Developing cost models by advanced modelling technology

Abstract

The aim of this paper is to examine the use of artificial neural network (ANNs) in the development of cost models. Although such advanced modelling techniques have been highly successful in many engineering areas, this success has been strongly dependent on the ability to choose the correct ANN structure. In this respect, choosing the most suitable structure for the individual processing elements that make up the ANN is essential. The research reported in this paper, therefore, makes use of the Taguchi methodology to identify best and worst structural elements for ANN processing elements. In order clearly to determine the accuracy of the models developed, cost information has been generated using a published cost model of a turning process. The cost information generated from this model has been used to train ANNs and test the resulting model for estimating accuracy. In order to measure accuracy, the 'percentage average absolute error' value has been adopted. Using this measure, the accuracy of models developed using the best and worst ANN structural elements have been compared with the use of regression analysis. The results indicate that the use of ANN to develop cost models is superior to regression analysis, although both methods fail to develop models that provide useful accuracies when large numbers of variables are involved.