “subject to”; ℎ(𝑣) and 𝑔(𝑣) represent the structural and physical property constraints, respectively, in the optimization problem, the right side of the constraint is normally kept as 0.
7 CAILD Objective function Constraints Solver Thermodynamic properties Structural, physical and chemical properties Generate-and-test, decomposition, etc.
Fig. 4. Structure of the CAILD problem.
In the above formulations, the frequency of the candidate groups 𝑣 is an integer variable, while the intermediate variables for calculating the objective function are always continuous. Additionally, the thermodynamic models and the QSPR models introduce many non-linearities and non-convexities to the calculation. These features make CAILD a challenging mixed-integer nonlinear programming (MINLP) problem in many cases. To solve this problem, different solvers or strategies can be used, e.g., generate-and-test, decomposition, mathematical optimization methods, etc. In summary, when using CAILD to design IL solvents for a separation process, the thermodynamic models for the objective function, the QSPR models for physical properties, and the solvers for the solution should be taken into consideration.
Table 1 summarizes the CAILD methods that have been applied in different separation processes. As can be seen, for these CAILD methods, the universal quasi-chemical functional-group activity coefficients (UNIFAC) and conductor-like screening model (COSMO) based thermodynamic models are the most popular. Moreover, the evaluation of the melting point and viscosity of ILs are considered in most of the CAILD methods since they are very important physical properties for the separation processes. The generate-and-test method is employed by many researchers to solve the CAILD problem due to its simplicity and reliability. It should be noted that because the CAILD methods are generally based on theoretical and semi-empirical
8
models, an experimental validation is often suggested after acquiring the IL candidates. However, as seen in Table 1, only a few CAILD works reported the experimental validation.
Table 1. Comparison of the CAILD works in the literature.*
Separation problems Thermodynamic models
Physical
properties Solvers Exp. Ref. Denitrification COSMO-RS NO Generate-and-test NO 61 Ethanol/water UNIFAC-IL Tm and Tb SBB and BARON NO 62
Toluene/heptane UNIFAC-IL Tm and η Decomposition approach NO 63
Pharmaceutical
intermediate UNIFAC-IL Td and Tm Generate-and-test NO 64 CO2 capture COSMO-SAC NO Generate-and-test NO 65
Benzene/cyclohexane and CO2 capture
COSMO-SAC Tm and η Branch and bound NO 66
Desulfurization UNIFAC-IL Tm and η Decomposition approach NO 67
CO2 capture COSMO-SAC Tm and η SA and GA NO 68
Ethanol/water and
acetone/methanol UNIFAC-IL Tm and η LINDOGLOBAL NO 47 Benzene/cyclohexane COSMO-RS NO Generate-and-test Yes 43 Phenol/water COSMO-SAC NO Generate-and-test Yes 69 Desulfurization Machine learning NO Generate-and-test Yes 39
* Tm means the melting point, Tb denotes the boiling point, Td represents the
decomposition temperature, and η means the viscosity.
3. Thermodynamic models
3.1. UNIFAC
The UNIFAC model was firstly proposed by Fredenslund et al.70 in 1975. It
combines the functional group concept for activity coefficients and quasi-chemical theory of liquid mixtures (UNIQUA). This model can be used to calculate the activity coefficient of compounds at any concentration in a mixture. The model has a combinatorial contribution (𝛾 ) and a residual contribution (𝛾 ):
ln 𝛾 = ln 𝛾 + ln 𝛾 (4)
The combinatorial contribution represents the differences in the size and shape of the molecules:
9 ln 𝛾 = 1 − 𝑉 + ln 𝑉 − 5𝑞 1 − + ln (5) 𝐹 = ∑ (6) 𝑉 = ∑ (7)
where 𝐹 and 𝑉 are surface area ratio and volume ratio, respectively; 𝑟 and 𝑞 are molecular van der Waals volumes and molecular surface areas, respectively, and they can be calculated as the sum of the group volume and group area parameters:
𝑟 = ∑ 𝑣( )𝑅 (8)
𝑞 = ∑ 𝑣( )𝑄 (9)
where 𝑣( ) is the number of groups of type 𝑘 in molecule 𝑖. The group parameters 𝑅 and 𝑄 are calculated by 𝑉 (van der Waals group volumes) and 𝐴 (surface areas), respectively:
𝑅 =
. (10)
𝑄 =
. × (11)
The residual part is due to energetic interactions and can be expressed as:
ln 𝛾 = ∑ 𝑣( ) ln Γ − ln Γ( ) (12)
where Γ is the group residual activity coefficient; Γ( ) is the residual activity coefficient of group 𝑘 in the reference solution containing pure compound 𝑖.
ln Г = 𝑄 [1 − ln(∑ 𝜃 𝜓 ) − ∑ 𝜃 𝜓 ⁄∑ 𝜃 𝜓 ] (13)
𝜃 = ∑ (14)
𝑋 = ∑
( )
10
𝑋 is the mole fraction of group 𝑚; 𝜓 is the group interaction parameter given by:
𝜓 = exp −(𝑎 ⁄ )𝑇 (16)
where the parameter 𝑎 characterizes the interaction between main groups 𝑛 and 𝑚. For the group-group interaction of main groups 𝑛 and 𝑚 , two different group interaction parameters 𝑎 and 𝑎 are required.
Although the UNIFAC model is very popular among chemists and chemical engineers, it was not applied to IL-related systems until recent years. Since the UNIFAC model is a group contribution (GC) based method, the ILs need to be divided into functional groups. So far, three methods for this purpose, as shown in Fig. 5, have been proposed. According to method (a), the IL is ideally divided into one cation and one anion group; for method (b), the skeletons of the cation and anion are treated as an electrically neutral group to avoid additional terms accounting for contributions caused by the strong electrostatic interaction between ionic pair71; in method (c), the IL
molecule is divided into several groups with the cation skeleton treated as a separate functional group.
Fig. 5. Methods for the segmentation of ILs.
Because method (a) cannot reflect the influence of the structural diversity of substituents on the thermodynamic properties, the other two segmentation methods are
11
mostly chosen by researchers. Lei et al.71 firstly developed a UNIFAC-IL model
(UNIFAC-Lei) in 2009 based on the segmentation method (b), and after that, the model has been revised and extended many times because more experimental data have become available72-74. The current group interaction parameters matrix of the
UNIFAC-Lei model is presented in Fig. 6. As can be seen, 50 kinds of IL main groups are included in the model. For most cases, the average relative deviation (ARD, Eq. 17) for the thermodynamic properties is less than 10%. The low ARD indicates that the UNIFAC-Lei model can provide good quantitative predictions for the thermodynamic properties of IL-involved systems.
ARD = ∑ (17)
where N is the number of data points; 𝑥 and 𝑥 are experimental and calculated values, respectively.
12
Fig. 6. Current UNIFAC-Lei parameter matrix74 (Copyright 2020, American Chemical
13
The segmentation method (c) treats cations, anions, and the substituents on cations as separate groups. Therefore, the UNIFAC model based on this method allows a larger design space, which is very beneficial to the CAILD problem. Moreover, most available GC-based methods for predicting physical properties of ILs also divided ILs in this manner75. Thus, these methods can be directly used when formulating the CAILD
problem. Song et. al75,76 developed a UNIFAC-IL model based on segmentation method
(c), the ARD for the 3654 experimental data of infinite dilution activity coefficient (𝛾 ) is 20.4%. Moreover, the obtained UNIFAC-IL model is successfully used in the CAILD problem to design ILs for the extractive desulfurization (EDS) of fuel oils.
