Mapping the input–output relationship in HSLA steels through expert neural network

Abstract

Modification of the architecture of the artificial neural network is done to accommodate the information available from the knowledge base in the field of materials science for thermomechanically processed HSLA steel. The complicated architectures of these networks are made to satisfy the well-understood physical metallurgy principles, which administer the property response to the combined actions of the compositional and process parameters. The networks developed have been found to give very good convergence during training. The number of epochs required to reach the targeted error was found less for these networks than the conventional networks.