Improving the predictive response using ensemble empirical mode decomposition based soft sensors with auto encoder deep neural network

Abstract

The keyhole instability is a key concern in laser deep-penetration welding of high reflectivity materials, potentially impacting the penetration status and weld quality. Monitoring and control the keyhole behavior still remain a great challenge for obtaining a desired welded joint. For the pulsed laser welding of thin-sheet aluminum alloy, an active visual monitoring system was established to systematically probe the dynamic keyhole behavior from multi-view sensing. Combining with the image processing method and process analysis, the keyhole surface area and depth were extracted to quantify the keyhole formation dynamics under different welding conditions. Furthermore, a data-driven deep learning model with hyperparameter optimization was constructed to identify different penetration states and it has a high accuracy and good reliability. The experiment results show that our proposed measurement scheme based on multi-view monitoring and deep learning approach could guide the development of real-time control of the pulsed laser welding process.