Prediction of Carbon Concentration Distribution in Carburizing Steel Based on Convolutional Neural Networks

The carburizing and quenching heat treatment process can form a carburized layer with high strength, high hardness and high wear resistance on the surface of the gear. At the same time, it still maintains high toughness in the gear core. According to the carburizing process parameters of specific materials, the accurate and rapid prediction of the surface carbon concentration distribution after carburizing can improve the production efficiency and quality of heat treatment. Therefore, this paper establishes accurate numerical models of the carburizing process in three different shapes of 20CrMnTi material, pentagon, circle and trapezoid, and sets 1440 sets of process parameter combinations for each shape for large-scale calculation. After collecting and processing the simulation data, CBNNet is proposed, which takes the structural geometry and process parameters as input, and the carbon concentration value as output. On the test set, the model showed high accuracy with a mean square error of 2.49x10 ^-5 . It shows that the model has strong generalization, and it is expected to improve the efficiency and quality of heat treatment in actual production.