| Abstract Scope |
The declining availability of skilled welding operators has created a growing need for intelligent welding systems that can improve consistency, productivity, and reliability while reducing operator burden. Over the past several years, EPRI has led development of an AI-driven adaptive welding platform that integrates real-time sensor data, machine vision, weld parameter acquisition, and artificial intelligence algorithms to move mechanized arc welding toward autonomous operation.
This presentation will summarize adaptive welding development efforts performed by EPRI, including system architecture, welding data collection, algorithm maturation, and transition from laboratory demonstration toward deployable industrial solutions. In collaboration with Fraunhofer ILT, EPRI has evaluated convolutional neural networks and a modified self-supervised learning approach (DINOv2) to segment key weld image features such as the weld pool, wire, torch, and joint geometry with reduced manual labeling requirements. These capabilities support real-time weld monitoring, wire position control, reinforcement prediction, and closed-loop adjustment of welding parameters.
The presentation will highlight the broader technology maturation pathway, including integration with commercial welding equipment, expansion across multiple welding processes, including additive manufacturing applications, and ongoing efforts to support technology transfer for high-consequence nuclear and industrial welding applications. |