Data-driven Prediction and Application of Steel Material Parameters

Yongshuai Xiu, Xiaohu Deng, Yixiao Sun, Yuedong Yuan, Gang Shen, Zunzhong Du, Xiaojun Yang, and Dongying Ju

DOI: https://doi.org/10.64486/m.65.4.5
Online publication date: March 3, 2026

Abstract: Accurate prediction of steel material parameters during heat treatment is essential for reliable finite element analysis (FEA) and process optimisation. Conventional experimental measurements and empirical models are often costly, time-consuming, and difficult to generalise to complex chemistries, microstructures, and temperature ranges. In this work, a data-driven prediction model is established using a comprehensive dataset that integrates simulation and experimental data, covering 18 elemental compositions, three typical microstructures, and a wide temperature range. Six key parameters are predicted simultaneously: thermal conductivity, specific heat capacity, yield stress, coefficient of thermal expansion, Young’s modulus, and density. Five machine learning models are evaluated, among which XGBoost shows the best performance for thermal parameters, while Gradient Boosting provides the highest accuracy for mechanical properties. After hyperparameter optimisation with grid search and cross-validation, all models achieve R² values above 0.99 and relative prediction errors within 5 %. An integrated Steel Materials Data Management System (S-MDMS) is further developed to combine data storage, visualisation, and online property prediction. The proposed model provides an efficient route for rapid acquisition and application of steel parameters in FEA-based heat treatment design and process optimisation.

Keywords: intelligent forecasting; machine learning; heat treatment; database; steels

This article is published online first and will appear in Metalurgija, Vol. 65, Issue 4 (2026).