Author ORCID Identifier
Semester
Spring
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
Fernando V. Lima
Committee Co-Chair
David S. Mebane
Committee Member
Yuhe Tian
Committee Member
Wenyuan Li
Committee Member
Qi Dong
Abstract
Systematic modeling, assessment, and optimization are essential for advancing emerging technologies and enhancing baseline process performance. In this work, different frameworks are proposed for hybrid physics–machine learning modeling, assessment, and optimization of chemical, energy conversion, and bioprocess systems. On the modeling side, the first proposed approach consists of a hybrid physics machine-learning framework. The method, named dynamic discrepancy reduced-order modeling (DD-ROM), balances discrepancies between reduced-order models (ROMs) and high-fidelity models (HFMs) using embedded Gaussian Processes (GPs), while providing significant computational gain. The framework is the first of its kind to offer comprehensive insight, guided by sensitivity and correlation analyses into where discrepancy terms should be introduced, as demonstrated by correcting dynamic mismatches between two distinct steam methane reforming (SMR) models. The results show the method’s ability to preserve model interpretability while achieving high predictive accuracy and improved computational efficiency. The second approach relates to the assessment of dynamic operations and presents a novel study on identifying control state changes in power systems using real closed-loop data. Specifically, Transfer Functions (TFs) and GPs within a Nonlinear AutoRegressive eXogenous model (GP-NARX) are employed. In the TF approach, deviations in the coefficients of transfer functions are used to indicate changes in the control states, while in the GP-NARX approach, changes are identified based on model uncertainties. Results obtained based on data from a power plant system showed that under specific conditions, both approaches are capable of identifying changes in the system’s control states, enabling informed recommendations for controller re-tuning and/or actuator re-ranging. The third framework consists of an integrated techno-economic and environmental analysis approach to assess emerging electrified chemical, energy conversion, and storage systems. The framework extends conventional cost assessment methods to evaluate capital and operating costs of large-scale emerging electrified technologies, providing profitability and sustainability metrics to quantify economic feasibility and environmental trade-offs. Multiple emerging electrified technologies are examined to demonstrate the framework, including: (i) low-temperature electrochemical CO2 reduction process; (ii) Hydrogen production via solid oxide electrolysis cells (SOEC); (iii) plastic depolymerization using Joule-heated graded-pore column reactors; and (iv) energy storage using Aluminum-CO2 batteries. Lastly, a dynamic optimization framework is proposed for enabling dynamic real-time economic optimization considering input disturbances and critical process constraints. Its effectiveness is demonstrated through the economic optimization of an industrial penicillin fermentation bioprocess, highlighting its capability to provide optimization-based decision-making for semi-batch operations.
Recommended Citation
Alves Nascimento, Claudemi, "Systematic Approaches for Hybrid Physics–Machine Learning Modeling, Assessment, and Optimization of Chemical, Energy Conversion, and Bioprocess Systems" (2026). Graduate Theses, Dissertations, and Problem Reports. 13293.
https://researchrepository.wvu.edu/etd/13293