Research On Fault Prediction Diagnosis Of AUV Thuster Based On SVR Method

As the land of renewable resources is dwindling,and countries in the world realize the importance of the development of marine resources.Extracting marine resources need advanced technology and equipment.As the only equipment of the underwater vehicle can work in the sea,it gradually attracts people's attention.Autonomous Underwater Vehicle works in the complex underwater environment independently.Security is one of the important issues to study in the process of AUV development and application.Fault Diagnosis technology is the basic technology to ensure the safety of AUV.The propeller is the highest use frequency,the heaviest work load components in AUV,and it's also the most common source of failure of the underwater vehicle.According to the characteristics of the underwater vehicle's work environment and its strongly nonlinear systems,studying the propeller fault diagnosis technology has important research significance and practical value for improving the safety and intelligent level of the underwater vehicle.This paper aims at the problems of the fault prediction of the thruster fault,and it conducts research respectively in four areas:fault diagnosis and prediction scheme;state quantity fault feature extraction method;control the amount of fault feature extraction method;fault feature prediction model method.AUV prediction of fault diagnosishas been researched.Because of the AUV work in ocean environment,easily affect by some random disturbance such as the current,and AUV itself has strong nonlinear,it is difficult to establish fault prediction model of AUV accurately,which make the traditional fault diagnosis technology based on AUV prediction model in the AUV thruster fault diagnosis is limited by a lot.For this problem,In this paper,based on support vector regression(Support Vector Regression,the SVR),and the empirical mode decomposition algorithm combines fault prediction diagnosis overall schemeI.In view of the AUV online data set,extract variety characteristic values of fault characteristics and establish AUV thruster fault prediction diagnosis model.Through "the beaver-?" propeller AUV simulation output fault diagnosis,verify the effectiveness of the scheme.The thruster fault feature of the AUV state quantity has been researched.Traditional extraction fault feature of state quantity based on fractal dimension and modified Bayesian method consists of random disturbance caused by sensor noise,eigenvalue of which may be greater than fault feature eigenvalue to lead to error results of fault diagnosis.To solve this problem,this paper proposes localization algorithm,fractal dimension the extraction method of fault feature from the state quantity based on mutation(Short-time Higher Frequency Component,SHFC)of empirical mode decomposition,frequency or magnitude difference.In view of "Beaver-?" AUV thruster fault simulation pool experimental data and by comparing the proposed method with the traditional modified Bayesian and fractal dimension method in fault feature extraction effect,the superior effect of this paper's solution is verified,Traditional fractal dimension often requires a large amount of data to reflect the system's internal law;however,there are less real samples in the process of AUV real-time operation.Due to the lack of data,the size of eigenvalue from extraction fault feature would be affected.For this problem,this paper puts forward to improved program of adding constant samples to fractal dimension,through fault feature actual extraction effect of experimental data,to validate fault feature reinforcing effect in this paper's improved method.In the experimental study,it was found that embedding dimension is introduced in the process of using fractal dimension to extract fault feature;as a result,the fault feature extraction effect depends on the selection of embedding dimension.Aiming at the problem of selecting the embedding dimension,this paper uses the direct experimental method to select it.The thruster fault feature of AUV control mass has been researched.Traditional fractal dimension method is a kind of fault feature extraction methods,itself does not have the fault recognition,and itself also has large amount of calculation and calculation time long defects.For this problem,This paper proposes the method to establish multiple known fault degree of ideal control data characteristics of SVR model fault feature extraction and recognition.In view of AUV online data collection,according to multiple sets of feature extraction of SVR model output,through the screening effect of feature extraction to accomplish AUV thruster control the fault feature extraction and the degree of recognition.This article is based on Beaver-II propeller AUV simulation output fault basin experimental data,by comparing the improved fault feature extraction method with fractal dimension method,verify the improved method of superior effect on the function.AUV fault feature prediction model has been researched,facing to the problem that AUV fault feature prediction model is too hard to be established,this paper took the control variable,state variable and the fusion fault feature of these two variables as input and output.After that,SVR fault feature prediction model was established,and the thruster fault can be detected real-timely by analyzing errors between output signals of this model and the raw signals.Based on pool-experiment data of Beaver-II AUV thruster fault,the on-line diagnosis performance was verified,because the results of fault prediction rely on the inputs and prediction steps when establishing the model.Due to this problem,the comparative experiments whether there were feedbacks,and the comparative experiment between single-step prediction and multi-step prediction has been done based on the specific pool-experiment data of Beaver-II AUV thruster fault.