Predicting mechanical properties of low-alloy steels using features extracted from Electron Backscatter Diffraction characterization

Machine learning (ML) approaches have recently been increasingly employed to establish quantitative relationships between material composition, processing, microstructure, and properties. However, the complexities of microstructure pose challenges for straightforward modeling, thereby complicating research efforts. This study introduces a series of multi-parametric quantification methods based on Electron Backscatter Diffraction (EBSD) data tailored to the microstructural characteristics of low-alloy steels. These methods include quantification of boundary densities across various misorientation angles, distinct types of boundaries, and geometrically necessary dislocation densities. Through thermomechanical simulation and micro-tensile testing of low-alloy steels, data on yield and ultimate tensile strengths were obtained, alongside EBSD-based extraction of microstructure characteristics. Several ML methods, including Random Forest (RF), Gradient Boosting Decision Trees (GBDT), and Extreme Gradient Boosting (XGBoost), were utilized to predict yield strength (YS) and ultimate tensile strength (UTS) using the aforementioned microstructural features. The GBDT model outperformed other algorithms, demonstrating high accuracy in predicting YS, UTS, and elongation. The model achieved an Mean Squared Error (MSE) of 972.18, an Mean Absolute Error (MAE) of 24.75 and an Coefficient of Determination (R 2 ) of 0.864 for YS, and an MSE of 812.28, an MAE of 22.87 and an R 2 of 0.823 for UTS. These results confirm GBDT's effectiveness in predicting mechanical properties from microstructural data.This successful integration of ML with multi-parametric description microstructural features underscores its potential in facilitating material design and development processes.