热力学计算与机器学习相结合逆向设计镍基高温合金

Reverse design of nickel-based superalloys based on thermodynamic calculation and machine learning

采用热力学计算与机器学习相结合的方法进行镍基高温合金面向热力学性能要求的逆向设计。结果表明:通过高通量热力学计算成功构建镍基高温合金热力学计算数据集,为采用机器学习方法实现面向热力学性能要求的镍基高温合金逆向设计提供数据基础。针对热力学目标性能建立若干C2P模型,模型精度均高于99%。采用MLDS方法进行合金成分逆向设计,推荐的8种合金均满足性能的要求(1100℃下的γ'相体积分数Vγ',1100℃≥60%,V_(γ,1100℃)+Vγ',1100℃≥99%,γ'相熔点Tγ′≥1300℃)。热力学性能预测误差最小的3种合金实验验证表明,Vγ',1100℃均大于80%,时效后的组织中V_(γ,1100℃)+Vγ',1100℃≥99%,且Tγ′≥1300℃,均满足设计的要求。

The combination of thermodynamic calculation and machine learning was used to reverse design nickel-based superalloys for thermodynamic performance requirements.The results show that the thermodynamic calculation dataset of nickel-based superalloys is successfully constructed by highthroughput thermodynamic calculations,which provides the data basis for the reverse design of nickelbased superalloys with thermodynamic performance requirements by using machine learning methods.Several C2P models are established for the thermodynamic target performance,and the accuracy of the models is higher than 99%.MLDS method is used to reverse design the alloy composition,and the eight alloys are recommended to meet the performance requirements(Vγ',1100℃≥60%,V_(γ,1100℃)+Vγ',1100℃≥99%and Tγ′≥1300℃).The experimental verification of the three alloys with the smallest prediction error of thermomechanical properties shows that the Vγ',1100℃are greater than 80%,V_(γ,1100℃)+Vγ',1100℃≥99%in the microstructure after aging,and Tγ′≥1300℃,which meet the design requirements.