Neural network modelling and system simulation for dynamic welding process during pulsed GTAW based on vision sensing

摘要

In this paper, the geometry information of weld pool of pulsed GTAW was acquired by vision sensing. A neural network dynamic model was set up to reflect the relations between topside geometry features and backside geometry features of weld pool during pulsed GTAW. The model was used to predict backside dimension of weld pool. The inputs of the model were topside geometry dimensions and welding parameters. In order to test the validity of the model, the simulating experiment combined with PID controller was carried, result showed the neural network model could be used in the intelligent control system for backside weld pool shape.