Physics-informed distributed modeling for CCF reliability evaluation of aeroengine rotor systems. (February 2023)

• Record Type: Journal Article
• Authors: Li, Xue-Qin; Song, Lu-Kai; Bai, Guang-Chen; Li, Dao-Guang
• Is Part Of: International journal of fatigue. Volume 167:Part B(2023)
• Publication Date: 2023-02
• Subjects: Physics-informed -- Aeroengine rotor systems -- Combined cycle fatigue -- Reliability evaluation; Materials -- Fatigue -- Periodicals; Materials -- Fatigue; Periodicals; 620.1122
• DOI: 10.1016/j.ijfatigue.2022.107342
• Languages: English
• ISSNs: 0142-1123

Abstract

Highlights: A physics-informed distributed modeling (PIDM) is proposed for the first time. The PIDM is built by absorbing the physics-informed idea into the distributed modeling. A system reliability framework concerning multiple frail sites and multiple failure modes is first built. The PIDM is validated to hold high accuracy and efficiency in rotor system reliability evaluation. The study shed the light on the development of physics-informed modeling in reliability evaluations. Abstract: Reliability evaluation of aeroengine rotor systems is often characterized by multiple correlated frail sites and multiple coupled failure modes, leading to the traditional integral modeling methods or separate modeling methods prone to unacceptable computing efficiency or accuracy. In this case, by absorbing the physics-informed thought into the distributed modeling, a novel physics-informed distributed modeling (PIDM) method is presented. In PIDM, based on the physical information of frail sites and failure modes, the complex reliability problem is first decomposed into several distributed sub-problems (i.e., multi-variate, multi-disciplinary and multi-objective); moreover, based on the extreme gradient boosting (XGB) algorithm, the mutually nested distributed sub-models of multi-variate mapping, static/dynamic multi-disciplinary and multi-objective reliability are established; finally, the multi-variate synchronous mapping, combined cycle fatigue (CCF) prediction, and integrated reliability analysis are achieved by the built distributed sub-models. To verify the effectiveness of the proposed method, the CCF reliability analysis of a typical high-pressure compressor rotor is performed. The comparisons of the direct Monte Carlo method, support vector regression (SVR), multilayer perceptron (MLP), extreme gradient boosting (XGB), MLP-based PIDM (MLP-PIDM), and XGB-based PIDM (XGB-PIDM) reveal that the proposed PIDM method holds the computing benefits on accuracy and efficiency. The current effort shed the light on the theoretical development of physics-informed modeling approaches for complex reliability evaluations.