Application of Bayesian Neural Network for modeling and prediction of ferrite number in austenitic stainless steel welds

As neural networks are extremely useful in recognizing patterns in complex data, Bayesian neural network analysis has been followed in the present work to reveal the influence of compositional variations on ferrite content for the austenitic stainless steel base compositions from the available database and to study the significance of individual elements on ferrite content in austenitic stainless steel welds based on the optimized neural network model. Bayesian neural network’s predictions are accompanied by error ba s and the significance of each input variable is automatically quantified in this type of analysis. Neural network model based on Bayesian framework for ferrite prediction in austenitic stainless steel welds has been developed using the database which was used for generating the WRC 92 diagram. The Bayesian framework uses a committee of models for generalization rather than a single model. The best model was chosen based on minimum in the test error and maximum in the logarithmic predictive error. The optimized model can be used for predicting the ferrite number in austenitic stainless steel welds with a better accuracy than the constitution diagrams. Using this model , the influence of variations in the individual elements such as carbon, manganese, silicon, chromium, nickel, molybdenum, nitrogen, niobium, titanium, copper, vanadium, and cobalt on the ferrite number in austenitic stainless steel welds has been determined. It was found that the change in ferrite number is a non-linear function of the