Neural Network Model of Weld Pool in Pulsed MIG Welding

Abstract

This paper deals with the problem concerning the modeling and sensing of the weld pool. In order to obtain the high quality of the welding result, it is important to measure and control the penetration depth in the robotic welding. Since it is difficult to directly measure the penetration depth and the relationship among the penetration depth, the welding current and other weld pool parameters such as the surface shape of the weld pool and the gap width, is non-liner, the weld pool model is constructed by neural network to estimate the penetration depth. The input and output variables of neural network model are determined from the information of the deferential equation for the penetration depth and the weld pool widths. Some steady state data and a transient response data of the fundamental experiment result are used as the training data of neural network model. The neural network model with less error can be learned by using these training data. When the experimental data of the weld pool widths and the gap width are input to the neural network model, this model is performed as the sensor to estimate the penetration depth. The validity of the neural network model was verified by the welding experiment.