Monitoring of resistance spot welding expulsion based on machine learning
Lei Zhou, Tianyi Zhang, Zhongdian Zhang, Zhenglong Lei, Shi-Liang Zhu
Abstract
There is still a lack of effective monitoring methods for resistance spot welding expulsion in manufacturing sites. In this study, more than 63,766 dynamic resistance curves of welding points were collected in manufacturing sites, and the machine learning method is applied to monitor the resistance spot welding expulsion. Generally, the model's generalisation ability is poor due to the complex conditions of the manufacturing site. This study solves this problem by improving the data preprocessing and model selection method. The experimental results show that when the process parameters do not change, the accuracy of expulsion recognition can easily reach 95%. When the process parameters change, it can reach 90% by selecting the appropriate data preprocessing method and model selection.