Research On Energy Feature Extraction And Multi-core Model Identification Of Underwater Thruster Fault Signal

With the development of Marine resources,the importance of Marine equipment has become increasingly prominent.As an indispensable Marine equipment for deep-sea exploration and resource development,the safety of underwater vehicle has been paid much attention by all sides.As the most important power component of the underwater vehicle,the thruster has the largest workload and the highest safety risk.Therefore,the research on the thruster fault diagnosis method of the underwater vehicle is the key to ensure the safety and reliability of the underwater operation,which has important research value and practical significance.Aiming at the problem of fault diagnosis of underwater vehicle thruster,this thesis studies the extraction of thruster fault features,and the identification of thruster fault degree,and verifies the effectiveness of the proposed method through the experiment of underwater vehicle pool.The method of fault feature extraction of underwater vehicle thruster is studied.Aiming at the problem that the fault feature value is not monotonous to the wavelet scale,the fault samples of different levels overlap,and the classification accuracy is not strong robust to the window length of MB algorithm in the process of fault extraction under the traditional PRE method,a fault feature extraction method based on the improved PRE algorithm is proposed in this thesis.To solve the above problems,the fusion of signals is omitted,a method of maximizing the time-frequency power spectral density is proposed and the samples are reconstructed by homomorphic transformation.The validity of the proposed method is verified by the experimental data of underwater vehicle pool.The method of fault identification of underwater vehicle thruster is studied.In this thesis,SVDD and FSVDD are respectively used to carry out the research on the fault identification method of underwater vehicle.Aiming at the problem of low identification accuracy under traditional SVDD method,a fault identification method based on least square SVDD is proposed in this thesis.By introducing the least square method into SVDD,a method of model selection and integration is adopted,and the boundary conditions of the model are set,so that the identification model is robust to the relative distance of the standard fault degree,and the accuracy of model identification is improved.The validity of the proposed fault identification method based on least square SVDD is verified by comparing the SVDD method with the proposed method.Aiming at the problem of insufficient sample utilization and low identification accuracy under the traditional FSVDD method,a fault identification method based on multi-core FSVDD is proposed in this thesis.Based on FSVDD,this thesis uses training samples corresponding to different fault levels as target samples to train multiple hypersphere models,and then uses training samples corresponding to fault levels other than the target level as nontarget samples to calculate monitoring coefficients.And establish multiple single-core identification models,and finally,integrate multiple single-core identification models into a multi-core identification model.In this article,the training samples corresponding to a certain fault level are used as target samples or non-target samples at different stages.Make full use of fault sample information and improve identification accuracy.The validity of the multicore FSVDD-based fault identification method proposed in this thesis is verified by comparing the FSVDD method and the method in this thesis.