Genetic algorithm (GA)–backpropagation (BP) network approach for hardness prediction of austempered ductile iron (ADI).

This article focuses on developing a predictive model for the hardness of austempered ductile iron (ADI), specifically the QT500-7 grade, based on heat treatment parameters using a genetic algorithm–backpropagation (GA–BP) neural network. Utilizing a dataset of 125 hardness measurements from samples subjected to varying austempering conditions, the GA was employed to optimize the initial weights of the backpropagation neural network, resulting in a model with a mean square error (MSE) of approximately 1.019 and a correlation coefficient of 0.986, indicating high accuracy and reliability. The study also correlates microstructural changes—such as increases in bainitic ferrite length and retained austenite volume fraction—with decreases in hardness as austenitizing and austempering temperatures rise. Overall, the GA–BP neural network model effectively predicts ADI hardness across a broad range of heat treatment parameters, offering a time-efficient alternative to experimental testing.