A Simulation-to-Real Transformation for Small-Sample Fault Diagnosis in Aeroengine Dual-Rotor Systems

Dual-rotor systems are critical components in aeroengines, where failures can lead to severe operational disruptions and significant economic losses. However, the high reliability of these systems results in limited fault data, creating a typical small-sample problem. Due to the uniqueness of dual-rotor systems, small-sample fault diagnosis faces two distinct challenges: 1) significant attenuation of fault characteristics during signal transmission and 2) complex dynamic modeling due to asynchronous vibration coupling and structural nonlinearity. To address these challenges, this article presents the first simulation-to-real transformation framework for dual-rotor systems. Specifically, an asymmetric Gaussian chirplet model (AGCM) is developed to preprocess data and enhance fault characteristics in the time–frequency domain, addressing the feature attenuation issue. For complex dynamic modeling, we propose a two-step approach: first, employing a Hertzian contact theory-based simulation model to generate labeled fault data. Nevertheless, due to the inherent complexity of real systems, a significant distribution gap exists between simulation and real data. To bridge this gap, we introduce an innovative adaptive multiscale style transfer network (AMSTN) to embed real-world style characteristics while preserving critical fault features. Experimental results demonstrate the framework’s superior performance under small-sample conditions.