Archived

This content is available here strictly for research, reference, and/or recordkeeping and as such it may not be fully accessible. If you work or study at University of Kentucky and would like to request an accessible version, please use the SensusAccess Document Converter.

Author ORCID Identifier

https://orcid.org/0009-0000-8423-3466

Date Available

5-27-2026

Year of Publication

2026

Document Type

Doctoral Dissertation

Degree Name

Doctor of Philosophy (PhD)

College

Engineering

Department/School/Program

Electrical and Computer Engineering

Faculty

YuMing Zhang

Faculty

Daniel Lau

Abstract

Arc welding processes demand real-time adaptive control that current robotic systems cannot achieve autonomously. This dissertation develops a systematic framework to robotize complex welding by learning from human demonstration, integrating generative modeling, physics-informed reconstruction, and model-based imitation learning. First, human--robot collaboration systems are established for both Gas Tungsten Arc Welding (GTAW) and Double-Electrode Gas Metal Arc Welding, combining robotic teleoperation with virtual reality interfaces to capture high-quality operator demonstrations. Second, a physics-informed neural network framework reconstructs complete molten pool flow fields from high-speed imaging, enriching process understanding beyond direct sensor observation. Third, generative models, including a hybrid latent variational autoencoder for predictive weld pool visualization and a dynamic variational autoencoder with particle filtering for robust state estimation, learn compact, temporally coherent representations of welding dynamics that enhance both human control and machine-interpretable process states. Fourth, model-based imitation learning, combining adversarial imitation with neural ordinary differential equation dynamics, extracts and refines human expertise for autonomous GTAW control. Fifth, a bidirectional multi-rate control architecture extends the paradigm to balance-critical humanoid platforms for precision manufacturing. Together, these contributions establish a transferable pipeline from human demonstration capture to autonomous robotic welding execution.

Digital Object Identifier (DOI)

https://doi.org/10.13023/etd.2026.19

Archival?

Archival

Funding Information

This research was supported by:

  1. National Science Foundation, "Optimal Transport Generative Adversarial Networks: Theory, Algorithms, and Applications" (Award No. 2327113), 2023–2026.
  2. National Science Foundation, "Intelligent Co-robots for Complex Welding Manufacturing through Learning and Generalization of Welders' Capabilities" (Award No. 2024614), 2020–2025.

Available for download on Wednesday, May 27, 2026

Share

COinS