Representation/prediction of physico-chemical properties of ionic liquids through different computational methods.
The “green” industrial chemical processes are of great interest to scientists and engineers due to elimination of environmental pollution, especially air pollution. One of the most important air pollutants is class of materials called volatile organic compounds (VOCs) which are widely used in different industrial chemical processes. The recent research has revealed that ionic liquids (ILs) are generally the best possible alternative to the conventional solvents; because in general, the ILs have interesting properties such as very low vapor pressure, nonflammability, and high physical and chemical stability. Ionic liquids are constituted of ions, typically a cation and an anion, and their thermophysical properties are strongly dependent on the type and chemical structure of the cation and anion. As a result, in theory, they can be designed for specific applications with certain properties by choosing the appropriate combination of anion/cation pair. For this purpose, a predictive model is required to estimate the target property based on the chemical structure of ions. At the initial step of this study, the NIST Standard Reference Database #103b as well as the published papers in the literature was chosen as the source of experimental data of ionic liquids. As a result, a large database was collected covering several thermophysical properties of ILs. Thereafter, the collected data were examined carefully and the duplicated and erroneous data were screened. Speed of sound, heat capacity, refractive index, viscosity, infinite dilution activity coefficient () , and critical temperature of various ionic liquids were modeled by means of two well-known property estimation methods, Group Contribution (GC) and Quantitative Structure-Property Relationship (QSPR) methods. These methods were combined with different computational and regression techniques such as genetic function approximation (GFA) and least square support vector machine (LS-SVM). The combined routines then were applied to select reasonable number of parameters from thousands of variables and to develop the predictive models for representation/prediction of chosen temperature-dependent thermophysical properties of ionic liquids. Speed of sound in ionic liquids was modeled successfully and two models were developed, one GC and one QSPR model. These models were the first GC and QSPR models developed for this property in the literature. Both models had better accuracy in terms of average absolute relative deviation (the AARD% of 0.36 for the GC and 0.92% for the QSPR models over 41 ILs) and covered a wider range of ionic liquids compared with the previous models published (AARD% of 1.96% over 14 ILs) and consequently, they were more applicable. Liquid heat capacity of ionic liquids was studied and one GC and one QSPR model were developed. Both models covered 82 ILs which was a larger number of ionic liquids compared with the best available model in the literature (32 ILs with an AARD% of 0.34%) and had relatively low AARD%. The AARD% of the models was 1.68% and 1.70% for the GC and QSPR models, respectively. In addition, the QSPR model was the first model developed for this property through the QSPR approach. For the refractive index of ionic liquids, little attention had been given to modeling and consequently, one new GC (AARD% = 0.34%) and the first QSPR (AARD% = 0.51%) models were developed to predict this property using the experimental data for 97 ionic liquids. Both models covered a wider range of ionic liquids and showed very good prediction ability compared with the best available model (an AARD% of 0.18% for 24 ILs). Viscosity of Fluorine-containing ionic liquids was studied because the insertion of fluorinated moieties in the molecular structure of ionic liquids could result in reduction of viscosity. As a result, one QSPR (AARD% = 2.91%) and two GC models were developed using two different databases, one with fewer number of ionic liquids but with more reliable data (AARD% = 3.23%), the one with larger number of ionic liquids but with lower reliability (AARD% = 4.85%). All of the models developed had better prediction ability compared with the previous models and covered a wider range of fluorinated ionic liquids. Infinite dilution activity coefficient (γ∞) of organic solutes was modeled by developing six different models for different types of solutes (alkane, alkene, aromatic, etc.). The model developed were the first GC models for the prediction of γ∞ of solutes in ionic liquids. They were much easier to use, more comprehensive, and much more accurate compared with the UNIFAC model. Ultimately, the theoretical critical temperature (Tc) of ionic liquids was tried to model using the GC and QSPR approaches. The experimental data of surface tension of 106 ionic liquids were used to calculate the critical temperature and then, these values were used to develop the models. It was found that the only available model in the literature was not accurate and predictive enough when its output was compared with the abovementioned Tc values. In addition, it was found that both of the models developed were not predictive enough to calculate the Tc of various types of ionic liquids as the models were developed using a few number of ionic liquids; however both models were accurate enough to fit the used values of Tc. The GC model has an AARD% of 5.17% and the QSPR model showed the AARD% of 4.69%. It this thesis, much larger databases were used to develop the models compared with the models published previously in the literature. It was found that thermophysical properties of ionic liquids can be modeled fairly well by combination of the GC or QSPR methods with an appropriate regression technique. In addition, the developed models improved significantly the quality of fit and predictions for a wider range of ionic liquids compared with the previous models. Consequently, the models proposed are more predictive and can be used to design the ionic liquids with desired property for specific applications.