Intelligent classification of neural network models for mild steel behaviour in hot forming

The rheological behaviour of mild steel subjected to hot forming was modelled through a parallel distributed processing paradigm based on the artificial neural network prediction of the metal material response. The evaluation of different feedforward back-propagation neural networks for the flow stress prediction was carried out on the basis of laboratory data of the stress–strain behaviour of mild steel subjected to compression tests with different temperature and strain rate conditions. The results obtained consist of a number of neural network models capable of describing the material flow stress under the considered processing conditions with diverse levels of agreement with the experimental data. The availability of a range of neural network models for the prediction of mild steel rheological behaviour under hot forming conditions evidenced the need for optimum model identification. The latter was carried out through the development and application of selected intelligent computing methods: a supervised neural network technique, an unsupervised neural network procedure and a neuro-fuzzy system. Each of the developed intelligent computation paradigms provided for satisfactory results in the selection of an optimum neural network model for the flow stress prediction. The choice of the intelligent computing approach could therefore be determined by the complexity of the tool development phase and the ease in the resulting evaluation.