Key Image Processing Techniques Used in Laser Welding

Welding image and weld pool edge Machine vision techniques are widely used with automatic laser-welding robots. Existing systems acquire geometrical information on both the weld seams and molten pool by using a single camera and then processing a single image. The recognition of weld seams is thus obviously affected by the arc, plasma, spatter, etc. This study examined a novel, cost-effective weld image acquisition and processing system with real-time performance, which is based on an advanced RISC machine (ARM). The software and hardware platforms of a weld seam image processing system were designed, and theoretical and experimental studies were undertaken. A novel image-processing method for the laser welding molten pool, based on a quaternion and self-organizing feature map (SOFM) neural network is also proposed. The edge characteristic vector of the molten pool is acquired based on the rotational characteristic of the quaternion in a three-dimensional vector space. The geometric features of the molten pool image are acquired by the self-organizing feature map neural network. The measured results pointed to the possibility of improving the reliability and precision of the automatic laser welding system by combining the two machine vision systems.