Fault diagnosis method of the USV based on improved Res2Net multi-scale feature fusion

With the application of unmanned surface vessel (USV) in military and civilian fields, people have put forward higher requirements for the safety and reliability of USV. When the USV fails, its autonomy will be affected, and may cause the USV unable to return back. It is important to accurately detect the faults of USV to ensure their safe operation. In response to the problem of difficult extraction of fault features for USV in complex water conditions, an improved Res2Net multi-scale feature fusion network based on Gramian Angular Field (GAF) is proposed. Firstly, the vibration signal is converted to GAF. Secondly, it is input into the constructed Depthwise Separable Inverted Res2Net residual block (DSIRes2Net) to extract the multiscale features. The multi-scale features are integrated into the low-dimensional features. Finally, these features are put into the Adaptive Time-Frequency feature Fusion block (ATFF) to adaptively fuse the time-domain and frequency-domain features, so as to realize the fault diagnosis of the USV. The experimental results show that the fault diagnosis accuracy of the USV obtained from the proposed algorithm can reach 98.5%, with a variance of 0.17, and a standard deviation of 0.41. The accuracy and stability of the fault diagnosis model have been effectively improved. It provides an important guarantee for the safe operation of the USV.