Search

Various welding features during high-power disk laser welding are extracted from the sensor optical signals. The association rules between the features and welding state parameters are considered as significant information for the welding state recognition. In this study, a multisensor detecting system for high-power disk laser welding was designed to obtain multisignals, including AI-sensing camera, UVV-sensing camera, X-ray TV system, visible light–sensing photodiode, reflected laser–sensing photodiode, and spectrometer. Forty-two original features were extracted from six sensor optical signals. The weld width and relative penetration were obtained as welding state parameters. Association analysis was conducted to select 16 pivotal features most closely associated with welding state parameters and provide quantitative analysis for the association rules between pivotal features and welding state parameters. By carrying out a series of experiments under different laser welding conditions, the relationship among features, welding state, and welding condition was revealed through the association rules. A molding weld was generated in different states, such as stabilization, hump, dimple, and blowout. To recognize the welding state, the Bayesian network imperfection-detection model was set up based on pivotal features. TAN Bayesian network model is better than Markov Bayesian network model, and the prediction accuracy can reach 86.04% and is 10.92% higher than Markov model. Experimental results show that the proposed method can be applied to recognizing high-power disk laser welding state in real time.