Quality control in die casting with neural networks

Die casting is an attractive manufacturing process for metal pieces of complicated shape which are produced in large quantities. In applications of high safety standards comprising parts exposed to high mechanical stress a 100% X-ray examination after production is required. In this paper it is shown that this expensive and time-consuming process can be replaced by employing a classifier based on an artificial neural net. All the process parameters considered as relevant for the quality are input to the net, which then calculates a quality index. The net is trained with a learning base of 120 items. Thereafter, the results obtained by means of a multilayer perceptron, a learning vector quantization and a dynamic learning vector quantization are compared. Our dynamic learning vector quantization, which represents an attractive new approach, is discussed in some detail.