PREDICTION OF DEPTH OF NITRIDE LAYER IN IRON DURING GAS NITRIDING USING NEURAL NETWORK

Abstract

Surface engineering of materials can add economic value to the material. Gas nitriding in iron is a typical thermochemical surface engineering process at eutectoid temperatures, where nitrogen diffuses to the surface to form nitride layers in the form of gamma phase and epsilon phase. In this study, a computational approach will be taken to predict the formation of the nitriding layer. In the prediction, a backpropagation neural network is used with input parameters of temperature, nitriding potential and time with an output of nitriding layer depth. This prediction does not distinguish the phase formed in the nitriding layer. The best results were obtained in the model of single hidden layer with 5 neurons and two hidden layers with the formation starting with 6 neurons followed by 5 neurons. The mean square error of the training data for the single hidden layer is 0.0027. While for two hidden layers the value is higher at 0.0032. The results obtained for the absolute mean error and root mean square values for the single hidden layer model are 0.6117 and 0.9670. For the two hidden layers model, the absolute error and root mean square values are 0.5894 and 1.0472. It can be seen from the correlation coefficient that both models can only predict well at depths of more than 10 μm.