面向低周疲劳可靠性评估的选择性集成协同代理建模方法

A selective ensemble collaborative surrogate modeling method for reliability assessment of low-cycle fatigue

摘要:

机械结构的低周疲劳可靠性评估对于保障结构的安全性和使用寿命至关重要。为有效进行低周疲劳可靠性评估，本文基于协同建模方法、基函数选择策略及Stacking集成学习，提出了一种选择性集成协同代理建模（Selective ensemble collaborative surrogate modeling，简称SECSM）方法。首先，基于协同建模方法将分析问题分解为主目标和子目标。然后，利用留一交叉验证剔除基函数模型库中精度不佳的代理模型。最后，从剩余的模型中选择基学习器和元学习器进行Stacking集成学习。通过航空发动机高压涡轮叶盘在流-热-固耦合作用下的低周疲劳可靠性评估，验证了不同代理模型方法的建模特性和仿真性能。结果表明，SECSM方法具有较低的平均绝对误差、较高的拟合优度和仿真精度，表现出良好的建模特性和仿真性能，为机械结构的低周疲劳可靠性分析提供了有益的见解。

Abstract:

Low-cycle fatigue reliability assessment of mechanical structures is crucial for ensuring the safety and service life of structures. To effectively conduct low-cycle fatigue reliability assessment, an elective ensemble collaborative surrogate modeling (SECSM) method was proposed based on collaborative modeling method, basis function selection strategy, and Stacking ensemble learning. First, the collaborative modeling method is used to decompose the analysis problem into sub-objectives and main objective. Then, leave-one-out cross-validation is applied to exclude surrogate models with poor precision from the basis function model family. Finally, base learners and meta-learner are selected from the remaining models for Stacking ensemble learning. The modeling characteristics and simulation performance of different surrogate models are verified through the low-cycle fatigue reliability assessment for high-pressure turbine blisk of an aeroengine under flow-thermal-structural coupling. The results show that the SECSM method has a low mean absolute error, high goodness of fit, and simulation precision, demonstrating excellent modeling characteristics and simulation performance. It provides valuable insights for low-cycle fatigue reliability analysis of mechanical structures.