Bridging Physics and AI: Large Language Models for Welding Defect Prediction and Fracture Analysis
- Sponsor
- Industrial and Enterprise Systems Engineering, Dept. Head office
- Speaker
- Professor Jingjing Li
- Contact
- BuuLinh Quach
- bquach@illinois.edu
- Phone
- 217-265-5220
- Originating Calendar
- ISE Seminar Calendar
*Presentation will be recorded.
Abstract:
This talk shows how large language models (LLMs) can accelerate discovery in welding research, especially when data are limited. Humping defects in high-speed laser welding for fuel cell manufacturing were first studied using in situ synchrotron X-ray imaging and computational fluid dynamics, revealing how keyhole shape, backward melt flow, and molten pool geometry drive defect formation. Building on these insights, an LLM-based framework, T2EGPT, was developed to extract process relationships from the literature and generate interpretable dimensionless equations for predicting humping onset from sparse experimental and simulation data. Compared with conventional data-driven models, T2EGPT achieved higher predictive accuracy and stronger transferability across materials. These findings also guided the design of a dual-beam laser welding strategy that suppressed humping at speeds above 90 m/min, reaching 4.5 times the current welding speed. The talk also presents an LLM-driven framework for autonomous fracture analysis in dissimilar aluminum-steel resistance spot weld micropillars. By combining image classification, retrieval-augmented generation, and welding knowledge, the framework explained crack initiation, crack growth, and stress-drop behavior under different corrosion conditions. These results highlight LLMs as an exciting new tool for physics-informed, scalable discovery in welding and beyond.
Bio:
Jingjing Li is a Professor in the Department of Industrial and Manufacturing Engineering at Penn State University, University Park. She holds a Ph.D. in Mechanical Engineering and an M.A. in Statistics from the University of Michigan, Ann Arbor, along with M.S. and B.S. degrees in Materials Science from Tsinghua and Beihang Universities, China. Her research advances the science and design of welding and metal additive manufacturing through in-situ characterization, process physics, and microstructure-performance relationships, with a growing focus on AI-driven approaches to enable next-generation smart manufacturing. She is a Senior Editor for Journal of Materials Processing Technology and has received several prestigious awards, such as the ASME Chao and Trigger Young Manufacturing Engineer Award and the NSF CAREER award. She was also elected as an ASME fellow.