Laser Welding Quality Monitoring Via Graph Support Vector Machine With Data Adaptive Kernel
Written by Sergey A. Shevchik, Farzad Vakili Farahani, Neige Faivre, Bastian Meylan, Tri Le-Quang, Silvio Zanoli, and

This contribution presents a new approach for in situ and real-time laser quality monitoring. This new approach uses a novel combination of the state-of-the-art sensors and machine learning for data processing. The investigations were carried out using laser welding of titanium workpieces. The signals from the laser back reflection, optical and acoustic emissions were recorded during the laser welding process and were decomposed with M-band wavelets. The relative energies of narrow frequency bands were taken as descriptive features. The correlation of the extracted features and the laser welding quality was carried out using Laplacian graph support vector machine classifier. Also, an adaptive kernel for the classifier was developed to improve the analysis of the complex features distributions and was constructed from Gaussian mixtures. The presented laser welding setup and developed adaptive kernel algorithm were able to classify the quality for every 2 μm of the welded joint with an accuracy ranged between 85.9- 99.9 %.

 

Reference
S.A. Shevchik, T. Le-Quang, F. Vakili-Farahani, F. Neige, B. Meylan, S. Zanoli, and K. Wasmer, Laser Welding Quality Monitoring Via Graph Support Vector Machine With Data Adaptive Kernel, IEEE Access, Vol. 7, Issue 1, pp: 93108 - 93122, 2019, doi: https://doi.org/10.1109/ACCESS.2019.2927661

Example typical movie typical of classification result during a laser welding experiment using AE and our algorithm, in a titanium plate.