Development and Commercialization of Adaptive Feedback Welding Technology for Fabrication and Repair Applications

The mechanized Gas Tungsten Arc Welding (GTAW) process has gradually evolved into the primary method for nuclear component fabrication and repair. While mechanized improvements have boosted parameter control, the need for a qualified welder to supervise operations persists. Recent advancements in monitoring and automation technologies have made the shift toward fully automated machine arc welding more feasible, reducing the necessity for continuous human oversight. This transition offers a promising route to cut costs in the energy sector, particularly in nuclear plants, and mitigate the shortage of skilled welders.

The mechanized GTAW system is equipped with comprehensive encoder feedback incorporating a camera, microphone, and laser profilometer. A distinct software interface has been developed to seamlessly integrate with all sensors and power supply on the weld head. Notably, the system boasts an automated homing sequence that precisely aligns the electrode within the groove and employs an automated wire dripping detection control loop utilizing acoustics. Furthermore, an image-based neural network, trained on real-time monitoring of welding parameters, identifies crucial features such as the weld pool, groove, wire, and electrode. A closed-loop control ensures a consistent wire position in the weld pool, compensating for wire cast and other unforeseen disturbances.

A separate neural network, trained on weld telemetry settings, electrode positioning, groove pre-scans, and post-scan analysis of weld beads, accurately predicts the weld bead profile for multi-pass welds and inconsistent groove geometries. Integration of both open and closed control loops enables autonomous planning and execution of multi-pass welds to fill grooves.

A viable trajectory towards the commercialization of Adaptive Feedback has emerged through collaboration with a third-party entity. This collaboration aims to integrate the neural networks into remote weld monitoring software, introducing an AI vision package to the public domain. Successful implementation holds the potential for incorporating additional adaptive feedback capabilities in the future.