Improved ferrite number prediction in stainless steel arc welds using artificial neural networks - Part 1: Neural network development

Neural network modeling is a powerful nonlinear regression analysis method that is extremely useful in identifying behavioral trends. This methodology was applied to the problem of predicting Ferrite Number in arc welds as a function of composition. This paper describes the details of the developmen t of the neural network model, named FNN-1999, including the identification of the optimum network architecture and network parameters. The model was trained on the same data as the WRC-1992 constitution diagram and covers a range of Ferrite Numbers from 0 to 117, with a corresponding wide range in composition. Results of the model are presented in Part 2. It is shown that the accuracy of the FNN-1999 model in predicting Ferrite Number is superio r to the accuracy of other models that are currently available, including t he WRC-1992 diagram.