IMR-HACSM: Hybrid adaptive coordination surrogate modeling-based improved moving regression approach for cascading reliability evaluation

The cascading reliability evaluation of multi-failure modes of complex system/structure usually needs to repeatedly establish mathematical models with the step-by-step modeling strategy, which weakens the correlation between multi-failure modes. To improve the efficiency and precision of cascading reliability evaluation, a hybrid adaptive coordination surrogate modeling-based improved moving regression (IMR-HACSM, short for) method is proposed based on hybrid adaptive coordination surrogate modeling (HACSM) and improved moving regression (IMR) technique. In this proposed method, the HACSM is developed from coordinative modeling idea, adaptive selection strategy and surrogate model, and the IMR technique is expanded by artificial protozoa optimizer (APO) algorithm and moving least square (MLS) method. Herein, the coordinative modeling idea is used to collaboratively establish these models of associated failures, the adaptive selection strategy is utilized to choose suitable forms of mathematical models of cascading failure and associated failures, the surrogate model is treated as the basis functions of top-objective and sub-objectives, the APO algorithm is employed to search the optimal radius of compact support region and to ensure effective modeling samples, and the MLS method is adopted to resolve these unknown coefficients of mathematical models. Besides, three examples are used to verify the effectiveness of the proposed method, including a two-level nested function approximation, an aircraft landing gear brake system temperature difference reliability assessment and an aeroengine turbine blisk low cycle fatigue (LCF) life reliability analysis. The results show that the IMR-HACSM method holds excellent modeling features and simulation performance relative to some existing different methods, including response surface method (RSM), Kriging, support vector machine (SVM), artificial neural network (ANN), RSM-based moving regression (MR-RSM), Kriging-based moving regression (MR-K), SVM-based moving regression (MR-SVM), ANN-based moving regression (MR-ANN), RSM-based IMR (IMR-RSM), Kriging-based IMR (IMR-K), SVM-based IMR (IMR-SVM) and ANN-based IMR (IMR-ANN). The efforts of this work provide useful ways for the cascading reliability evaluation of complex system/structure, and contribute to the design of high reliability products.