A COMPARATIVE-STUDY OF ARTIFICIAL NEURAL NETWORKS FOR THE PREDICTION OF CONSTITUTIVE BEHAVIOR OF HSLA AND CARBON-STEELS

Backpropagation neural networks are utilized to store and predict the flow stresses of several steels. A convergence algorithm using a varyi ng learning factor is developed which is shown to save one sixth of th e learning time when compared with the algorithm in which a constant l earning factor is utilized. A performance test shows that the well-tra ined neural network can interpolate flow stresses very well if the inf ormation for interpolation is sufficient in the training pairs. The ca pability of the network to extrapolate is found not to be impressive. The neural network can handle several groups of data during adaptive l earning simultaneously without losing accuracy. The time needed for ad aptive learning to reach a reasonable level of accuracy is short. Comp aring the predicted results to other models, the output of neural netw ork is shown to have the highest accuracy.