Yue Cao

To improve human control of welding process, a teleoperated Human Robot Collaboration welding system is developed, where a generative AI model can predict the future weld pool evolution, to reduce human cognitive burden. This work exemplifies a human-centric HRC paradigm that unites human expertise, robotic execution, and computational intelligence for intelligent manufacturing.

Double-electrode Gas Metal Arc Welding (DE-GMAW) is a variant of GMAW that decouples heat and mass transfers, well suitable for additive manufacturing (AM). I developed a VR-based teleoperated Human Robot Collaboration system to robotize it by leveraging human adaptability. An object detection network (YOLO) was implemented and a human model was established. Ongoing studies include transferring human operational knowledge using Imitation Learning.

In additive manufacturing (AM), Labels are typically assigned at the segment level during post-process inspection, giving identical labels to all frames in a segment even though anomalies may occur only in part of it, leading to inevitable misclassifications. In this work, an iterative label refinement strategy is proposed, using Variational Autoencoder and Gaussian Mixture Models, to learn unbiased feature for anomaly detection.

Achieve adaptive control of weld penetration in pulsed GMAW, using adaptive Model Predictive Control, nonlinear identification, and Kalman Filtering, with characteristic arc voltage signals as feedback.