WPD-PCA-Based Laser Welding Process Monitoring and Defects Diagnosis by Using FNN and SVM

Abstract

Recent years have seen increasing attention paid to laser welding monitoring. This paper introduces an innovative approach to perform laser welding process monitoring and welded defect diagnosis. The laboratory-scale sensor can be replaced with industrial-scale sensors after the data-driven model has been established by applying multivariate statistics and machine learning methods. In addition, industrial-scale sensor makes effective diagnosis of welded defect by using pattern recognition. Experimental results show that the feature vector affecting estimation and classification accuracy can be obtained by using wavelet packet decomposition principal component analysis. Image processing technology was applied to quantify geometrical parameters of welding process. The feedforward neural network prediction model and the support vector machine classification model built in this research help to guarantee accurate estimation on welding status and effective identification of welded defect. The method proposed by this paper provides an innovative data-driven-based approach for laser welding process monitoring and defects diagnosis.