A neural network approach for selection of powder metallurgy materials andprocess parameters

The artificial neural network (NN) methodology presented in this paper has been developed for selection of powder and process parameters for Powder Me tallurgy (PM) part manufacture. This methodology differs from the statistic al modelling of mechanical properties in that it is not necessary to make a ssumptions regarding the form of the functions relating input and output va riables. Employment of a NN approach allows specification of multiple input criterion, and generation of multiple output recommendations. The inputs c omprise the required mechanical properties for the PM material. The system employs this data within the NN in order to recommend suitable metal powder compositions and process settings. Comparison of predicted and experimenta l PM materials data has confirmed the accuracy of the NN approach, for pred icting the materials and process settings needed for attainment of required process outcomes. (C) 2000 Elsevier Science Ltd. All rights reserved.