当前位置:首页 > 预发布刊期->2022年43卷第7期
机器学习在高温合金粉末盘构件疲劳寿命预测中的应用
Prediction of Fatigue Life of Superalloy Powder Disk via Machine Learning
浏览(1045) 下载(0)
- DOI:
- 作者:
- 张国栋
ZHANG Guodong
- 作者单位:
- 西北工业大学
Northwestern Polytechnical University
- 关键词:
- 粉末冶金;机器学习;寿命预测;高温合金
powder metallurgy; machine learning; life prediction; superalloy
- 摘要:
- 粉末冶金工艺制备高温合金,由于制粉的污染、容器材料剥落等原因,不可避免地会引入非金属夹杂物。而这种夹杂物的存在会导致粉末冶金涡轮盘低周疲劳失效,对高温合金涡轮盘的力学性能有着极大的危害。但目前大部分的疲劳寿命模型是由Manson-Coffin公式发展而来的,且未考虑到疲劳过程中弹性模量的变化以及合金中缺陷特征等因素的影响,导致实际预测结果存在较大的误差。为此我们利用机器学习算法建立了夹杂距表面的距离和夹杂的大小与涡轮盘低周循环次数之间的定量预测模型,通过对比不同算法,即支持向量机、随机森林、梯度提升机、核岭回归和套索算法,发现梯度提升树算法能够更好的预测疲劳寿命。
Impurity are inevitably introduced in the preparation of superalloys in powder metallurgy process due to the possible pollution of powder and the peeling of container. The existence of such impurities will lead to low cycle fatigue failure of powder metallurgy turbine disk, which is very harmful to the mechanical properties of superalloy turbine disk. However, most of the current fatigue life models are developed from Manson-Coffin formula, and do not take into account the changes of elastic modulus in the fatigue process and the characteristics of defects in the alloy, resulting in large errors in the actual prediction results. We use machine learning algorithm to establish the quantitative prediction model that maps the impurities distance from the surface and the impurities size to the low cycle numbers. We use support vector regression, random forest, gradient boosting machine, kernel ridge regression and lasso to build the model. The comparison of the model performance shows that the gradient boosting tree can capture the tendency underlying the data better.