A MLP Neural Network Approach for Prediction of Amount of High Carbon Retained Austenite in Austempered Ductile Irons

Important characteristics such as high strength to weight ratio, toughness, ductility, fatigue strength, wear resistance, high heat conductivity, excellent damping capacity and low notch effect sensitivity beside low production costs result in that Austempered Ductile Irons (ADI) to be so significant for designers and metallurgists. Retained austenite, which is enriched with carbon during isothermal austempering transformation, is the main reason of these unique properties. According to the intricate, costly and inaccurate typical method for determining the amount of retained austenite volume fraction, the purpose of this investigation is presenting an artificial neural network based model to estimate this phase by utilizing an extended database of published authoritative data. Therefore, 2141 data (including 10 inputs; %C, %Si, %Mn, %Ni, %Cu, %Mo, austenitizing time, austenitizing temperature, austempering time, austempering temperature and an output; retained austenite content after austempering) were used for generating a Multi Layer Perceptron (MLP) neural network with 3 to 10 cells in a hidden layer. Activation functions in hidden layer are forms of non-linear, continuous and differentiable sigmoid tangent and output functions are linear. Evaluation test results showed that the selected synthesized MLP-ANN can easily and precisely estimate the amount of retained austenite according to chemical composition and heat treatment parameters, with a negligible error.