Keyhole morphology monitoring of laser welding based on image processing and principal component analysis

The keyhole is a specific phenomenon produced by the intense interaction between laser and material. Keyhole morphology can reflect welding stability and welding quality to a certain extent. Nowadays, the keyhole is observed and image processed by a high-speed camera and related algorithms, respectively. However, the binarization threshold is fixed in keyhole extraction, and conventional binarization methods are not adaptive. This will affect the feature extraction of keyhole morphology. In this paper, a dynamic threshold adjustment method is proposed, which can combine the quick positioning of the Otsu method and the weight balance of the average method. Furthermore, seven defined features of the keyhole region are divided into dynamic parameters and shape parameters. The dimension of these parameters is reduced by principal component analysis (PCA). The first three PCs occupy more than 92%, which covers most of the keyhole information. At last, the influence of dynamic parameters and shape parameters on keyhole morphology is presented. This research plays a positive role in monitoring the keyhole morphology of laser welding.