镍基粉末高温合金的室温和高温力学性能:机器学习建模与实验验证

Room/High-temperature Mechanical Properties of Nickel-Based Powder Metallurgy Superalloys:Machine Learning Modeling and Experimental Validation

本文采用机器学习方法实现了镍基粉末高温合金在不同温度下的力学性能精准预测。首先,通过文献评估构建了一个包含166条数据的镍基粉末高温合金数据集,经数据清洗和特征分析后,从合金成分、热处理工艺和测试温度出发,建立了预测镍基粉末高温合金不同温度下的屈服强度与延伸率的高精度预测模型。其次,利用构建的模型预测了未在训练数据集中使用的新型镍基粉末高温合金FGH4113A在不同热处理工艺和测试温度下的力学性能,并进行了实验验证。结果表明,所构建的模型对FGH4113A合金的室温屈服强度预测的平均绝对误差(Mean absolute error,MAE)为31.49 MPa,室温延伸率的MAE为2.45%,而高温屈服强度预测的MAE为76.14 MPa,高温延伸率的MAE为2.06%,表明模型具有良好的泛化能力。最后,通过分析模型的可解释性,提出在一定范围内提高二级时效温度并降低一级时效温度可以同时提高镍基粉末高温合金的屈服强度和延伸率,这为合金的热处理工艺设计提供了新思路。

In this research,machine learning(ML)method was employed to realize the accurate prediction of the temperature-dependent mechanical properties of nickel-based powder metallurgy superalloys.Firstly,a dataset comprising 166 pieces of data was first compiled through comprehensive literature review.Following the data cleaning and feature analysis,the ML models with high predictive performance on the yield strength(YS)and elongation(EL)of nickel-based powder metallurgy superalloys at varying test temperatures were constructed by using the composition,heat treatment technologies,and test temperature as inputs.After that,the YS and EL of a novel nickel-based powder metallurgy superalloy(FGH4113A)at different heat treatment states and test temperatures were predicted and experimentally validated without its information in the training dataset.The mean absolute error(MAE)of constructed models on room-temperature YS prediction of FGH4113A alloy is 31.49 MPa,and the MAE on room-temperature EL prediction is 2.45%,while the MAE on high-temperature YS and EL prediction is 76.14Mpa and 2.06%,respectively.The results demonstrated the robust generalization capability of the constructed ML models.Finally,through interpretability analysis on the validated models,the increase of secondary aging temperature and the decrease of primary aging temperature can improve YS and EL simultaneously,which presents a novel avenue for heat treatment technology design for advanced powder metallurgy superalloys.