Deep-learning-based porosity monitoring of laser welding process

Abstract

A porosity monitoring scheme for laser welding process was developed based on a deep learning approach. The in-process weld-pool data were sensed with a coaxial high-speed camera and labelled with the porosity attributes measured from welded specimens. A convolutional neural network (CNN) model with compact architecture was designed to learn weld-pool patterns to predict porosity. In laser welding experiments of 6061 Aluminum alloy, the CNN-based monitoring model achieved a classification accuracy of 96.1% for porosity occurrence detection, though the prediction of micro (less than 100 µm) and deep subsurface pores still remains challenging.