Vision-based keyhole detection in laser full penetration welding process

Abstract

In a laser full penetration welding process, keyhole behavior affects the welding quality most. Due to the variation of wall thickness, constant welding speed and power will lead to partial penetration or over heat. So, effective monitoring and control method is needed to improve the quality of laser full penetration welding. In this paper, a vision based full penetration hole detection system is studied. The effects of exposure time, zooming lens, and optical filters on the quality of coaxial images are investigated. Among these factors, narrow bandwidth filters affect coaxial images most, because according to Plank law, workpiece under different temperature will emit light of different wavelengths, and the images under different narrow bandwidth filters reflect the temperature distribution of workpiece. The full penetration hole detection algorithm is developed based on region grow algorithm. Images of different welding speed and power are captured, and the relationship of the diameter of full penetration hole with welding speed and laser power is studied. The detection algorithm can identify partial penetration welding and laser cutting. It can be used to closed loop control of laser welding in future.