In-situ monitoring of the penetration status of keyhole laser welding by using a support vector machine with interaction time conditioned keyhole behaviors

Abstract

Laser welding is a complex manufacturing process for which diagnosis and control are essential to ensuring quality control regardless of any disturbances to the welding process. In this study, the interaction time conditioned keyhole behaviors (ITCKBs) were proposed and analyzed for different penetration status, and an innovative method was developed that uses ITCKBs as the input of a support vector machines (SVM) to realize in-situ monitoring of the weld penetration status independent of process parameters. Four different ITCKBs were extracted to guarantee accurate estimation at different penetration status: the average area of the keyhole, average area of the full penetration hole, frequency of the full penetration hole, and maximum area of the full penetration hole. Penetration statuses were classified with different combinatorial inputs for comparison. Compared with using the keyhole area, full penetration hole area, or keyhole area and full penetration hole area as the input, using the four ITCKBs as the input improved the classification accuracy of the SVM by 39.1%, 33.3%, and 24.1%, respectively. This improvement in performance was because the point clusters of the ITCKBs are linearly separable. The results also indicated the presence of a positive synergistic effect because the classification accuracy of the SVM improved from 63.5% to 96.5% as the number of ITCKBs increased.