| Abstract Scope |
This study introduces an extensible machine learning (ML) framework to enhance resistance spot welding quality across the welding lifecycle. By integrating multi-stream data, including process parameters, material stackups, and in-situ signals, the framework establishes quantitative relationships between materials, processes, and joint performance. A key innovation is the framework's extensibility, which enables the predictive design of process windows for new, untested material stackups. Meanwhile, the model analyzes real-time signal patterns and anomalies to correlate process signatures with weld quality metrics. This flexible architecture supports diverse material stackups, including advanced high-strength steels, aluminum alloys, and dissimilar combinations, ensuring consistent performance across material stacks. Results demonstrate that this integrated approach effectively improves the weld process design and automated monitoring, offering a scalable, data-driven pathway for high-performance welding in high-volume manufacturing. |