However, due to the huge number of ILs and the scarcity of experimental data, there are still a lot of interaction parameters that are missing for the UNIFAC-IL model as shown in Fig. 6. In order to fill the gaps in the UNIFAC-IL model, many researchers chose to use the pseudoexperimental data produced by the prior models, such as conductor-like screening model for real solvent (COSMO-RS) and conductor-like screening model for segment activity coefficient (COSMO-SAC), to regress the vacant parameter pairs for UNIFAC-IL model. Based on the UNIFAC-Lei model, Dong et al.77
developed a COSMO-UNIFAC model to predict the thermodynamic of IL-involved systems. This model combines the respective advantages of COSMO-SAC and UNIFAC models and can provide a moderate quantitative prediction of the IL-involved systems when the UNIFAC model parameters are yet to be determined. Song et al.78
extending the UNIFAC model for IL-involved systems by combining experimental and computational databases using COSMO-RS. As shown in Fig. 7, over 96.0% of the calculated 𝛾 are located within the region with ±0.5 absolute error of log 𝛾 .
14
Fig. 7. Comparison of experimental and calculated 𝛾 from the UNIFAC-IL mode proposed by Song et al.78 (Copyright 2019, American Institute of Chemical Engineers).
The combination of UNIFAC and COSMO-based models can overcome the strong dependence on experimental data for the UNIFAC model. However, the pseudoexperimental data need first to be produced by other thermodynamic models and then used to regress the group interaction parameters for the UNIFAC model. This makes the development procedure of this method very complex and indirect. Moreover, the resolution of the COSMO-UNIFAC model is not as good as that of the UNIFAC model.
3.2. COSMO-SAC
The COSMO-SAC model79-81 is derived from the COSMO-RS model which is
originally proposed by Klamt et al.82. Because COSMO-SAC is a predictive model
based on quantum mechanics calculation, it is independent of experiments and has been proven to be a reliable tool for predicting thermodynamic properties83-86. It is worth
15
the code for the COSMO-SAC model became open-source and freely available to academic and non-commercial users87.
In the COSMO-SAC model, the interactions of molecules in a mixture are regarded as pairwise interactions of charged surface segments of the molecule that can be obtained from quantum mechanical calculations when the molecule is placed in a perfect conductor. Therefore, this model is generally made up of two steps: (1) calculate the surface charge density profiles (σ-profiles, Fig. 8) and cavity volumes (𝑉 ) for each species in a mixture based on the results of COSMO calculation by quantum chemistry package (e.g. DMol3); (2) calculate the activity coefficient using COMSO-SAC model based on the acquired σ-profiles and VCOSMO.
Fig. 8. The σ-profiles of several organic compounds and ILs. The σ-profile of molecular i is calculated by:
𝑃 (𝜎 ) = ( ) (18)
where 𝐴 (𝜎 ) is the surface area with screening charge density 𝜎 of molecule i and 𝐴 is the entire surface area of the molecule i.
16
The σ-profile of the mixture s can then be obtained by:
𝑃 (𝜎 ) = ∑ ∑ ( ) (19)
where 𝑥 is the mole fraction of component i.
The activity coefficient of molecule i in the mixture can be calculated from:
ln 𝛾, = ln 𝛾, + ln 𝛾, (20)
where the combinatorial part ln 𝛾, accounts for the size and shape differences of the molecules. This quantity is usually described by the Staverman-Guggenheim combinatorial term:
ln 𝛾, = ln + 𝑞 ln + 𝑙 − ∑ 𝑥 𝑙 (21)
with 𝜃 =∑ and 𝜙 =∑ ; 𝑟 and 𝑞 are the normalized volume and surface area, respectively; 𝑧 is the coordination number and set to 10.
𝑙 = (𝑟 − 𝑞 ) − (𝑟 − 1) (22)
with 𝑟 = 𝑉 𝑟 and 𝑞 =𝐴 𝑞 ; where 𝑉 is the molecular volume of component i; 𝑟 (66.69 Å3) and 𝑞 (79.53 Å2) are the normalized parameters for volume and surface
area, respectively.
The residual part ln 𝛾, , which is also called the restoring free energy part, mainly due to electrostatic interactions between the molecules in the mixture:
ln 𝛾, = 𝑛 ∑ 𝑃 (𝜎 ) ln Γ (𝜎 ) − ln Γ (𝜎 ) (23)
where 𝑛 denotes the number of surface segments of molecule i with a standard segment surface area 𝑎 and can be calculated by:
17
where the sum on the right-hand side goes overall charge densities 𝜎 in the mixture. The quantity Δ𝑊(𝜎 , 𝜎 ) is called the exchange energy and can be calculated from:
Δ𝑊(𝜎 , 𝜎 ) = (𝜎 + 𝜎 ) + 𝑐 max[0, 𝜎 − 𝜎 ]min[0, 𝜎 − 𝜎 ] (25)
where the first term on the right-hand side is the misfit energy, accounting for the electrostatic interactions, and the second term on the right-hand side accounts for hydrogen-bonding interactions. The values of the generalized parameters are 𝛼 = 16466.72 kcal Å4 mol-1 e-2, 𝑐 = 85580 kcal Å4 mol-1 e-2, and 𝜎 = 0.0084 e Å-2.
As a powerful predictive model, the COSMO-SAC has been proved to be a promising alternative for predicting the thermodynamic properties of IL-related systems and is widely used to screen ILs for different separation tasks. Nevertheless, the COSMO-SAC model has two disadvantages, which confine its application in CAILD: (1) a pre-specified σ-profile database containing all molecular information is required for the COSMO-SAC model which is time-consuming, thus this method only suitable for screening ILs in a relatively small size of design space; (2) the results of COSMO-SAC strongly influenced by the accuracy of the quantitative prediction.
In order to solve the first problem of the COSMO-SAC model, many researchers66,88 using a GC-based COSMO-SAC method (GC-COSMO) to regress the
σ-profile for every group in ILs. By this method, the σ-profile of an IL can be simply calculated by adding up the σ-profile of every group belonging to the IL (Fig. 9). This method provides a fast and reliable way to generate the information needed in the SAC model. More importantly, the GC concept is introduced to the COSMO-SAC model, which makes it possible to be used in CAILD with other GC-based physical properties prediction models.
18
Fig. 9. Comparison of σ-profile of [apeMIM]+ from the GC-COSMO method (dark
cyan dash line) and DMol3 (magenta solid line), and the σ-profile of MIM (black short dash-dot line), CH2 (red dash-dot line) and CH2NH2 (blue dash-dot-dot line)66
(Copyright 2016, Elsevier).
The second problem can be tackled by revising the original COSMO-SAC model (COSMO-SAC-2002). Wang et al.80 present a refinement to the COSMO-SAC model
(COSMO-SAC-2010) by revising the definition of hydrogen bonding; Hsieh et al.81
adding the dispersive interactions to the COSMO-SAC model (COSMO-SAC-2013) which allows a more accurate prediction of thermodynamic properties. Although the COSMO-SAC model has been consistently refined, as a predictive model, the resolution of COSMO-SAC is still worse than that of the UNIFAC model. Fingerhut et al.89 made a comprehensive assessment of COSMO-SAC models for the predictions of
fluid phase equilibrium, the mean absolute deviation (MAD, Eq. 26) between the calculation and experimental 𝛾 of 29173 data points for different models are listed in Table 2. Moreover, Bharti et al.90 reported that the COSMO-SAC model is more
accurate in predicting the IL-free systems, indicating that this method still requires further development.
19
MAD = ∑ (ln 𝛾 ) − (ln 𝛾 ) (26)
Table 2. Comparison of different thermodynamic models for the prediction of 𝛾 .
Model MAD (%)
COSMO-SAC-2010 95
COSMO-SAC-2013 86
UNIFAC 73
3.3. Machine learning
In recent years, machine learning (ML) algorithms have been well developed and used to build the QSPR and GC models for predicting different types of properties of IL-involved systems. For the thermodynamic properties of IL-involved systems, Paduszyński91 developed three models for the prediction of 𝛾 based on three ML
algorithms, i.e., stepwise multiple linear regression (SWMLR), feed-forward artificial neural network (FFANN), and least-squares support vector machine (LSSVM). A comprehensive 𝛾 database including more than 34000 data points for 188 ILs and 128 solutes is used to establish the models. The GC method is employed to describe ILs, while the Abraham solvation parameters are adopted to quantify the influence of solute structure. The performance of the ML algorithms are compared in Fig. 10, and LSSVM shows the lowest overall ARD of 12%. This means the ML-based models own a much larger design space and comparable accuracy compared to the UNIFAC-Lei model74.
