Predicting the Tensile Strength of Steel Based on Different Deep Learning Models

This article is based on the Chinese Materials Genome Engineering Database (MGED) to construct deep learning models for predicting the tensile strength of steel. The predictive performance of the model is compared by three performance characterization data: average determination coefficient, average absolute percentage error, and relative mean square error. The Pearson parameter results show that the composition of steel and heat treatment process are significantly related to the tensile strength of steel. The random forest model has good predictive ability before and after training, and the K-nearest neighbor model has similar predictive ability to the random forest model after training. The decision tree model has good predictive ability for the tensile strength of steel after training, but is inferior to the K-nearest neighbor model and random forest model. By comparing the predictive performance parameters of the multi-layer neural network model, it shows that the model has certain predictive ability, but further data training and structural parameter optimization are needed.