Development of Carbon Concentration Prediction Technology from Microstructure Images of Carburizing Heat Treatment Products using Machine Learning

Abstract

Carburizing, quenching and tempering heat treatments are widely used as surface hardening, and it is known that their performances are changed depending on the carbon concentration. We had verified the effectiveness of the technique for predicting the carbon concentration from the microstructure images after heat treatment by using image recognition technologies and machine learning technologies that were focused in recent years. These technologies extracted the black and white objects of the image, and then machine learning of each shape information and brightness information as a features．Based on these technologies, we succeeded in predicting 0.78 mass％ of microstructure in the range of 0.75 mass％ to 0.79 mass％ by using regression models according to the characters of microstructure images.