Monitoring of laser welding process using PCA based 2-step sensor fusion Available to Purchase

The classification rate of laser welding process monitoring can be greatly increased by sensor fusion methods. But these methods frequently suffer from slow time varying behaviors; outliers due to inadequate training sets; and unreliability of some sensors. In this study, we have developed a 2-Step sensor fusion approach based on Principal Component Analysis (PCA). Power Spectral Density (PSD) of the signals from four sensors including infrared, ultraviolet, audible sound, and acoustic emission is chosen to be the feature source. PCA is conducted in both the sensor space and the frequency space, accompanied by a Class Mean Scatter (CMS) optimizing procedure. The method is applied to classify penetration qualities of laser welding. Results are very promising in the sense that 100% classification rates are achieved by using relatively small training set (10 ∼ 20 samples) and unreliable signals are automatically rejected. Given these initial results, we can reasonably expect that this approach will be highly suited for the adaptive monitoring after incorporating some recursive algorithms.