Abstract Autonomous aviation platforms, such as unmanned helicopters, rely heavily on reliable drive systems to ensure flight safety. However, complex operational environments, heterogeneous sensing data, and limited fault samples pose significant challenges to lifecycle health management of transportation equipment. To improve autonomous fault perception and decision-making capability, this paper proposes a multi-sensor fusion fault diagnosis framework (MsWPD-ViT) based on wavelet packet decomposition (WPD) and vision transformer (ViT) for intelligent fault diagnosis of helicopter tail drive systems. The framework first extracts fault-sensitive frequency sub-bands using WPD and employs short-time Fourier transform (STFT) and local binary pattern (LBP) encoding to generate lightweight texture descriptors that preserve weak fault information. Subsequently, shared convolutional layers are employed to reduce feature redundancy, and the dual-position encoding mechanism is designed to model both sensor spatial relationships and sub-band frequency orders. The ViT is finally introduced to capture global correlations across sensors and frequency components, facilitating intelligent feature fusion and accurate health state identification. Experimental results on a helicopter tail drive system demonstrate that MsWPD-ViT significantly outperforms other advanced methods, particularly with limited training data, while maintaining low computational cost. This work provides an effective solution for intelligent drive health management in autonomous aviation and offers insights for the development of future self-maintaining transportation equipment.
Wei et al. (2026) studied this question.