Dynamic Behavior Analysis and Control of Welding Pool Based on Machine Learning

The dynamic behavior of molten pool in welding process has a decisive influence on weld quality, while traditional control methods rely on empirical judgment and post-event detection, so it is difficult to realize real-time monitoring and adjustment. In this study, real-time monitoring, accurate prediction and intelligent control of molten pool dynamic behavior are realized by using machine learning algorithm. By extracting time domain, frequency domain, spatial domain and statistical features, and combining recursive feature elimination method and random forest model for feature selection, an LSTM-SVM hybrid model combining Long Short-Term Memory (LSTM) and Support Vector Machine (SVM) is constructed, which effectively processes high-dimensional and time-series correlation data in welding process. The experimental results show that the hybrid model has high accuracy, low average absolute error and high correlation coefficient in predicting the temperature, velocity and direction of molten pool, and the prediction accuracy is over 90%. This research not only improves the intelligence and automation level of welding process, but also provides strong support for the innovation and progress of welding technology.