Seminar on Deep Learning Based Monitoring of Weld Penetration: Where we are and what are the needed revolutionary solutions?

Dr YuMing Zhang

James R Boyd Professor of Electrical Engineering

 University of Kentucky, USA

This seminar will be chaired by A/P Zhou Wei.

Weld penetration should be monitored during manufacturing. Unfortunately, it occurs underneath workpieces and is not considered directly observable so that its real-time in-situ monitoring is challenging. In the last half century, researchers have focused on finding promising real-time observable phenomena and correlating such phenomena to penetration. This has been difficult as it is unclear what are critical in observed phenomena and trial-and-error iterative processes are practiced. Such iterative process is not automated, finding/fitting right features/model takes months if not longer, and success is not assured. Deep learning automates and combines featuring and fitting to maximally use the raw information directly to achieve highest accuracies. Computation is drastically increased, but the iterative process is replaced by an automated one so that the time frame is still drastically reduced. This talk will review and analyze (1) various raw information that has been used as the observed phenomena to input into deep learning models and analyzes why they may correlate to penetration; and (2) various deep learning models and analyzes why/how they may and are needed to correlate different raw information to penetration. It will also briefly review major techniques that have been used to train deep learning models in penetration prediction. It will then identify major achievements for efforts taken so far. Finally, we identify two fundamental issues that require revolutionary solutions in order to move the deep learning technologies from laboratory studies to manufacturing as directions for future efforts. These two issues are analyzed and some preliminary solution directions are proposed.