20
Fig. 10. Calculated vs. experimental 𝛾 of training set resulting from the SWMLR, FFANN, and LSSVM-based models proposed by Paduszyński91 (Copyright 2016,
American Chemical Society).
Song et al.92 proposed a GC-based method based on two different ML algorithms,
i.e., artificial neural network (ANN) and support vector machine (SVM), to directly correlate the relationship between CO2 solubility and IL structure, temperature, and
pressure. The predictive abilities of the two ML algorithms are compared in Fig. 11. The ANN-GC model is slightly better than the SVM-GC model, and their mean absolute errors (MAE) are 0.0202 and 0.0240, respectively. The results indicate that both models can give reliable predictions on the solubility of CO2 in ILs, which could
21
(a)
(b)
Fig. 11. Comparison between the experimental and predicted CO2 solubility by Song et
al.92: (a) ANN-GC and (b) SVM-GC (Copyright 2020, Elsevier).
Because the methods described above are all GC-based, theoretically, they can be directly used in CAILD. However, despite their high quantitative accuracy, the limitations of these models should not be ignored. Firstly, the model based on the ML algorithm is not derived from thermodynamic principles, it is more like a black box that can be only viewed in terms of its inputs and outputs, without any knowledge of its internal workings. On the contrary, the UNIFAC and COSMO-SAC models have a solid theoretical basis, and the procedure for calculating the results is clear. Secondly, although the ML-GC models can give a correct prediction of thermodynamic properties for IL-involved systems, they are not available in the process simulators (e.g. Aspen
22
Plus). It means that the ML-GC models can be only used in the CAILD problem which does not include process evaluation.
4. Physical properties and toxicity
4.1. Melting point
In the separation processes using ILs, the melting point (Tm) of the ILs should
always lower than the operating temperature, which can guarantee the ILs stay in the liquid phase. It is well known that the thermal behavior of ILs is relatively complex93,
especially for the melting point, which strongly varies with the combination of anion and cation. The typical Tm of ILs varies from 190 to 500 K94. In order to predict the
melting point of ILs, many QSPR and GC-based methods are developed in the literature, and the comparison of some of the previous models is shown in Table 3. These models not only provide a fast and reliable prediction for the melting point of ILs but also show the relationships between melting point and the structure of ILs. For example, Lazzs95
proposed a method (Eq. 27) based on GC:
T = 288.7 + ∑ 𝑛 ∆𝑡 + ∑ 𝑛 ∆𝑡 (27)
where 𝑛 and 𝑛 are the frequency of the groups i and j in the IL, respectively; ∆𝑡 and ∆𝑡 are the contribution of the cation and anion group to the melting point, respectively.
Table 3. Comparison of the models for the prediction of melting point.
Year Method No. of ILs Results Ref.
2008 QSPR/ANN 126 MAE = 19.37 K 96 2008 QSPR/ANN 97 ARD = 1.3% 97 2009 GC 190 ARD = 5.86% 98 2010 QSPR/GA 50 MAE = 8.77 K 99 2011 GC 136 ARD = 7.8% 100 2012 GC/GA 200 ARD = 7% 95 2012 QSPR/GFA 808 ARD = 7.3% 101 2012 GC 799 ARD = 5.82% 102
23 2016 GC/GA 63 ARD = 5% 103 2017 DFT 136 ARD = 8.5% 104 2018 QSPR/ML 243 MAE = 26.1 K 105 2018 QSPR/ML 2212 MAE = 14 K 106 2019 GC 111 ARD = 4.74% 107
By comparing the contribution of the groups the influence of the structure on the melting point of IL can be evaluated. The pyridinium main group has the lowest ∆𝑡 among all the cation main groups, it means that the pyridinium ILs possess a lower melting point compared to other cation types. The -CH2- group on cation owns a ∆𝑡
of -3.759, that is to say, increasing the length of alkyl chain in the cation will decrease the melting point of ILs, which is favorable for using the ILs as solvents in the separation processes. Moreover, the presence of -OH ( ∆𝑡 = −14.994 ) and -O- (∆𝑡 = −10.468) will also decrease the melting point of ILs. The structure of anion also has a significant impact on the melting point. ILs with [DCA]- and [AlCl
4]-own
the lowest melting point while ILs with [Br]-, [Cl]-, and [I]- have the highest melting
point. In general, the melting point of ILs with same cation are in the order: [AlCl4]- <
[DCA]- < [Tf
2N]- < [SCN]- < [CH3SO4]- < [BF4]- < [PF6]- < [Cl]-. Unlike in the cation,
the CH2 group in the anion owns a ∆𝑡 of 4.264, which means the increase of the length
of alkyl chains in the anion will increase the melting point of ILs. 4.2. Viscosity
The viscosity data of fluids is important in the chemical engineering processes for the design of liquid-liquid extractors, distillation columns, piping systems, reactors, and other units. The flow behavior of fluids significantly affected by their viscosity, thus this property should be included in the process simulation and optimization108. Low
viscosity is generally desired for a solvent since it can minimize pumping costs and increase mass transfer rates. Because ILs tend to possess a higher viscosity compared to conventional solvents, their viscosity should be considered in the CAILD method.
24
To accurately estimate the viscosity of ILs, different models are published based on QSPR and GC methods which are summarized in Table 4.
Table 4. Comparison of the models in the literature for the prediction of viscosity.
Year Method No. of ILs T (K) ARD (%) Ref.
2007 GC/MLR 77 283 - 362 18.7 109 2007 QSPR + GC/non-linear 162 273 - 353 21.7 110 QSPR + GC/MLR 162 273 - 353 28.4 2008 QSPR/MLR 32 293 111 2009 GC 24 293 - 393 7.5 112 2011 QSPR 99 298 113 2011 QSPR 72 253 - 409 34.0 114 2011 QSPR 255 258 - 433 13.6 115 2011 QSPR/ANN 58 298 - 333 4.8 116 2012 GC/MLR 443 253 - 433 31.0 117 2012 QSPR/MLR 293 253 - 373 8.8 118 2012 QSPR/MLR 696 119 2013 QSPR/ANN 81 273 - 388 6.6 120 2013 QSPR/MLR 26 258 - 433 9.5 121 2013 QSPR/MLR 146 283 - 343 122 2014 GC/ANN 1484 253 - 573 11.4 123 2015 GC 326 253 - 395 22.3 124 2015 QSPR/MLR 89 253 - 395 59.4 125 QSPR/LSSVM 89 253 - 395 30.2 2015 QSPR/MLR 27 298 126 2016 QSPR/CMIS 23 273 - 388 127 2017 QSPR/ANN 59 273 - 393 1.3 128 2017 QSPR/ELM 89 253 - 395 ≈ 20 129 2018 QSPR/MLR 349 253 - 573 8.2 130 2019 QSPR/ANN 38 258 - 408 6.9 131 2019 GC/LSSVM 2068 290 - 360 30.2 132
In general, the viscosity of ILs is decided by temperature and their structure. The viscosity decreases dramatically with the increase of temperature, this is to say, it is
25
possible to optimize the operating temperature to meet the requirement for the viscosity of IL in a CAILD problem. Meantime, the viscosity of ILs with same anion follow the order119: imidazolium < pyridinium < pyrrolidinium < oxazolidinium < piperidinium <
morphonium. The viscosity of the ILs with aromatic cations are lower than ILs with nonaromatic cation. This is because an aromatic cation has a more delocalized positive charge, which decreases the interaction between cation and anion. Moreover, the viscosity values of the phosphonium ILs are lower than those of the ammonium ILs, despite the fact that phosphonium ILs display higher atomic weight133. The functional
groups on cations also influence the viscosity. Except for the -O- group, all the other functional groups such as -OH, -COOH, -C≡N, and -F could increase the viscosity. The ILs with -O- tend to have lower viscosity due to the more flexible rotation caused by the increase of conformational degrees of freedom134. The increase in viscosity from
other functional groups may be related to the strong interaction between cation and anion caused by the hydrogen bond and van der Waals forces.
