Hardness Prediction of Heat Treated Iron Parts Manufactured by Powder Metallurgy Using Neural Network and Fuzzy Rule-Based Models

Most of the parameters in PM processes are nonlinearly related and coupled, therefore selection of materials and processing parameters involves lots of trial-error efforts which are time taking and expensive. Many attempts have been made to model these processes in order to select processing parameters to achieve desired properties precisely before committing experimental resources and therefore reduce the final cost. This paper is the result of a case study in an effort to model material behavior during heat treatment using artificial neural network (ANN) and fuzzy rule-based models. To prepare an appropriate model, cold compacted-sintered parts were hardened in heat treatment furnace to reach desired hardness (output argument) at various temperature and time duration (input arguments). The performance of the models is compared with the Multiple Linear Regression (MLR) model with respect to Mean Square Error (MSE), Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and co-efficient of determination (R2). It has been observed that although ANN model had acceptable accuracy, the fuzzy rule-based model with max relative error of 7% had the best performance and capability of prediction.