Author ORCID Identifier

https://orcid.org/0009-0001-0446-7906

Semester

Summer

Date of Graduation

2026

Document Type

Dissertation (Campus Access)

Degree Type

PhD

College

Statler College of Engineering and Mineral Resources

Department

Chemical and Biomedical Engineering

Committee Chair

Yuhe Tian

Committee Member

Fernando V. Lima

Committee Member

Wenyuan Li

Committee Member

Brent Bishop

Committee Member

David S. Mebane

Abstract

Chemical and energy industries are undergoing a transition toward sustainable, flexible, and efficient operations to meet increasing energy demands while reducing environmental impact and ensuring safety. This transition occurs in the context of inherent uncertainty and variability, driven by factors such as raw material fluctuations and evolving market demands. Addressing these challenges requires innovative strategies that integrate process modeling and artificial intelligence/machine learning (AI/ML) methodologies to enhance operational performance, safety, and operability. This work develops hybrid modeling and optimization frameworks for the design and operations of chemical and energy systems under uncertainty, with a focus on improving flexibility, sustainability, and safety. Three main contributions are presented. First, a hybrid mechanistic/data-driven modeling framework is developed for Proton Exchange Membrane Water Electrolyzer (PEMWE) systems. A Physics-Informed Recurrent Neural Network (PI-RNN) is proposed for improved dynamic prediction of system behavior. The model incorporates physics-based loss functions and leverages both state variables and their time derivatives to ensure accuracy while preserving physical consistency. Second, an operability-based framework is introduced for process design, operations, and risk-aware optimization. This methodology quantifies how design and operational decisions affect process safety, enabling the identification of safe operating regions and providing comprehensive risk assessment across steady-state, open-loop, and closed-loop conditions. Third, a hybrid methodology combining multi-parametric programming and machine learning is developed to characterize feasible operating regions under uncertainty. Nonlinear constraints are reformulated through outer approximation, leading to explicit piecewise-affine representations of feasibility boundaries. These expressions are subsequently reformulated as Y-Wise Affine Neural Networks (YANNs) using rectified linear unit activation functions, allowing exact representation of the parametric solutions. For highly nonlinear or nonconvex systems, additional layers are incorporated and refined through adaptive boundary-focused sampling, enhancing accuracy near critical feasibility transitions. Overall, the proposed frameworks provide an integrated computational approach for modeling, design, and optimization of chemical and energy systems.

Download

Share

COinS