Prediction of mechanical properties of 16Mn large tube plates based on machine learning

Yi Zhong, Feng Mao, Xin Li, Kunlin Miao, Ruxing Shi, Baoning Yu, Chong Chen, Changji Wang, Hua Yu, Shizhong Wei,
Prediction of mechanical properties of 16Mn large tube plates based on machine learning,
Materials Today Communications,
Volume 49,
2025,
113712,
ISSN 2352-4928,
https://doi.org/10.1016/j.mtcomm.2025.113712.
(https://www.sciencedirect.com/science/article/pii/S235249282502224X)
Abstract: Large tube plates, used extensively in nuclear energy and similar industries, are vital for efficient heat exchange and fluid transport. However, predicting their mechanical properties through traditional methods is costly, time-consuming, and lacks precision. This study leverages machine learning to predict the mechanical properties of 16Mn steel large tube plates using real production data from a steel plant. After preprocessing the data, machine learning methods, including Support Vector Machine (SVM) and Backpropagation Neural Networks (BPNN), were applied to construct predictive models. Among these, the Support Vector Machine with Radial Basis Function (SVM-RBF) algorithm demonstrated exceptional performance, achieving R² values exceeding 0.95 for various mechanical properties (e.g., Testing set of YS(RT): R2= 0.98, RMSE= 6.5 MPa, MAE= 5.45 MPa, MAPE= 1.47 %). In addition, the model’s generalization ability and predictive reliability were systematically evaluated using cross-validation techniques. Sensitivity analysis was conducted using the Mean Impact Value (MIV) method, revealing the key factors influencing the production process, such as pouring temperature and finishing forging temperature. Experimental validation showed the model’s accuracy, with prediction errors for mechanical properties lower 2 %. These results suggest that the integrated approach of data-driven modeling and experimental validation offers a robust solution for optimizing the production and performance prediction of 16Mn steel large tube plates, advancing the development of materials with tailored mechanical properties.
Keywords: 16Mn Large tube plate; Machine learning; Mechanical properties; Sensitivity analysis