The viscosity of the ILs with the same cation increase in the order119: [DCA]- <
[Tf2N]- < [TfO]- < [BF4]- < [PF6]- < [OAc]-. The lower viscosity of the ILs with [DCA]
-and [Tf2N]- anions is due to the highly delocalized negative charge distribution on the
anion which can weaken the cation-anion charge interaction. The higher viscosity for [BF4]- and [PF6]- anion is because of their rigid structural characteristic and the lack of
conformational degrees of freedom135. The ILs with [OAc]- anion possess the highest
viscosity, this is because of the localization of negative charge on -COO group introduces a strong hydrogen bonding and electrostatic interaction to the IL136.
26 4.3. Toxicity
Because of their negligible vapor pressure, atmospheric pollution by ILs is unlikely, and for this reason, ILs are often been referred to as “green” solvents compared to the VOCs. However, in recent years, the hazardous potential of ILs towards the ecosystem has been recognized. Due to their significant solubility in water, ILs could be released to the environment by industrial wastewater (Fig. 12). Moreover, because of their non-volatile nature and resistance to decomposition by aquatic microorganisms, ILs can bioaccumulate in the environment. The high thermal and chemical stability of ILs could further accelerate this bioaccumulation. Although ILs can possess wide-ranging toxicity137,138, this important property is not included in
current CAILD works as shown in Table 1.
Fig. 12. Transport and transformation of ILs in the environmental system139 (Copyright
2015, American Chemical Society).
The toxicity of ILs has been studied using different receptors including vibrio fischeri (v. fischeri), staphylococcus aureus (s. aureus), leukemia rat cell line (IPC-81) scenedesmus vacuolatus (s. vacuolatus), daphnia magna (d. magna), etc., where v. fischeri and IPC-81 are the most commonly used. The main index of IL toxicity is EC50 (μM) which denotes the half-maximal effective concentration. According to the criteria
27
proposed by the UFT research unit from the University of Bremen, the cytotoxicity of ILs towards IPC-81 can be divided into four categories, i.e., very high (EC50 < 1.0), high (1.0 < EC50 <100), moderate (100 < EC50 < 5000), and low (EC50 > 5000). Because of the huge amount of cation-anion combinations, the experimental toxicity measurements are time-consuming and expensive. In order to predict the toxicity of ILs efficiently, different models have been developed and summarized in Table 5.
Table 5. Summary of the reported models for the perdition of toxicity of ILs.
Year Method No. of ILs R2 Ref.
Vibrio fischeri (V. fischeri)
2007 GCM 43 0.925 140 2010 QSPR/MLR 75 0.91 141 2011 QSPR/heuristic 51 0.85 142 2013 QSPR/MLR 97 0.762 143 2015 QSPR/MLR 24 0.954 144 2015 QSPR/LSSVM 69 0.933 145 2015 QSPR/MLR 157 0.908 146 2015 QSPR/MLR 40 0.91 147 2016 QSPR/MLR 56 0.78 148 2017 QSPR/MLR 110 0.91 149 QSPR/MLP 110 0.979
Leukemia Rat Cell Line (IPC-81)
2011 QSPR/MLR 227 0.91 150 QSPR/MLP 227 0.98 2013 QSPR/MLR 97 0.778 143 2013 GCM 281 0.925 151 2014 QSPR/MLR 100 0.918 152 QSPR/SVM 100 0.958 2015 QSPR/MLR 100 0.813 153 2015 QSPR/classfication 253 > 0.9 154 2015 QSPR/MLR 289 0.869 155 2016 QSPR/MLR 17 0.999 138 2017 QSPR/MLR 304 0.772 156 2018 QSPR/MLR 119 0.92 157 QSPR/SVM 119 0.941 QSPR/ELM 119 0.969 2018 QSPR/PLS 269 0.86 137 QSPR/SVM 269 0.89 2020 QSPR/MLR 127 0.925 158 QSPR/MLP 127 0.97
28
The structure-activity models listed in Table 5 could also be used for identifying the factors affecting the toxicity of ILs (Fig. 13). Generally, phosphonium and imidazolium ILs are more toxic. Moreover, the dominant role of the length of the alkyl chain in cation was discovered for different types of organisms used for toxicity assessment of ILs159. To be specific, ILs with a longer side chain tend to be more toxic.
This is because the ILs with longer alkyl chains could be incorporated into the polar head groups of the phospholipid bilayer and damage the cell membrane158. For the
structure of the anion, it was found that ILs with fluorine-containing anion are more toxic, and the toxicity increase with the number of F atom in the anion: [BF4]- < [PF6]
-< [Tf2N]- < [eFAP]-. It should be noted that compared to the cation, the anion has a
minor influence on the toxicity of ILs158.
Cation
Phosphonium and imidazolium ILs
Branched cation
Positively charged atoms and N-atoms in the cations Anion
Fluorine-containing ILs Hydrophobic anions Substituent
Long hydrophobic alkyl chains
Cation
Pyridinium and morpholinium ILs with short polar linear alkyl chains
Anion
Alkyl sulphates and organic anions
Substituent
Short polar linear alkyl chains
Hazardous Ionic Liquids Green Ionic Liquids
Fig. 13. Overview of the influence of the different factors on the toxicity of ILs160
(Copyright 2019, Elsevier).
Concerning the substituents on the cation, as described above, because lipophilicity is an essential factor that influences the toxicity of ILs, the functional groups or structures increasing the lipophilicity will also increase the toxicity of ILs. For this reason, increasing branching and the presence of nitrogen functional groups in the cation were proven to significantly increase toxicity towards d. magna and v. fischeri161. On the contrary, the introduction of oxygen groups into the alkyl side chain
29
5. Solving the CAILD problem
5.1. Generate-and-test
Because many models for the prediction of thermophysical properties used in CAILD are not very complex, this entails only a modest need for computational time and costs. Most of the methods discussed above can provide the prediction for millions of ILs in a matter of minutes. Therefore, as shown in Table 1, many CAILD methods generate a large number of IL candidates and then evaluate their thermophysical properties of interest one by one. This approach is known as the “generate-and-test”, which is often used in the IL screening problem by the COSMO-based thermodynamic model, where the σ-profiles for every ILs are generated and then used to calculate the activity coefficient. Besides, the generate-and-test method is also suitable for the CAILD problem that is potentially difficult to solve, e.g., CAILD problem with complex constraints. This is because the generate-and-test method can skip the step of solving the mathematical problem by testing all the possible combinations. However, CAILD problems with a large design space are not efficiently solved with this approach since it can take a quite long time to evaluate all the potential structures of ILs. In this situation, the use of optimization methods is a better choice.
5.2. Decomposition
Many CAILD problems are characterized by a large design space, challenging non-linear nature in thermodynamic of process models, and different types of variables. As a result, these CAILD problems are very hard to be directly solved as optimization problems. Therefore, it is more practical to divide the original problem into several subproblems, which are easier to be solved. As shown in Fig. 14, two strategies can be used to decompose the optimization problem. In strategy A, the design space can be decomposed into several parallel subproblems, the solution for each subproblem is compared and the best solution among all subproblems is the solution for the optimization problem. This strategy has been used to solve the CAILD problem for the
30
benzene/cyclohexane extraction in Chapter 3 of this thesis. Strategy B is more popular among researchers63,67, typically, the subproblems successively apply by increasing the
number or difficulty of the constraints from the original problem. As a result, the feasible set of ILs upon the solution of each subproblem will be decreased. This step-wise reduction in design space makes many CAILD problems significantly easier. Strategy B has been used to solve the biogas upgrading problem in Chapter 4 in this thesis.
