Doctor of Philosophy (PhD)


Gordon A. and Mary Cain Department of Chemical Engineering

Document Type



The discovery of new materials like catalysts, polymeric films, and biomolecules, is driven by industrial needs such as improving reaction or separation selectivity, enhancing therapeutic effects on medical treatments, or reducing costs of replacement. However, deployment of these advances in industrial applications is often hindered by the lack of models needed for design and optimization. Due to the novelty of materials and devices, experimental data and first principles' knowledge are scarce, making it hard to build models either via data-driven or knowledge based approaches. In this context, a way to efficiently combine domain knowledge with data could provide a pathway to streamline the deployment of new discoveries in industrial settings. The notion of hybrid modeling provides a useful platform to address these issues. Transfer Learning (TL) is an extension of Machine Learning (ML) in which knowledge learned for a particular task can be leveraged to ease the training for a new task. In terms of modeling for electrochemical systems, models developed for a given device or material can be used to reduce the amount of data needed to develop models for new materials, devices, or configurations. When the data for the initial training stage comes from a simulation, domain knowledge can be easily incorporated into the data-driven model, while at the same time reducing the number of experiments to be conducted. This approach can be used to generate surrogate models that approximate the real behavior of the systems with adequate accuracy at a reasonable cost. This dissertation introduces a methodology for the incorporation of Machine Learning-based surrogate models and TL into hybrid modeling with applications in electrochemical processes.



Committee Chair

Romagnoli, Jose A.



Available for download on Monday, February 23, 2026