Joining Process Monitoring Using ML-Based Joint Quality Estimation

The system performs preliminary action by capturing process data (current, voltage, time signals) during the joining process itself, before the actual quality assessment is needed. This preliminary data collection enables subsequent machine learning analysis to predict quality outcomes without requiring separate inspection steps, thus maintaining high reliability while improving efficiency.

The patent replaces the mechanical/manual ultrasonic inspection system with an automated machine learning-based evaluation system. The machine learning model processes electrical process signals (current, voltage, time) to predict joint quality, substituting the need for physical ultrasonic testing devices and manual inspection operations, thereby achieving both high reliability and improved productivity.

The patent replaces time-consuming physical inspection methods with automated machine learning analysis of process signals. The system evaluates joint quality by analyzing electrical signals already generated during the joining process, eliminating the need for separate inspection operations and achieving comprehensive quality control without additional time loss.

The system maintains continuity of useful action by performing quality evaluation continuously during the joining process itself, rather than interrupting production for separate inspection cycles. The machine learning model processes data in real-time as joints are created, enabling continuous quality monitoring that does not halt or slow down the manufacturing process.

The patent introduces an intermediary approach by using actual process signals from the joining machine as training data, rather than relying solely on complex simulation data. The machine learning model is trained on real electrical signals (current, voltage, time) captured during actual joining operations, with quality labels provided by subsequent ultrasonic inspection, creating a practical training pathway that balances accuracy with reduced complexity.

The system implements feedback by using the results of ultrasonic inspection to label the quality of joints, which then serves as ground truth for training the machine learning model. This feedback loop allows the model to learn from actual outcomes and continuously improve its predictive accuracy without requiring complex simulation frameworks, as the real inspection results provide direct training signals.