Design space
Subproblem 1 Subproblem 2 Subproblem 1 Subproblem 2 Strategy A Strategy BFig. 14. Two strategies are applied in the decomposition method. 5.3. Mathematical optimization
In some cases, CAILD problems can be addressed straightforwardly with the mathematical optimization algorithm. This approach is best suited to problems with many variables or large design space. In cases where there are only a few possible variables, the CAILD problem can be efficiently enumerated using the generate-and-test method. The mathematical optimization method can be combining with other approaches, such as the decomposition method. As shown in Chapter 3, the CAILD problem is firstly decomposed into several subproblems and each subproblem is then solved by the mathematical algorithms. Because the CAILD is always formulated as an
31
MINLP problem, the solvers (e.g. BARON, CONOPT, LINDOGLOBAL) based on the outer-approximation algorithm and branch-and-bound algorithm can be applied47,66.
5.4. Heuristics
Although the mathematical algorithms are useful tools for the CAILD problem, they are not capable of the high-dimensional and non-linear optimization problem. This type of problem always contains a lot of variables, which leads to a large search space with enormous possible combinations of descriptors. To solve this kind of CAILD problem effectively, the heuristic approaches, such as genetic algorithm (GA) and tabu search (TS) are employed by researchers60,68,163. In the heuristic algorithms, the
high-level selection strategies are used to produce a series of trial points towards the complex optimization problems. For the CAILD problem, the structure of the ILs is the variable needs to be optimized, which is generated based on certain constraints. Firstly, the structure of IL is treated as an input to the entire CAILD problem, and the objective value is calculated. Then, a new IL structure (or series of structures) is selected for the next evaluation based on the relationship between the objective function value and certain characteristics (structures, physical properties, etc). Finally, the algorithm is terminated when converging to a point of optimality or reaching the maximum execution time and iterations.
32
6. Aim and scope of this work
This thesis aims at developing the CAILD methods to design ILs with relatively low toxicity for different separation problems including extraction and gas absorption. Unlike the CAILD methods available in the literature as shown in Table 1, the toxicity of ILs is taken into consideration and treated as a constraint in the CAILD methods. Moreover, some of the required IL candidates were experimentally verified for the extraction problems which provides solid proof for the reliability of the proposed CAILD methods.
In Chapter 2, an efficient and accurate QSPR model for the prediction of IL toxicity is built which can be used in the CAILD framework. The GC-COSMO method is applied to generate the σ-profile of ILs and the related descriptors for the model. Then, the linear and nonlinear QSPR models are developed using Multi-Linear Regression (MLR) and Multi-Layer Perceptron technique (MLP), respectively. Compared to other QSPR models in the literature, the QSPR method developed in this chapter is more efficient and can be easily integrated into CAILD methods to design green ILs with low toxicity.
In Chapter 3, a CAILD method is proposed to design suitable ILs for the extractive separation of benzene/cyclohexane which have close physical properties and hard to be separated by traditional distillation. The UNIFAC-IL model is employed to estimate the activity coefficient of solutes in ILs while the QSPR models are used to predict the physical properties including the toxicity which is calculated by the method proposed in Chapter 2. After generating the promising IL candidates by decomposition method, the liquid-liquid equilibrium (LLE) experiments are performed using one of the ILs to validate the CAILD method.
In Chapter 4, a CAILD method is developed for biogas upgrading. As a potential alternative for fossil fuels, biogas needs to be upgraded to biomethane to increase the calorific value. Considering that the main components in biogas are CH4 and CO2, the
33
biogas upgrading problem can be regarded as the separation of CH4/CO2. For this
purpose, a CAILD method is presented, where the solubility of gases in ILs and the physical properties are calculated by the UNIFAC-IL model and QSPR models, respectively. The toxicity of ILs is also included in the constraint to ensure the designed ILs are environmentally friendly. The generate-and-test method is used to solve the MINLP problem by testing 880 possible IL structures. Finally, the selected ILs are evaluated in the corresponding process simulated in Aspen Plus, and the results are compared with the traditional water-scrubbing process.
In Chapter 5, a hierarchical hybrid method for the screening of ILs as extraction solvents is presented and exemplified by the extractive desulfurization process. The common practice for screening ILs is based on the COSMO-SAC or COSMO-RS model. However, for the IL-involved systems, the accuracy of the COSMO-based models for the prediction of infinite dilution activity coefficients (IDAC) is inferior to other models due to its fully predictive character. Considering that there is an increasing number of experimental IDAC data published nowadays, it is possible to build a large experimental IDAC database and use it to screen ILs for specific separation problems. For this purpose, an IDAC database contains 47424 datapoints is built where the name of IL and solute, classification, IDAC value, temperature, and references are recorded. The database is then used to rank the separation performance of different ILs towards the given extraction problems by a Matlab program. The ILs exhibit promising extractive performance and proper physical properties (low melting point, viscosity, and toxicity) are selected. After that, those ILs are used as solvents to perform the LLE experiments, and the corresponding parameters for the non-random two-liquid model (NRTL) are regressed. Finally, the process simulation based on the acquired NRTL parameters is developed and compared with the process using conventional solvents.
34
Abbreviations
ANN artificial neural network
ARD average relative deviation
CAILD computer-aided ionic liquids design
CAMD computer-aided molecular design
COSMO-RS conductor-like screening model for real solvent
COSMO-SAC conductor-like screening model for segment activity coefficient
EC50 half-maximal effective concentration
EDS extractive desulfurization
FFANN feed-forward artificial neural network
GA genetic algorithm
GC group contribution
GC-COSMO GC-based COSMO-SAC
IDAC infinite dilution activity coefficients
ILs Ionic liquids
IPC-81 leukemia rat cell line
LLE liquid-liquid equilibrium
LSSVM least-squares support vector machine
MAD mean absolute deviation
MAE mean absolute errors
MINLP mixed-integer nonlinear programming
ML machine learning
MLP multi-layer perceptron technique
MLR multi-linear regression
NRTL non-random two-liquid model
QSPR quantitative structure-property relationship
RTILs room-temperature ionic liquids
SVM support vector machine
SWMLR stepwise multiple linear regression
TS tabu search
UNIFAC universal quasi-chemical functional-group activity coefficients
35
VOCs volatile organic compounds
[DCA]- dicyanamide [AlCl4]- tetrachloroaluminate [Tf2N]- bis(trifluoromethylsulfonyl)imide [SCN]- thiocyanate [CH3SO4]- methylsulphate [BF4]- tetrafluoroborate [PF6]- hexafluorophosphate [TfO]- triflate [OAc]- acetate [eFAP]- tris(nonafluoroethyl)trifluoro-phosphate
36
References
1 J. P. Hallett and T. Welton, Chem. Rev., 2011, 111, 3508-3576. 2 Walden, Bull. Acad. Imp. Sci. St.- Petersbg., 1914, 6, 405-422.
3 R. Hayes, G. G. Warr and R. Atkin, Chem. Rev., 2015, 115, 6357-6426.
4 V. N. Emel’Yanenko, G. Boeck, S. P. Verevkin and R. Ludwig, Chem. Eur. J., 2014, 20, 11640-11645.
5 F. H. Hurley and T. P. WIer, J. Electrochem. Soc., 1951, 98, 207-212. 6 M. V. Fedorov and A. A. Kornyshev, Chem. Rev., 2014, 114, 2978-3036. 7 J. S. Wilkes and M. J. Zaworotko, J. Chem. Soc., Chem. Commun., 1992,
965-967.
