Neural network and fuzzy control of weld pool with welding robot

The sensing and control of weld pool depth in robotic welding are discussed. A neural network-based method for measuring the depth is proposed, since the depth cannot be directly measured in real time. The weld pool depth is estimated by using the information obtained from the welding side. The surface shape of the weld pool and the width of the groove gap can be measured during the welding. The weld pool depth can also be measured after the welding. Training data were constructed from these numerical data. When the width of the groove gap changes, the weld pool depth changes too. The feedforward control system for the variation of the groove gap width just under the electrode can be constructed by observing the groove gap width before the electrode. The feedback control system was constructed in order to keep the output of the neural network constant. The fuzzy control system was constructed from the feedback control part and the feedforward control part. The validity of a neuro-fuzzy controller was verified by welding experiments.<<ETX>>