Date of Award
2026-05-01
Degree Name
Doctor of Philosophy
Department
Metallurgical And Materials Engineering
Advisor(s)
David A. Roberson
Abstract
The martensite start temperature Ms governs heat treatment design and mechanical property development in martensitic steel alloys. Accurate prediction of Ms from chemical composition alone would accelerate alloy development, yet existing empirical equations lack accuracy for multi-component steels and purely data-driven models provide no physically principled behavior outside their training domain. This dissertation develops a hybrid machine learning-physics framework for predicting Ms in steel alloys from composition. The MAP Cambridge dataset (1,020 cleaned records, 15 alloying elements) is processed through a five-stage cleaning pipeline and a 21-feature physics augmented representation, including the Ghosh-Olson Ms thermodynamic prediction as an explicit feature. Kernel PCA with an RBF kernel reduces this to a 30-component latent space; the first two components correspond physically to a carbon/atomic radius axis (PC1, 20.0%) and a nickel/electronic structure axis (PC2, 14.0%), confirming that the latent space encodes the dominant thermodynamic mechanisms of martensite stabilization.
Language
en
Provenance
Received from ProQuest
Copyright Date
2026-05
File Size
262 p.
File Format
application/pdf
Rights Holder
Agniprava Banerjee
Recommended Citation
Banerjee, Agniprava, "Effective Prediction Of Martensite Start Temperature By Hybrid Machine Learning And Ethical A.I. In Materials Science" (2026). Open Access Theses & Dissertations. 4628.
https://scholarworks.utep.edu/open_etd/4628