Fault Diagnosis of Aero‐Engine Gas Path System Under Time‐varying Operating Condition

ABSTRACT An aero‐engine usually works in time‐varying operating condition. The coupling effect between gas path fault mode and operating condition makes the characteristics of fault mode vary with the operating condition. It increases the difficulty of fault diagnosis of gas path system. Moreover, class imbalance commonly exists between normal data and fault mode data. To address these challenges, we propose a multi‐stage semi‐supervised fault diagnosis method for gas path system, considering time‐varying operating condition and class imbalance. At the first stage, a multilayer perceptron (MLP) is constructed to identify the operating condition based on altitude, Mach number, fuel flow, and high and low‐pressure rotor speed. At the second stage, an auxiliary classifier Wasserstein generative adversarial network with gradient penalty (ACWGAN‐GP) model is developed to create fault samples during the self‐training process. All the data are divided into three categories, that is, pressure, speed, and temperature data, which are used to train fault diagnosis mode, respectively. The Dempster–Shafer evidence theory is employed for the information fusion at the decision level. The effectiveness of our proposed method is validated by data generated by a gas turbine simulation program.