Improved ferrite number prediction in stainless steel arc welds using artificial neural networks - Part 2: Neural network results

The development of a neural network model, named FNN-1999, for predicting F errite Number in are welds as a function of alloy composition is described in Part 1. In this paper, the results of the model are compared to other me ans of predicting Ferrite Number in stainless steel welds. It was found the accuracy of the FNN-1999 model in predicting Ferrite Number is superior to that of the WRC-1992 diagram, the Function Fit model and a preliminary neu ral network model developed earlier. The error in fitting the current model to the training set was 40% less than that for the WRC-1992 diagram. In ad dition, the FNN-1999 model removes the restriction found in WRC-1992 and ma ny other constitution diagrams that each element's contribution to the Ferr ite Number is constant, regardless of the overall composition. Examples are given that show that with this added flexibility of the FNN-1999 model, th e impact of alloying additions varies as a function of concentration, and i n some cases the variation can be quite significant.