Detection of causes of casting defects assisted by artificial neural networks

Abstract

Defects in castings often appear unexpectedly and it is difficult to identify their source as they can be brought about by a large number of randomly changing production parameters. Artificial neural networks were used for detection of the causes of gas porosity defects in steel castings. The applied procedure included systematic storing of two types of information: about the process parameters, materials used and even employees involved in the production (as the network inputs) and about the appearance of a given defect (as the network output). The trained network was able to detect interdependences among various factors influencing water vapour pressure in the mould and thus to indicate the most probable cause of porosity.