基于MEA-BP神经网络的涡轮盘低循环疲劳寿命预测与概率评估  

Low⁃Cycle Fatigue Life Prediction and Probability Evaluation of Turbine Disks Based on MEA⁃BP Neural Network

作　　者：闫文慧 高海峰 YAN Wenhui;GAO Haifeng(School of Mechatronic Engineering and Automation,Shanghai University,Shanghai 200444,China)

机构地区：[1]上海大学机电工程与自动化学院,上海200444

出　　处：《力学季刊》2023年第3期568-580,共13页Chinese Quarterly of Mechanics

基　　金：国家自然科学基金(51705309);中国博士后科学基金(2017M621481)。

摘　　要：作为航空发动机的关键部件,涡轮盘在高温、高转速的严酷条件下工作.低循环疲劳作为涡轮盘的主要失效模式,对其失效寿命和可靠性有重要影响.涡轮盘的结构复杂性,使得基于仿真数据的疲劳寿命预测和概率评估的难度显著增加,直接使用蒙特·卡罗方法(Monte Carlo Method,MCM)的计算量非常大,而传统响应面法(Response Surface Method,RSM)精度达不到计算要求.鉴于思维进化算法(Mind Evolutionary Algorithm,MEA)优化反向传播(Back Propagation,BP)神经网络的方法具有很强的非线性逼近能力,本文探索MEA-BP神经网络和MCM抽样相结合的方法,开展涡轮盘的低循环疲劳寿命预测与概率评估.与MCM、RSM、BP神经网络法进行对比,验证MEA-BP神经网络的优越性.结果表明,MEA-BP神经网络不但可以满足精度要求,而且可以显著提高计算效率.通过灵敏度分析,得出了影响寿命的主要因素,为涡轮盘可靠性设计提供有效依据.As a key component of aero-engines,turbine disks work under severe conditions of high-temperature and high speed.As one of the main failure modes of turbine disks,low-cycle fatigue(LCF)has an important impact on the fatigue life and reliability.The structural complexity of turbine disks makes the fatigue life prediction and probability assessment based on simulation data significantly difficult.For example,direct Monte Carlo method(MCM)requires a large amount of calculation,while the traditional response surface method(RSM)cannot meet the requirements of the computational accuracy.In view of the strong nonlinear approximation ability of the back-propagation(BP)neural network optimized by the mind evolutionary algorithm(MEA),this paper explores the combination of MEA-BP neural network and MCM to carry out LCF life prediction and probability evaluation of the turbine disk.Compared with RSM,MCM and BP neural network,the superiority of the MEA-BP neural network is verified.The results show that the MEA-BP neural network can not only meet the requirement of accuracy,but also significantly improve the computational efficiency.Through the sensitivity analysis,the main factors affecting the LCF life are obtained,providing an effective basis for the reliability design of turbine disk.

关 键 词：涡轮盘 低循环疲劳 MEA-BP 神经网络 概率评估 

分 类 号：V231.95[航空宇航科学与技术—航空宇航推进理论与工程]