NEURAL-NETWORK APPROACH TO FLOW-STRESS EVALUATION IN HOT DEFORMATION

With increase in the use of finite-element methods to characterize the workpiece behaviour under different processing conditions in metal-forming operations, an effective means of establishing the complicated relationship between the flow stress of the material and the process variables may be explored. In view of the widespread use of neural networks in addressing problems that are intractable and cumbersome with traditional methods, the present study tried to investigate the feasibility of utilizing a neural network to extract the complex relationships involved in hot-deformation process modelling. Flow stress data obtained on a medium carbon steel under conditions of constant strain rate and temperature was used in conjunction with a back-propagation neural network for the purpose of training the network, which could in turn be used to predict the flow-stress values for any given processing conditions. It has been found that the Bow-stress values predicted by the network, within the input pattern range, agree closely with actual experimental values, thus indicating the possibility of using the neural network approach to tackle hot deformation problems.