A study on the horizontal fillet welding using neural networks

Generally, weld bead shape is a serious factor in falling-off in weld quality among various kinds of weld defect. In GMAW, weld bead shape affects a number of welding parameters including; welding current, voltage, speed and weaving length. To detect weld bead shape, we select proper welding parameters. We have difficulty analyzing the relationship between weld bead shape and welding parameters due to non-linearity of the welding process. In the case of an arc sensor, though it has a signal processing problem, it is still a widely-used method in industry because it is low cost and easily automated. Proper welding parameters were selected and a weld bead shape detecting system was proposed, using neural networks which were able to identify the relation between weld bead shape and the welding parameters. Also, in the neural controller, the time delay neural network (TDNN) was used in this proposed neural network, due to non-linearity of the welding process. Besides, in welding quality testing, it is important to analyze the weld bead shape. Proper welding parameters and fifteen points that represent weld bead shape were selected, and a real-time monitoring system of weld bead shape was proposed to find the effects of various welding parameters and estimate weld quality using neural networks.