Stochastic analysis of multispectral data during laser beam welding of stainless steel with different gap sizes

Abstract As industrial automation advances, effective process monitoring methods are crucial. This paper investigates laser beam welding LBW of ferritic stainless steel 1.4003 at butt joints with varying gap widths using a multispectral sensor. By analysing z-standardized stochastic signals and associated statistical figures, the percentage error rate for classifying weld seams as »OK« or »not OK« is evaluated, differentiating between visible VIS and near-infrared NIR spectra. Over 1,500 linear discriminant analyses are performed, showing that classification capability improves with more stochastic features (boosting), particularly scattering parameters for identifying »not OK« welds. Targeted data analysis reduces data load and enhances classification efficiency, with different wavelengths showing varying effectiveness in a priori prediction. The choice of statistical figures impacts the selection and number of wavelengths for process monitoring.