Performance prediction of nickel-based high-temperature alloys based on machine learning

Abstract

This paper proposes a machine learning-based approach for predicting the mechanical properties of nickel-based superalloys, aiming to achieve high-precision prediction and optimized design of alloy performance. By integrating multi-source data, we established a comprehensive database covering chemical composition and tensile mechanical properties of nickel-based superalloys, and systematically evaluated the performance of various machine learning algorithms in computational modeling, thereby providing a data-driven methodology for alloy design. First, a dataset of nickel-based superalloys was constructed and standardized to ensure data consistency and validity. Based on this dataset, a performance prediction model for nickel-based superalloys was developed. Algorithms such as Random Forest (RF), Gradient Boosting Decision Tree (GBDT), and Extreme Gradient Boosting (XGBoost) were employed to predict tensile strength, with Bayesian optimization applied to refine the model. Results demonstrated that the RF model performed optimally in tensile strength prediction, achieving an R2 value of 0.96. The Bayesianly optimized RF model has an R2 of 0.98. making it particularly suitable for this application. Analysis using the RF model revealed that elements such as Al, Ti, and Nb significantly influence tensile strength via γ′-phase strengthening mechanisms, whereas elements B and C play a critical role in reducing cross-sectional shrinkage. This finding is consistent with previous experimental and research results, making the model a valuable tool for subsequent high-temperature alloy design.