Description
A Master of Science thesis in Chemical Engineering by Samira Jihad Zeinab entitled, “Predicting the Heats of Fusion of Ionic Liquids via Group Contribution Modeling and Machine Learning”, submitted in April 2022. Thesis advisor is Dr. Paul Nancarrow and thesis co-advisor is Dr. Nabil Abdel Jabbar. Soft copy is available (Thesis, Completion Certificate, Approval Signatures, and AUS Archives Consent Form).
Abstract
Energy security, environmental pollution, and global warming have become major concerns due to significant population and economic growth. The transition from fossil fuels, which can be used to generate power constantly, to intermittent renewable energy sources, such as solar or wind, requires the development of effective energy storage methods. Phase change materials (PCMs), a type of thermal energy storage technology that can absorb, store, and release thermal energy based on the temperature of the environment, hold significant potential in the energy storage mix. However, many such materials suffer from major drawbacks such as wide melting point ranges, supercooling, phase separation, evaporation, thermal degradation, or corrosion. Ionic liquids (ILs) have been considered as a promising substitute for standard PCMs in recent years, due to their non-volatility and thermal stability. While ILs are often described as designer materials, this potential must be utilized by developing structure-property models for the prediction of their key physiochemical properties. Heat of fusion is one of the most important material properties for PCM applications. In this work, the group contribution modelling (GCM) approach has been used as the basis for the development of a new IL heat of fusion predictive model. A database of IL heat of fusion data was compiled from a variety of literature sources with 344 data points and 289 unique ILs after data refinement. Extensive analysis of the structure-property relationships was used to develop several novel structural parameters that were incorporated into the GCM. A range of machine learning algorithms were investigated in combination with the GCM approach, including ridge regression, lasso regression, multi-layer neural networks, and gradient boosting regression, with the latter giving the best performance against the test set. A test set mean absolute error of 5.9 kJ.mol⁻¹ and R² of 0.67 between the predicted and experimental data was obtained, indicating that the model displays reasonable accuracy for the comprehensive range of ILs studied. However, significant discrepancies were found to be widespread in the literature data for IL heat of fusion, limiting the predictive power of the data-driven models and highlighting the need for improvements in measurement protocols.