About this Abstract |
Meeting |
Advances in Welding and Additive Manufacturing Research 2022
|
Symposium
|
Advances in Welding and Additive Manufacturing Research 2022
|
Presentation Title |
Defect Detection and Control through Multi-source Visual Sensing and Deep Learning |
Author(s) |
Zongyao Chen, Zhili Feng, Jian Chen, Yunhe Feng, Dali Wang |
On-Site Speaker (Planned) |
Zongyao Chen |
Abstract Scope |
An intelligent welding control system has been developed to monitor and control weld penetration in GTAW process in real time. The system consists of following major individual parts, and integration thereof to realize the goal of real-time intelligent control. First, both passive vision and active vision sensing methods are combined into one system to acquire multiple types of weld pool images, to overcome the interference from the intense arc light. New algorithms are developed to determine weld pool key 3D geometric features based on the two-dimensional passive vision weld pool images and the image of reversed electrodes. Second, machine learning was applied to correlate weld pool penetration to the key features extracted from vision images in welding. The accuracy of prediction was verified through additional welding experiments. Decision making algorithms derived from deep machine learning were implemented with a personal computer, to monitor the weld pool status from multi source images, and to command necessary adjustments to welding parameters in a close loop welding control system, to realize real time welding process control for full penetration during welding. The system was successfully tested on bead on plate welding of stainless steel. |
Proceedings Inclusion? |
Undecided |