JATCO Ltd, The University of Tokyo, and MI-6 Ltd. have jointly developed a method for predicting abnormal grain growth in the manufacturing process of carburized drivetrain components. The results were published in the technical journal JATCO Technical Review.
In the carburizing process of steel used for drivetrain components in automobiles and other machinery, the occurrence of "abnormal grain growth (hereafter referred to as G.G.)"—where crystal grains become locally coarse—causes a decline in product strength, making its suppression a critical challenge. However, because factors such as microstructure, strain, and carburizing temperature interact in complex ways, it has been difficult to capture the specific conditions under which G.G. occurs simply by examining individual factors in isolation.
To address this issue, we utilized a "Bayesian Network," a probabilistic model capable of handling relationships among multiple interrelated factors. Considering the entire manufacturing process from raw material to cold forming and carburizing, we structured the relationships between parameters based on material and manufacturing domain expertise. By combining this structure with experimental data, we developed a model that predicts the occurrence of A.G.G. directly from material characteristics and manufacturing conditions—without requiring direct measurement or input of intermediate properties during the process. As a result, the model successfully predicted non-G.G. manufacturing conditions with 88% precision.
The establishment of this technology enables the advance prediction of G.G. risk without measuring intermediate characteristics along the process. Furthermore, because the model allows users to change parameter values in the model and see how the predicted probability changes when varying individual parameters, it can be utilized both to evaluate manufacturing conditions for suppressing G.G. and to analyze contributing factors. This is expected to streamline the design and improvement of manufacturing conditions, as well as investigate possible causesduring anomaly occurrences.
Article Information
- Title: Prediction of Abnormal Grain Growth in Carburized Parts Using Bayesian Networks
- Journal: JATCO TECHNICAL REVIEW No.25
- Publication URL:https://www.jatco.co.jp/wp-jatco/wp-content/themes/JATCO/assets/document/review/english/2025/JTR25_08_EN.pdf
- Authors ※Affiliations are based on the information at the time of publication.
- Yasuaki Ito, Go Kato, Makoto Maeda (JATCO Ltd)
- Junya Inoue (The University of Tokyo)
- Takumi Yoshida, Tsubasa Yamashita, Shuhei Kojima (MI-6 Ltd.)
