A Transformer-based TTSC fault Diagnosis Method for Aircraft Brushless Generator

Turn-to-turn short-circuit fault (TTSC) can lead to significant losses and accidents, making its accurate diagnosis essential for safe aircraft operation. In recent years, transformer-based models have achieved good results for fault diagnosis. Inspired by it, this paper presents a fault diagnosis method for aircraft brushless generator stator windings TTSC, utilizing a transformer-based no decoder. The approach takes the no-load induced electromotive forces (EMFs) in the αβ reference frame as input features and outputs both the ratio of shorted turns and the α-axis component of the EMFs third harmonic. By relying solely on no-load stator voltage measurements, this method provides a clear overview of the stator windings TTSC. The adopted transformer-based model leverages a multi-head attention mechanism. This design enhances model robustness by reducing dependency on single-model interpretations and enabling more comprehensive feature extraction. To validate the proposed method, a dataset was generated using a simulated fault aircraft generator. Different degrees of fault severity are simulated by parallel connection of different resistors at different positions on the windings. Experimental results demonstrate a diagnostic accuracy of 95% on the test dataset, highlighting the method's acceptable potential in terms of accuracy and practicability.