No abstract

In the manufacturing industry, quality degradation due to a decrease in skilled operators possessing domain knowledge has become a problem. Digitalization using IoT technology has emerged as a means to tackle this problem. In the resistance welding process, welding quality fluctuates depending on aging of electrodes. Experienced operators adjust the welding voltage to keep the quality constant. As this knowledge is difficult to share, the success rate of voltage changes at the time of quality degradation tends not to be improved. Therefore, we developed an automatic voltage determination system that improves both quality and productivity by improving the success rate, which is one of the main measures. The system learns past sensor data and voltage change logs, determines the voltage according to input real-time sensor data, and sets the voltage for the welding machine. We propose three voltage determination methods: a similarity search method, a voltage prediction method using a regression model that outputs voltage, and a quality prediction and voltage search method that searches for the optimum voltage in a classification model to predict the success or failure of voltage changes. Our evaluation of these methods shows that the success rate improves by up to 12.4 percentage points compared to when the operators performed the process manually. This result demonstrates that we can achieve quality stabilization and productivity improvement by implementing our system in the welding process.

No abstract

No abstract

No abstract