Theses and Dissertations
Date of Award
7-1-2026
Document Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
Department
Mathematics
First Advisor
Jianzhi Li
Second Advisor
Hansapani Rodrigo
Third Advisor
Aldair E. Gongora
Abstract
Advances in materials science, process engineering, and artificial intelligence (AI) are creating new opportunities to automate the discovery of novel materials, at speeds far beyond traditional experimental approaches. Fully realizing this potential requires a principled pathway for selecting, validating, and deploying machine learning frameworks within physically constrained design spaces that are too large to sample exhaustively. This dissertation establishes a four-step framework, Represent, Learn, Evaluate, and Decide, for using AI as a bounded instrument of scientific discovery, demonstrated through three contributions. First, a deep feedforward neural network trained with a domain-aligned Kullback–Leibler divergence loss on sparse equation-of-state tables learns the full multiphase equilibrium structure of a binary alloy system, achieving 98.75% accuracy in thermodynamic phase prediction across unseen compositions. Second, a multimodal automation framework, yielded from benchmarking five natural language processing strategies and six object detection architectures on real-world data, fuses real-time speech recognition and visual scene classification, executing and validating robotic experimental workflows that enable human-on-the-loop self-driving laboratories. Third, ARTEMIS (Autonomous Robotic Toolpathing, Error Monitoring, and Intelligent Self-correction) establishes a roadmap for AI-powered robotic fused deposition modeling (FDM). To that aim, frontier and efficient large language models from OpenAI and Anthropic families are benchmarked by their ability to select context-aware corrective actions across five FDM defect classes. Collectively, these contributions establish AI, constrained to the physically validated actions a domain permits, as a precise scientific instrument, thus forging a path towards increasingly autonomous research cycles and manufacturing processes.
Recommended Citation
Hallas, K. L. (2026). From Equations to Discovery: AI-Driven Parameter Space Navigation in Materials Science and Process Engineering [Doctoral dissertation, The University of Texas Rio Grande Valley]. ScholarWorks @ UTRGV. https://scholarworks.utrgv.edu/etd/2027

Comments
Copyright 2026 Kristen L. Hallas. All Rights Reserved. https://proquest.com/docview/3409162495