A Novel Active Learning Kriging-Based Reliability Analysis Method for Aero-Engine Gear

Abstract

This paper proposes the active learning Kriging (ALK)-based reliability method for high-cycle fatigue reliability analysis of aero-engine gears. Uncertainties to affect the reliability of aero-engine gears are quantified with random variables, and the finite element simulation model of gears is refined to align with experimental data. Based on the Basquin equation, the S–N curve of the gear is fitted to the stress–life data obtained from experiments. The stress under given loads is obtained through simulation, and the corresponding life is derived from the S–N curve. Using the given permissible lifespan, the limit state function for gear fatigue reliability analysis is established. This function is then approximated using an active learning surrogate model, and the probability of failure is subsequently estimated. Furthermore, to enhance computational efficiency and accuracy, this paper reviews the origin of active learning strategy and defines an improvement function aimed at structural reliability analysis by drawing an analogy to the derivation process of the expected improvement (EI) learning function in the efficient global optimization (EGO) algorithm. Consequently, a novel learning function for active learning Kriging-based reliability analysis is derived. The application of this method to aero-engine gears made of 17CrNiMo6 steel verifies that it effectively enhances the efficiency of fatigue reliability analysis under ensuring a certain accuracy.