Monitoring of laser weld penetration using sensor fusion

Abstract

Depth of penetration is a critical parameter in laser welding. In many applications, full penetration is desired, but difficult to detect in real time. A sensor fusion system using infrared, ultraviolet, audible sound, and acoustic emission has been implemented for real time monitoring of CO2 laser lap welds in both laboratory and industrial production settings. Signals from the welds were analyzed by: (1) singular value decomposition with data fusion, (2) class mean scatter with decision fusion, (3) class mean scatter with feature fusion, and (4) singular value decomposition with decision fusion using minimum distance and quadratic classification. A classification rate of 100% was obtained for detection of full penetration for both laboratory and production settings.