Abstract
An example of application of neural network is introduced. The backpropagation (BP) model is applied to the weld bead visual inspection. The bead shape data are picked up from the actual weld bead and divided into three categories. The weight values of the connections which joint the units in the different layer of the neural network are formed through the learning process made by using these data for each category. As the result of the test, it is shown that the discrimination between the sound bead and the defect bead can be done properly if the selection of the network parameters and the learning process are made suitably.
