The review of visual monitoring and recognition algorithm in laser welding

Abstract

There is abundant information in molten pool and keyhole during laser deep penetration welding. Their stability and characteristics are closely related to welding quality. At present, visual monitoring is an important means to obtain the characteristics of molten pool and keyhole. However, how to obtain a clear image, and how to optimize or develop an accurate and efficient algorithm are two major problems facing the visual monitoring of molten pool and keyhole. In view of the above problems, the paper combs the factors affecting molten pool visual monitoring and molten pool recognition algorithm. Firstly, the effects of camera type, auxiliary light source and filter, image acquisition method, the shielding gas and the air pressure on the image clarity of molten pool and keyhole are emphatically analyzed, and a method conducive to visual monitoring of molten pool and keyhole is proposed. Secondly, the influence of different algorithms and models on the monitoring accuracy of the system is analyzed. It is found that image pre-processing can highlight molten pool image information, and DBN (Deep Belief Networks) model has higher prediction accuracy through the processed molten pool image. Pointed out in the end of this article, the development of a laser welding monitoring device with clear image, stable and high dynamic response, the integrated development of multi-sensor synchronization visual monitoring device, and the development of laser welding process of adaptive fusion image processing algorithms will be the focus in the future, which provides a research direction for monitoring of molten pool and keyhole.