Ranked Feature-Based Laser Material Processing Monitoring and Defect Diagnosis Using k-NN and SVM

In this study, an in-situ monitoring system was developed using a spectrometer for laser welding on galvanized steel. The spectrometer monitored the emission spectra generated from laser-induced plasma for the purpose of classifying welding defects because the plasma generated during the laser welding process contains a considerable amount of information about the on-going process. Temporal features extracted from the emission spectra were used for in-situ monitoring. In order to extract the best features, Fisher's criterion was adopted to rank and select the features. The monitoring performance of a photodiode and spectrometer were compared by using the selected features. The emission spectrum proved to be a better feature than the photodiode signal. Additionally, the emission spectrum was combined with statistical properties such as mean, root mean square, standard deviation, peak, skewness, and kurtosis to increase the classification rate. The ranking of the emission spectra depended on the statistical features. The k-nearest neighbors (k-NN) algorithm and support vector machine (SVM) algorithm were used as classifiers of the ranked features. Three groups, which are sound welding, underfill defect and bead separation defect, were successfully classified for quality assurance.