AUV Thruster Multi-source Fusion Fault Diagnosis Method Based On Deep Learning

The underwater vehicle is widely used in military and civil fields.The propeller is an important equipment of the underwater vehicle,which needs to be monitored and diagnosed in real time to ensure the safe navigation of the underwater vehicle.The fault diagnosis method based on deep learning can carry out "end-to-end" diagnosis without relying on any expert experience,and is not limited by the specific architecture of AUV,and has good generalization performance,so it is the mainstream method at present.In this thesis,onedimensional residual shrinkage network is taken as the backbone network,focusing on the strong noise interference problem,small sample problem,diagnosis results based on single sensor signal with uncertainty,specific work is summarized as follows:(1)In order to improve the fault diagnosis accuracy of AUV thruster in noisy environment,this thesis improved the one-dimensional residual shrinkage network(1dDRSN),and proposed a fault diagnosis model of AUV thruster based on multi-scale depth residual shrinkage network.First,a wide convolution layer is used to increase the number of input signal channels and reduce the dimension of the signal.Then,in the feature extraction layer,1d-DRSN of different sizes is used to extract the feature and information fusion of the output of the wide convolutional layer.The cross-layer connection advantage of the residual shrinkage network is used to improve the depth of the network,and the soft threshold function is used to remove the noise interference in the sample signal.Finally,the classification of fault signals is realized through the full connection layer.Experiments show that compared with DCNN,Res Net and 1d-DRSN,the proposed model has better anti-noise performance and the highest fault diagnosis accuracy,and can be used for fault diagnosis of AUV propeller in noisy environment.(2)Aiming at the problem that the method mentioned above has poor diagnostic effect in the case of small samples,a fault diagnosis model of AUV thruster based on the twin depth residual shrinkage network is established.The input of the model changes from a single sample to a pair of randomly matched samples,which realizes the doubling of the sample size.The feature extraction module of the model is a one-dimensional residual shrinkage network with the same structure parameters,which improves the feature extraction ability of the model under strong noise interference.Finally,the Euclidean distance between the two samples is obtained through the similarity calculation module.Experimental results show that compared with the traditional deep learning method,the proposed model has higher fault diagnosis accuracy in the case of small samples of AUV thrusters.(3)The signal of single sensor is easy to be interfered by external noise,which may not be enough to accurately represent the health condition of AUV thruster,thus leading to the uncertainty of fault diagnosis results.A fault diagnosis model of AUV thruster based on multi-source signal fusion and depth residual shrinkage network is proposed in this thesis.The model has three independent channels,and the angular velocity signals collected by gyroscopes in different directions are used as input vectors for each channel.The residual shrinkage network is used to remove the noise interference of each sensor signal and extract useful fault features.Finally,the full connection layer is introduced to fuse the features of each channel signal and output the diagnosis results.Experimental results show that the accuracy of the multi-source signal fusion diagnosis model proposed in this thesis reaches100%,effectively eliminating the uncertainty in the process of fault diagnosis,and has good anti-noise performance in the case of small samples,and has the feasibility of application in actual scenarios.