8 H. Qi, Y. Ren, S. Guo, Y. Wang, S. Li, Y. Hu and F. Yan, ACS Appl. Mater. Interfaces, 2020, 12, 591-600.
9 D. Y. Oh, Y. J. Nam, K. H. Park, S. H. Jung, K. T. Kim, A. R. Ha and Y. S. Jung, Adv. Energy Mater., 2019, 9, 1802927-1802936.
10 M. Forsyth, L. Porcarelli, X. Wang, N. Goujon and D. Mecerreyes, Acc. Chem. Res., 2019, 52, 686-694.
11 S. H. Lee, Y. D. Lim, D. W. Seo, M. A. Hossain, H. H. Jang, H. C. Lee and W. G. Kim, J. Ind. Eng. Chem., 2013, 19, 322-326.
12 M. Wang, Z. Shan, J. Tian, K. Yang, X. Liu, H. Liu and K. Zhu, Electrochim. Acta, 2013, 95, 301-307.
13 K. Asaka and M. Hashimoto, Smart Mater. Struct., 2020, 29, 25003-25012. 14 F. Ma, Z. Zhang, W. Yan, X. Ma, D. Sun, Y. Jin, X. Chen and K. He, ACS Sustain.
Chem. Eng., 2019, 7, 4675-4683.
15 S. Qu, M. Li, C. Zhang, Y. Sun, J. Duan, W. Wang, J. Li and X. Li, Polymers, 2018, 11, 1-14.
16 A. Wang, H. Xu, X. Liu, R. Gao, S. Wang, Q. Zhou, J. Chen, X. Liu and L. Zhang, Polym. Chem., 2017, 8, 3177-3185.
17 R. González, M. Bartolomé, D. Blanco, J. L. Viesca, A. Fernández-González and A. H. Battez, Tribol. Int., 2016, 98, 82-93.
18 R. Gusain, S. Dhingra and O. P. Khatri, Ind. Eng. Chem. Res., 2016, 55, 856-865.
37
G. Freire, C. S. R. Freire and A. J. D. Silvestre, ACS Sustain. Chem. Eng., 2019, 7, 14126-14134.
20 E. E. L. Tanner, A. M. Curreri, J. P. R. Balkaran, N. C. Selig‐Wober, A. B. Yang, C. Kendig, M. P. Fluhr, N. Kim and S. Mitragotri, Adv. Mater., 2019, 31, 1901103-1901112.
21 D. Monti, E. Egiziano, S. Burgalassi, P. Chetoni, C. Chiappe, A. Sanzone and S. Tampucci, Int. J. Pharm., 2017, 516, 45-51.
22 N. Adawiyah, M. Moniruzzaman, S. Hawatulaila and M. Goto, Medchemcomm, 2016, 7, 1881-1897.
23 J. N. Pendleton and B. F. Gilmore, Int. J. Antimicrob. Agents, 2015, 46, 131-139. 24 K. Kuroda, Y. Kohno and H. Ohno, Chem. Lett., 2017, 46, 870-872.
25 P. Lozano, C. Gomez, A. Nicolas, R. Polo, S. Nieto, J. M. Bernal, E. García-Verdugo and S. V Luis, ACS Sustain. Chem. Eng., 2016, 4, 6125-6132.
26 P. Lozano, J. M. Bernal, E. Garcia-Verdugo, G. Sanchez-Gomez, M. Vaultier, M. I. Burguete and S. V Luis, Green Chem., 2015, 17, 3706-3717.
27 P. Xu, G.-W. Zheng, P.-X. Du, M.-H. Zong and W.-Y. Lou, ACS Sustain. Chem. Eng., 2015, 4, 371-386.
28 J. V Rodrigues, D. Ruivo, A. Rodríguez, F. J. Deive, J. M. S. S. Esperança, I. M. Marrucho, C. M. Gomes and L. P. N. Rebelo, Green Chem., 2014, 16, 4520-4523.
29 R. Teixeira and N. M. T. Lourenço, Tetrahedron: Asymmetry, 2014, 25, 944-948. 30 G. M. J. Al Kaisy, M. I. Abdul Mutalib, M. A. Bustam, J.-M. Leveque and N.
Muhammad, J. Environ. Chem. Eng., 2016, 4, 4786-4793. 31 A. Ramalingam and A. Balaji, J. Mol. Liq., 2015, 212, 372-381.
32 S. S. AlSaleem, W. M. Zahid, I. M. Alnashef and H. Haider, Sep. Purif. Technol., 2017, 184, 231-239.
33 W. Bai, Y. Dai, X. Pan, Z. Zhu, Y. Wang and J. Gao, J. Chem. Thermodyn., 2019, 132, 76-82.
34 C. Ma, S. K. Shukla, R. Samikannu, J.-P. Mikkola and X. Ji, ACS Sustain. Chem. Eng., 2020, 8, 415-426.
35 M. Mu, J. Cheng, C. Dai, N. Liu, Z. Lei, Y. Ding and J. Lu, Green Energy Environ., 2019, 4, 190-197.
38
37 G. Yu, C. Dai, H. Gao, R. Zhu, X. Du and Z. Lei, Ind. Eng. Chem. Res., 2018, 57, 12202-12214.
38 M. Jin, Y. Hou, W. Wu, S. Ren, S. Tian, L. Xiao and Z. Lei, J. Phys. Chem. B, 2011, 115, 6585-6591.
39 K. P. Paduszyński, M. Kroíikowski, M. Zawadzki and P. Orzeł, ACS Sustain. Chem. Eng., 2017, 5, 9032-9042.
40 Y. Zhang, X. Ji, Y. Xie and X. Lu, Appl. Energy, 2016, 162, 1160-1170. 41 Z. Song, C. Zhang, Z. Qi, T. Zhou and K. Sundmacher, AIChE J., 2017, 64,
1-13.
42 S. Yuan, Y. Chen, X. Ji, Z. Yang and X. Lu, Fluid Phase Equilib., 2017, 445, 14-24.
43 Z. Lyu, T. Zhou, L. Chen, Y. Ye, K. Sundmacher and Z. Qi, Chem. Eng. Sci., 2014, 115, 186-194.
44 P. Wang, P. Yan, J. A. Reyes-Labarta, J. Gao, D. Xu, L. Zhang and Y. Wang, Fluid Phase Equilib., 2019, 485, 183-189.
45 M. P. Cumplido, A. Cháfer, J. de la Torre and H. Poy, J. Chem. Thermodyn., 2020, 140, 105889-105896.
46 F. S. Oliveira, R. Dohrn, A. B. Pereiro, J. M. M. Araújo, L. P. N. Rebelo and I. M. Marrucho, Fluid Phase Equilib., 2016, 419, 57-66.
47 Y. Chen, R. Gani, G. M. Kontogeorgis and J. M. Woodley, Chem. Eng. Sci., 2019, 203, 402-414.
48 F. Cai and G. Xiao, Fluid Phase Equilib., 2015, 386, 155-161. 49 Y. Zhang, X. Ji, Y. Xie and X. Lu, Appl. Energy, 2018, 217, 75-87.
50 R. Farahipour, A. Mehrkesh and A. T. Karunanithi, Chem. Eng. Sci., 2016, 145, 126-132.
51 X. Zhang, Z. Liu and W. Wang, AIChE J., 2008, 54, 2717-2728.
52 D. Wappel, G. Gronald, R. Kalb and J. Draxler, Int. J. Greenh. Gas Control, 2010, 4, 486-494.
53 B. E. Gurkan, J. C. De La Fuente, E. M. Mindrup, L. E. Ficke, B. F. Goodrich, E. A. Price, W. F. Schneider and J. F. Brennecke, J. Am. Chem. Soc., 2010, 132, 2116-2117.
54 L. Fischer, T. Falta, G. Koellensperger, A. Stojanovic, D. Kogelnig, M. Galanski, R. Krachler, B. K. Keppler and S. Hann, Water Res., 2011, 45, 4601-4614.
