Machine Learning-Aided Process Design: Modeling and Prediction of Transformation Temperature for Pearlitic Steel

In this article, different machine learning (ML) algorithms are provided to predict the transformation temperature of pearlite using relevant material descriptors, austenitizing temperature, and cooling rate. To search for an appropriate model, the predictive performance of ML model including artificial neural network (ANN), generalized regression neural network (GRNN), radial basis function neural network (RBFNN), and extreme learning machine (ELM) is evaluated and compared on testing dataset. To quickly find the appropriate parameters, artificial fish swarm algorithm (AFSA) is applied to further improve the prediction accuracy of ANN model. In view of the distinguished prediction performance, the regression analysis shows that the GRNN model performs favorably against the other learning models. To verify the superiority of proposed model, a type of pearlitic steel with hypereutectoid composition is prepared. The phase transition temperature and transformation products are determined using diametral dilatometric and microscopy technique through continuous cooling transformation with cooling rates of 1 and 2 degrees C s-1. As a result, the experimental results agree well with the predictive results. The work proposes a reliable model to predict the phase transition temperature which can facilitate the optimization of process parameters to achieve the best possible microstructure.