Vision-based Sensing Of The Weld Pool Penetration In PAW

PAW (plasma arc welding) is a high efficiency and advanced welding methods. It has the feature of high energy density, less welding distortion and high efficiency. It is widely used in heavy machinery, pressure vessels, aerospace and other industries. Vision-based sensing of the weld pool penetration is a fundamental work for weld joint quality control and it can also provide fundamental data for numerical simulation of keyhole plasma arc welding.In this paper, based on the experimental system of plasma arc welding, a synchronous vision system with three CCD cameras is developed for observing the weld pool and the keyhole. The filter group is designed to weaken the interference from arc light and efflux plasma. The proper viewing angle, exposure time and the focal length of the lens are selected, and clear images are captured.Bead-on-plate welding tests are conducted on stainless steel under different welding conditions. Clear images are obtained and the CCD cameras are calibrated based on the pinhole camera model.Lots of images are analyzed and multi-parameter constraints algorithm is designed to extract the edge of the weld pool and the keyhole. According to the calibration coefficients, the geometrical parameters of weld pool and keyhole are obtained.Based on the experimental results, the relationship among the welding parameters and the geometrical parameters of the weld pool and the keyhole is studied, and a proper BP neural network forecasting model is developed. This paper lays a solid foundation for welding quality control system.