39
55 E. Reyhanitash, B. Zaalberg, S. R. A. Kersten and B. Schuur, Sep. Purif. Technol., 2016, 161, 61-68.
56 S. S. Alsaleem, W. M. Zahid, I. M. Alnashef and M. K. Hadj-Kali, J. Chem. Eng. Data, 2016, 60, 2926-2936.
57 M. Fuerhacker, T. M. Haile, D. Kogelnig, A. Stojanovic and B. Keppler, Water Sci. Technol., 2012, 65, 1765-1773.
58 U. Domańska and M. Wlazło, Fuel, 2014, 134, 114-125.
59 S. Trohalaki and R. Pachter, QSAR Comb. Sci., 2005, 24, 485-490.
60 S. E. McLeese, J. C. Eslick, N. J. Hoffmann, A. M. Scurto and K. V. Camarda, Comput. Chem. Eng., 2010, 34, 1476-1480.
61 R. Anantharaj and T. Banerjee, Ind. Eng. Chem. Res., 2010, 49, 8705-8725. 62 L. M. Chávez-Islas, R. Vasquez-Medrano and A. Flores-Tlacuahuac, Ind. Eng.
Chem. Res., 2011, 50, 5153-5168.
63 A. T. Karunanithi and A. Mehrkesh, AIChE J., 2013, 59, 4627-4640.
64 M. Harini, S. Jain, J. Adhikari, S. B. Noronha and K. Yamuna Rani, Sep. Purif. Technol., 2014, 155, 45-57.
65 B. S. Lee and S. T. Lin, Chem. Eng. Sci., 2015, 121, 157-168.
66 D. Peng, J. Zhang, H. Cheng, L. Chen and Z. Qi, Chem. Eng. Sci., 2017, 159, 58-68.
67 Z. Song, C. Zhang, Z. Qi, T. Zhou and K. Sundmacher, AIChE J, 2017, 64, 1013-1025.
68 J. Wang, Z. Song, H. Cheng, L. Chen, L. Deng and Z. Qi, ACS Sustain. Chem. Eng., 2018, 6, 12025-12035.
69 Y. Chen, S. Zhou, Y. Wang and L. Li, Fluid Phase Equilib., 2017, 451, 12-24. 70 A. Fredenslund, R. L. Jones and J. M. Prausnitz, AIChE J., 1975, 21, 1086-1099. 71 Z. Lei, J. Zhang, Q. Li and B. Chen, Ind. Eng. Chem. Res, 2009, 48, 2697-2704. 72 Z. Lei, C. Dai, X. Liu, L. Xiao and B. Chen, Ind. Eng. Chem. Res, 2012, 51,
12135-12144.
73 Z. Lei, C. Dai, W. Wang and B. Chen, AIChE J., 2014, 60, 716-729.
74 Y. Dong, Y. Guo, R. Zhu, J. Zhang and Z. Lei, Ind. Eng. Chem. Res, 2020, 59, 10172-10184.
40 1013-1025.
76 Z. Song, X. Li, H. Chao, F. Mo, T. Zhou, H. Cheng, L. Chen and Z. Qi, Green Energy Environ., 2018, 4, 154-165.
77 Y. Dong, S. Huang, Y. Guo and Z. Lei, AIChE J., 2020, 66, 1-5. 78 Z. Song, T. Zhou, Z. Qi and K. Sundmacher, AIChE J., 2020, 66, 1-15. 79 S.-T. Lin and S. I. Sandler, Ind. Eng. Chem. Res, 2002, 41, 899-913.
80 S. Wang, S. I. Sandler and C.-C. Chen, Ind. Eng. Chem. Res, 2007, 46, 7275-7288.
81 C. M. Hsieh, S. T. Lin and J. Vrabec, Fluid Phase Equilib., 2014, 367, 109-116. 82 A. Klamt and F. Eckert, Fluid Phase Equilib., 2000, 172, 43-72.
83 C.-M. Hsieh and S.-T. Lin, Fluid Phase Equilib., 2009, 285, 8-14.
84 and Y. H. Li Yang, Xiaochun Xu, Changjun Peng, Honglai Liu, AIChE J., 2010, 56, 2687-2698.
85 Y. Chen, S. Zhou, Y. Wang and L. Li, Fluid Phase Equilib., 2017, 451, 12-24. 86 L. Bong-Seop and S.-T. Lin, VTT Publ., 2017, 63, 3-194.
87 I. H. Bell, E. Mickoleit, C. M. Hsieh, S. T. Lin, J. Vrabec, C. Breitkopf and A. Jäger, J. Chem. Theory Comput., 2020, 16, 2635-2646.
88 W. Tu, L. Bai, S. Zeng, H. Gao, S. Zhang and X. Zhang, J. Mol. Liq., 2019, 282, 292-302.
89 R. Fingerhut, W. L. Chen, A. Schedemann, W. Cordes, J. Rarey, C. M. Hsieh, J. Vrabec and S. T. Lin, Ind. Eng. Chem. Res., 2017, 56, 9868-9884.
90 A. Bharti and T. Banerjee, Fluid Phase Equilib., 2015, 400, 27-37. 91 K. Paduszyński, J. Chem. Inf. Model., 2016, 56, 1420-1437.
92 Z. Song, H. Shi, X. Zhang and T. Zhou, Chem. Eng. Sci., 2020, 223, 115752-115758.
93 P. Dyson and G. Tilmann, Eds., Metal Catalysed Reactions in Ionic Liquids, Springer Netherlands, Dordrecht, 2005.
94 S. Zhang, N. Sun, X. He, X. Lu and X. Zhang, J. Phys. Chem. Ref. Data, 2006, 35, 1475-1517.
95 J. A. Lazzús, Fluid Phase Equilib., 2012, 313, 1-6.
96 R. Bini, C. Chiappe, C. Duce, A. Micheli, R. Solaro, A. Starita and M. R. Tiné, Green Chem., 2008, 10, 306-309.
41
97 J. S. Torrecilla, F. Rodríguez, J. L. Bravo, G. Rothenberg, K. R. Seddon and I. López-Martin, Phys. Chem. Chem. Phys., 2008, 10, 5826-5831.
98 Y. Huo, S. Xia, Y. Zhang and P. Ma, Ind. Eng. Chem. Res., 2009, 48, 2212-2217. 99 C. Yan, M. Han, H. Wan and G. Guan, Fluid Phase Equilib., 2010, 292,
104-109.
100 C. L. Aguirre, L. A. Cisternas and J. O. Valderrama, Int. J. Thermophys., 2011, 33, 34-46.
101 N. Farahani, F. Gharagheizi, S. A. Mirkhani and K. Tumba, Thermochim. Acta, 2012, 549, 17-34.
102 F. Gharagheizi, P. Ilani-Kashkouli and A. H. Mohammadi, Fluid Phase Equilib., 2012, 329, 1-7.
103 J. A. Lazzús, Period. Polytech. Chem. Eng., 2016, 60, 273-281.
104 L. Chen and V. S. Bryantsev, Phys. Chem. Chem. Phys., 2017, 19, 4114-4124. 105 M. H. Keshavarz, H. R. Pouretedal and E. Saberi, Process Saf. Environ. Prot.,
2018, 116, 333-339.
106 V. Venkatraman, S. Evjen, H. K. Knuutila, A. Fiksdahl and B. K. Alsberg, J. Mol. Liq., 2018, 264, 318-326.
107 Y. Chen, G. M. Kontogeorgis and J. M. Woodley, Ind. Eng. Chem. Res., 2019, 58, 4277-4292.
108 W. D. Monnery, W. Y. Svrcek and A. K. Mehrotra, Can. J. Chem. Eng., 1995, 73, 3-40.
109 H. Matsuda, H. Yamamoto, K. Kurihara and K. Tochigi, J. Comput. Aided Chem., 2007, 8, 114-127.
110 K. Tochigi and H. Yamamoto, J. Phys. Chem. C, 2007, 111, 15989-15994. 111 R. Bini, M. Malvaldi, W. R. Pitner and C. Chiappe, J. Phys. Org. Chem., 2008,
21, 622-629.
