Date of Award
August 2024
Document Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
Department
Chemical Engineering
First Advisor
Ali Alshami
Abstract
Experiments are generally time-consuming, expensive, and rely on exhaustive planning, analysis, and testing efforts. The limitations of the traditional experimental methods in materials development can be overcome by the use of machine learning (ML) algorithms to guide and assess the discovery and fabrication processes. This dissertation involves developing different ML models, focusing on their utilization in three distinct, yet overlapping areas: Prediction of organic compound aqueous solubility, guiding separation membranes fabrication and predicting scale formation in commercially produced water systems.Prediction of organic compound aqueous solubility: A reliable and practical determination of a chemical species’ solubility in aqueous systems continues to be examined using empirical observations and exhaustive experimental studies alone. Predictions of the solubility using data-driven algorithms can allows creating a rationally designed, efficient, and cost-effective tool for the next-generation materials and chemical formulations. Molecular-descriptors, the most used method in previous studies, and Morgan fingerprint, a circular-based hash of the molecules' structures, were used to predict the water solubility of various organic compounds. Furthermore, the most effective features for the aqueous solubility quantification were reported using the Shapley Additive exPlanations (SHAP) and thermodynamic analysis. Guiding separation membranes fabrication: Fabrication and/or modifying separation membranes is a time-consuming process and requires numerous experiments and analysis to discover new poly/monomers building blocks. Additionally, there is currently no commercially available computational software available to predict the behavior of a mono/polymer and calculate their diffusional properties. To address this challenge, we explored the potential of using ML to graft the polyamide (PA) surface of a reverse osmosis (RO) membrane to increase the water flux, and overcome the limitations of the permeability-selectivity tradeoff. The moieties with positive and negative contributions toward water permeability were identified using SHAP analysis. We attempted to improve the subunits of the PA’s structure with positive Shapley values and graft the polyamide RO membrane layer of a commercial membrane, Dupont XLE, resulting in a substantial increase in water flux. Predicting scale formation in produced water systems: Predicting scale formation under real-world condition is a significant challenge due to continuous variation in salt concentrations affecting inorganic scale composition. This study aimed to identify which type of scale would form at different ion concentrations and pH levels, and interpreted the impact of each feature on scale formation using ML methods. The ML methods used could uncover hidden patterns in the data in order to make a scale formation predictions and decisions by shifting from thermodynamic models to data-driven, black box prediction models. We trained the saturation index of potential mineral scales in PW samples under 60°F and 60 psi conditions, using three different ML techniques on a database comprising 2313 PW’s quality data points from different locations in Bakken Shale Formation, including ionic compositions, and pH. We identified significant features influencing scale formation by applying SHAP analysis to our model. We deployed this model on a publicity accessible website, at https://alshamiresearch.com/ where users who may have limited knowledge of thermodynamics, kinetics, computer programming, and numerical analysis can seamlessly use it for their specific experiments.
Recommended Citation
Tayyebi, Arash, "Machine Learning-Assisted Materials Development: Molecular Solubility Prediction, Polymeric Membrane Fabrication, And Scale Formation" (2024). Theses and Dissertations. 6458.
https://commons.und.edu/theses/6458