부산대학교 도서관

Deep learning has demonstrated notable proficiency in autonomous underwater vehicle (AUV) fault diagnosis applications. Nevertheless, the signals of AUVs working underwater are unstable and easily disturbed by noise, which makes it difficult for traditional deep-learning models to be applied to complex real-world scenarios. Therefore, it is crucial to overcome the impact of noise on feature extraction capabilities. To this end, this paper proposes an FFT-driven Transformer (FGF-Trans) framework for AUVs fault diagnosis. Specifically, the framework combines a frequency global filter block (FGF-Block) with a Transformer, which is able to capture the global contextual information of the entire feature instead of only local information. By virtue of this integration, the proposed framework can better extract fault feature information in noisy environments and exhibit excellent diagnostic accuracy. We experimentally validate the FGF-Trans method on the AUVs fault dataset. Experimental results show it has higher application potential and accuracy in actual scenarios than traditional deep learning models.