112 R. L. Gardas and J. A. P. Coutinho, AIChE J., 2009, 55, 1274-1290.
113 I. Billard, G. Marcou, A. Ouadi and A. Varnek, J. Phys. Chem. B, 2011, 115, 93-98.
114 P. Eiden, S. Bulut, T. Köchner, C. Friedrich, T. Schubert and I. Krossing, J. Phys. Chem. B, 2011, 115, 300-309.
115 C. Han, G. Yu, L. Wen, D. Zhao, C. Asumana and X. Chen, Fluid Phase Equilib., 2011, 300, 95-104.
42
116 J. O. Valderrama, J. M. Muñoz and R. E. Rojas, Korean J. Chem. Eng., 2011, 28, 1451-1457.
117 F. Gharagheizi, P. Ilani-Kashkouli, A. H. Mohammadi, D. Ramjugernath and D. Richon, Chem. Eng. Sci., 2012, 80, 326-333.
118 S. A. Mirkhani and F. Gharagheizi, Ind. Eng. Chem. Res., 2012, 51, 2470-2477. 119 G. Yu, D. Zhao, L. Wen, S. Yang and X. Chen, AIChE J., 2011, 58, 2885-2899. 120 N. V. K. Dutt, Y. V. L. Ravikumar and K. Y. Rani, Chem. Eng. Commun., 2013,
200, 1600-1622.
121 B.-K. Chen, M.-J. Liang, T.-Y. Wu and H. P. Wang, Fluid Phase Equilib., 2013, 350, 37-42.
122 G. Yu, L. Wen, D. Zhao, C. Asumana and X. Chen, J. Mol. Liq., 2013, 184, 51-59.
123 K. Paduszyński and U. Domańska, J. Chem. Inf. Model., 2014, 54, 1311-1324. 124 J. A. Lazzús and G. Pulgar-Villarroel, J. Mol. Liq., 2015, 209, 161-168.
125 Y. Zhao, Y. Huang, X. Zhang and S. Zhang, Phys. Chem. Chem. Phys., 2015, 17, 3761-3767.
126 R. Alcalde, G. García, M. Atilhan and S. Aparicio, Ind. Eng. Chem. Res., 2015, 54, 10918-10924.
127 Z. Esmaeili-Jaghdan, A. Shariati and M. R. K. Nikou, J. Mol. Liq., 2016, 222, 14-27.
128 M.-R. Fatehi, S. Raeissi and D. Mowla, J. Mol. Liq., 2017, 227, 309-317. 129 X. Kang, Z. Zhao, J. Qian and R. Muhammad Afzal, Ind. Eng. Chem. Res., 2017,
56, 11344-11351.
130 F. Yan, W. He, Q. Jia, Q. Wang, S. Xia and P. Ma, Chem. Eng. Sci., 2018, 184, 134-140.
131 R. Haghbakhsh and S. Raeissi, Phys. Chem. Liq., 2018, 57, 401-421. 132 K. P. Paduszyński, Ind. Eng. Chem. Res., 2019, 58, 17049-17066.
133 S. Seki, T. Kobayashi, Y. Kobayashi, K. Takei, H. Miyashiro, K. Hayamizu, S. Tsuzuki, T. Mitsugi and Y. Umebayashi, J. Mol. Liq., 2010, 152, 9-13.
134 Z.-B. Zhou, H. Matsumoto and K. Tatsumi, Chem. Eur. J., 2004, 10, 6581-6591. 135 Z.-B. Zhou, H. Matsumoto and K. Tatsumi, Chem. Eur. J., 2005, 11, 752-766. 136 X. Liu, G. Zhou, S. Zhang, G. Wu and G. Yu, J. Phys. Chem. B, 2007, 111,
5658-43 5668.
137 S. R. Farahani, M. R. Sohrabi and J. B. Ghasemi, Ecotoxicol. Environ. Saf., 2018, 158, 256-265.
138 M. A. Salam, B. Abdullah, A. Ramli and I. M. Mujtaba, J. Mol. Liq., 2016, 224, 393-400.
139 M. Amde, J. F. Liu and L. Pang, Environ. Sci. Technol., 2015, 49, 12611-12627. 140 P. Luis, I. Ortiz, R. Aldaco and A. Irabien, Ecotoxicol. Environ. Saf., 2007, 67,
423-429.
141 P. Luis, A. Garea and A. Irabien, J. Mol. Liq., 2010, 152, 28-33.
142 S. Bruzzone, C. Chiappe, S. E. Focardi, C. Pretti and M. Renzi, Chem. Eng. J., 2011, 175, 17-23.
143 C.-W. Cho, J. Ranke, J. Arning, J. Thöming, U. Preiss, C. Jungnickel, M. Diedenhofen, I. Krossing and S. Stolte, SAR QSAR Environ. Res., 2013, 24, 863-882.
144 C. Wang, Z. Wei, L. Wang, P. Sun and Z. Wang, Ecotoxicol. Environ. Saf., 2015, 115, 112-118.
145 S. Ma, M. Lv, F. Deng, X. Zhang, H. Zhai and W. Lv, J. Hazard. Mater., 2015, 283, 591-598.
146 F. Yan, Q. Shang, S. Xia, Q. Wang and P. Ma, J. Hazard. Mater., 2015, 286, 410-415.
147 R. N. Das, K. Roy and P. L. A. Popelier, Ecotoxicol. Environ. Saf., 2015, 122, 497-520.
148 M. Grzonkowska, A. Sosnowska, M. Barycki, A. Rybinska and T. Puzyn, Chemosphere, 2016, 159, 199-207.
149 O. Ben Ghanem, M. I. A. Mutalib, J.-M. Lévêque and M. El-Harbawi, Chemosphere, 2017, 170, 242-250.
150 M. H. Fatemi and P. Izadiyan, Chemosphere, 2011, 84, 553-563.
151 M. Cruz-Monteagudo and M. N. D. S. Cordeiro, Toxicol. Sci., 2013, 138, 191-204.
152 Y. Zhao, J. Zhao, Y. Huang, Q. Zhou, X. Zhang and S. Zhang, J. Hazard. Mater., 2014, 278, 320-329.
153 E. B. de Melo, Phys. Chem. Chem. Phys., 2015, 17, 4516-4523.
12699-44 12710.
155 R. N. Das, K. Roy and P. L. A. Popelier, Chemosphere, 2015, 139, 163-173. 156 A. Sosnowska, M. Grzonkowska and T. Puzyn, J. Mol. Liq., 2017, 231, 333-340. 157 L. Cao, P. Zhu, Y. Zhao and J. Zhao, J. Hazard. Mater., 2018, 352, 17-26. 158 D. Peng and F. Picchioni, J. Hazard. Mater., 2020, 398, 122964-122975. 159 O. Ben Ghanem, M. I. A. Mutalib, M. El-Harbawi, G. Gonfa, C. F. Kait, N. B.
M. Alitheen and J.-M. Leveque, J. Hazard. Mater., 2015, 297, 198-206.
160 N. Abramenko, L. Kustov, L. Metelytsia, V. Kovalishyn, I. Tetko and W. Peijnenburg, J. Hazard. Mater., 2020, 384, 121429-121442.
161 K. Roy, R. N. Das and P. L. A. Popelier, Chemosphere, 2014, 112, 120-127. 162 A. Tot, M. Vraneš, I. Maksimović, M. Putnik-Delić, M. Daničić, S. Belić and S.
Gadžurić, Ecotoxicol. Environ. Saf., 2018, 147, 401-406.
163 J. Zhang, L. Qin, D. Peng, T. Zhou, H. Cheng, L. Chen and Z. Qi, Chem. Eng. Sci., 2017, 162, 364-374.
