Research On Underwater Thruster Fault Diagnosis Method Based On Sequence Reconstruction And Deep Learning

Much of the Earth’s surface is covered by bodies of water such as oceans,rivers and lakes,many of which remain unexplored.With human’s exploration of the ocean,underwater robots can be found in the Arctic,Mariana Trench,seaside and so on.It can be seen that underwater robots are playing a wide role in ocean exploration.However,it is a great challenge for underwater vehicle thrusters to work continuously under harsh conditions of long time and high strength.In order to enable underwater vehicle to perform safe and stable tasks,it is of great practical significance to study the fault diagnosis of underwater thrusters.Aiming at the fault diagnosis problem of underwater thruster,this thesis studies the fault signal of underwater thruster from three perspectives: time domain,frequency domain and time frequency domain.The main contents of this thesis are as follows:(1)In order to effectively study the problem of fault diagnosis of underwater thrusters,the development of underwater robot comprehensive experiment system provides a guarantee for subsequent experimental analysis and verification.According to the basic operation requirements of underwater robot experiment,the underwater robot body is built.According to the functional requirements of underwater robot,the underactuated multi-degree of freedom underwater manipulator is designed.Aiming at the safety and attitude adjustment requirements of underwater robot,buoyancy adjustment device is designed.In order to ensure the performance of buoyancy regulating device,a test device of buoyancy regulating device is designed.(2)For weak fault features and changes in the position of fault information in the monitoring signal sequence,the fault diagnosis accuracy will be reduced.In this thesis,a deep learning diagnosis method of underwater thruster fault based on time sequence feature enhancement and reconstruction is proposed.The fault features are enhanced by signal processing methods such as wavelet decomposition and modified Bayes algorithm,and the time sequence is reconstructed according to the position of maximum fault feature value.The fault degree identification of thruster is carried out based on deep learning model.By comparing the proposed method with the unreconstructed time series,it is verified that the proposed method is beneficial to improve the efficiency and accuracy of fault identification.(3)When fault diagnosis is carried out for a single sequential channel and time domain signal,there are problems such as incomplete fault information characterization and limited fault identification accuracy.A fault identification method of underwater thruster based on time series channel extension and frequency sequence fusion is proposed.Based on signal data processing methods such as wavelet decomposition method and evidence theory fusion,the single time sequence channel is extended,multiple time sequences are transformed into multiple frequency sequences based on fast Fourier transform,and the diagnosis results of convolution neural network are compared between the single frequency sequence and the two-dimensional matrix fused by different sorting of multiple single frequency sequences.It is proved that the proposed method is beneficial to improve the efficiency and accuracy of fault identification.(4)Aiming at the limitations of local fault features that cannot be characterized in frequency domain signals and the limited ability of two-dimensional convolutional neural networks to extract time-frequency domain information,a three-dimensional convolutional neural network diagnosis method for underwater thruster faults based on time-frequency power spectrum reconstruction and fusion is proposed.The time-frequency power spectrum of the underwater vehicle’s dynamic signals is obtained based on the smooth pseudo-Wigner-Ville distribution algorithm,and the instantaneous Shannon entropy curve is obtained based on the instantaneous probability density function.The time-frequency power spectrum is reconstructed and fused according to the position of the minimum value of the instantaneous Shannon entropy curve.By comparing the results of fault identification without reconstruction and after reconstruction fusion,it is verified that the proposed method is beneficial to improve the efficiency and accuracy of fault identification.