Research On Thruster Fault Feature Separation And Fault Degree Identification Method For Underwater Vehicle

With the increasing depletion of the land non-renewable resources,the development of ocean resources space has become the focus of all countries.Autonomous underwater vehicle(AUV),as the only equipment capable of working in deep ocean environment,plays an irreplaceable role in the development of marine resources.AUV works in the complex ocean environment with no human intervention on board and no umbilical cable connected to the mother ship,security is an important feature.Fault diagnosis is the basis and key technology to ensure the safety of AUV.Thruster is the key component with the heaviest load.Approaches associated with thruster fault diagnosis have great research significance and practical value in improving the security of AUV and speeding up the process of its application.The fault feature separation and fault degree identification in the AUV thruster fault diagnosis,containing AUV thruster fault feature extraction and fusion,AUV thruster fault feature and external disturbance feature separation,AUV thruster fault degree identification and time series prediction compensation of fault degree,are mainly researched in this dissertation.Study on AUV thruster fault feature extraction and fusion.On extracting fault feature from AUV speed signal based on the modified bayes classification algorithm(MB algorithm),the fault feature value,the difference and ratio of the fault feature value and the noise feature value are both relatively small,so this dissertation proposes a wavelet MB algorithm.Different from the MB algorithm,which extracts fault feature from AUV original speed signal,the proposed method extracts fault feature from the wavelet approximate component of speed signal.On fusing the speed signal fault feature and the control signal fault feature based on the evidence theory algorithm,the two classification plane interval of fault feature curve and noise feature curve in the fusion fault feature is relatively small.So a normalized mapping method is proposed.Different from the evidence theory algorithm,which separates fault feature curve and noise feature curve in the fusion fault feature space,the proposed method transforms the fusion fault feature to a new feature space,and separates fault feature curve and noise feature curve in the new feature space.The effectiveness of the proposed methods is verified by pool experiments of the experimental prototype.Study on AUV thruster fault feature and external disturbance feature separation.On decomposing AUV speed signal into intrinsic mode functions(IMFs)based on ensemble empirical mode decomposition algorithm(EEMD algorithm),the intrinsic mode functionmatrix(IMFM)dimension is high,therefore a method combining wavelet decomposition with EMD is proposed.Different from the EEMD algorithm,which adds white noise to the speed signal,and then uses the EMD algorithm to decompose the preprocessed speed signal,the proposed method extracts the wavelet approximation component of speed signal based on wavelet decomposition,and then uses the EMD algorithm to decompose the wavelet approximation component.On extracting thruster fault feature from independent component based on the MB algorithm,both the difference and ratio of the fault feature value and the noise feature value are relatively small,therefore a wavelet detail components-assisted feature extraction method is proposed.Different from the MB algorithm,which extracts fault feature from independent component,the proposed method extracts fault feature from independent component added with wavelet detail components of the original speed signal.The effectiveness of the proposed methods is verified by pool experiments of the experimental prototype.Study on AUV thruster fault degree identification: When fusion feature modulus maximum method is used to extract thruster fault feature from speed signal and control signal,the extracted fault feature is non-monotonic to fault degree,so a peak region energy-based fault feature extraction method is proposed.Different from the fusion feature modulus maximum method,which regards the maximum of the fusion fault feature as fault feature,the proposed method convolutes the fusion fault feature to transform the fusion fault feature to energy distribution,and regards the maximum of peak region energy in the energy distribution as fault feature.When the grey relational analysis algorithm(GRA algorithm)is applied to estimate fault degree,the estimation results are none but standard fault degrees.To further get an estimation result between adjacent standard fault degrees,a least square grey relational grade-based fault degree estimation method is proposed.Different from the GRA algorithm,which chooses the standard fault degree corresponding to the biggest grey relational grade(GRG),the proposed method determines the fault degree interval according to the highest and second highest GRG,and then estimates fault degree in the fault degree interval according to the mapping relationship between GRG and fault degree.The effectiveness of the proposed methods is verified by pool experiments of the experimental prototype.Study on time series prediction compensation of fault degree.On establishing the fault identification model based on the fuzzy support vector domain description method(FSVDD method),the relative fitting error of the model to the true fault degree is large.So this dissertation proposed a homomorphic membership function method.Different from the FSVDD method,which uses the fault feature value in time domain to calculate fuzzymembership degree coefficient,the proposed method uses the fault feature value in log domain to calculate fuzzy membership degree coefficient.On identifying fault degree by the FSVDD method,the identification result lags behind the true fault degree.So a low frequency trend prediction method is proposed.Different from the FSVDD method,which extracts fault feature from current speed signal and control signal,the proposed method firstly forward predicts speed signal and control signal,and then extracts fault feature from the prediction speed signal and control signal.The effectiveness of the proposed methods is verified by pool experiments of the experimental prototype.