Research On Weak Thruster Fault Detection And Prediction Method Of Underwater Vehicle

Unmanned cableless Autonomous Underwater Vehicle(AUV)works in a complex and changeable marine environment.Security is the primary consideration of doing AUV research,Propeller is the core part of load system,its fault diagnosis technology is of great importance for improving the security of AUV.Weak propeller fault features are faint and have relatively lower discrepancy compared with marine external random noises.Currently,there is no mature theory and agreed-upon solution.However,researches on weak propeller fault feature diagnosis theory and its application are of great significance and high practical values for improving the security of AUV and its development process.With respect to the weak propeller fault feature diagnosis of AUV,the research is mainly conduct form following four aspects,which are the augmentation of propeller features,the diagnosis of weak propeller fault features,the identification of propeller faults and the prediction of weak propeller fault feature variation magnitude tendency:With respect to over-attenuation caused by wavelet decomposition,when attenuating the external stochastic disturbance for longitude velocity of AUV,a novel algorithm was proposed to enhance fault feature and attenuate stochastic disturbance for AUV longitude velocity.The stochastic resonance of longitude velocity was obtained based on adaptive stochastic resonance.The energy of external disturbance was transmitted to fault signal,thus enhanced the fault features and suppressed the external random disturbance simultaneously.The pool experiment results tested the validation of the proposed method.Based on experiment data from outputs of bi-stable system driven by the aperiodic longitude velocity and periodic signal,respectively,the motion state of Brownian particle driven by the longitude velocity was analyzed.The analysis shows that: Transition movement between traps was not appeared when the Brownian particle drove by the longitude velocity of AUV,just moved in negative potential well.The pool experiment results tested the validation of the proposed method.Fault feature could be enhanced for the general propeller fault in AUV based on bi-stable stochastic resonance method integrating with particle swarm optimization,but it cannot satisfy the detection requirement of weak thruster fault.As for this problem,the paper makes an improvement in the bi-stable stochastic resonance method.Different form bi-stablestochastic resonance method,the paper transforms the bi-stable potential function into mono-stable one by constructing mono-stable stochastic resonance system,further improved the reinforcing effects.In the process of optimizing parameters of mono-stable stochastic resonance system driven by AUV signal,the global convergent time would be long when particle swarm optimization is adopted to adjust the parameters.In order to reduce the global convergent time,the paper improves the linear inertial weighted function as nonlinear one by adding cosine function in the inertial weighted function.The pool experiment results tested the validation of the proposed method.Wavelet reconstruction can be problematic when applied to AUV fault detection,the reconstruction of external interference values may outrun weak fault feature reconstruction values,thus make it impossible to make a right prediction.The paper develop an approach based on stochastic resonance combined with wavelet reconstruction method for AUV propeller weak fault detection.Different form the traditional way of wavelet reconstruction which detects fault by decomposing and reconstructing the wavelet with highest decomposition scale,the method in this paper is based on calculate the wavelet entropy of all decomposition scale and then find the optimal reconstruction scale though wavelet entropy,based on that optimal scale and then detect the weak propeller fault features.The pool experiment results tested the validation of the proposed method.There exist some problems when the fractal feature method is applied to identify thruster faults for AUV.Sometimes it could not identify the thruster fault,or the identification error is large,even the identification results are not consistent for the repeated experiments.The paper analyzes the reasons resulting in these above problems according to the experiments on AUV prototype with thruster faults.On the basis of these analyses,in order to overcome the above deficiency,an improved fractal feature integrated with wavelet decomposition identification method is proposed for AUV with propeller fault.Different from the fractal feature method where the signal extraction and fault identification are completed in the time-domain,the paper makes use of the time-domain and frequent-domain information to identify propeller faults.In the paper,the propeller fault could be mapped multi-source and described redundantly by the fault feature matrix constructed based on the time-domain and frequent-domain information.In identification phase,different from the fractal feature method where the fault is identified based on fault identification model,the fault sample bank is builtat first in the paper,and then pattern recognition is achieved by calculating the relative coefficients between the constructed fault feature matrix and the elements in the fault sample bank.The pool experiment results tested the validation of the proposed method.Applying traditional GM(1,1)method to predict the magnitude of weak fault variation exists some problems,for instance gray background have deviations,and white differential equation bias and the prediction results are unable to amend.This paper presents an approach to improve the performance of grey forecasting model.Different form typical GM(1,1)method,which generate gray background by equal weight of neighbor sequence calculated though once accumulation,the paper computes the neighbor integration of accumulative generated sequence,and assign it as the grey background value;The traditional way of solving white differential equation of grey differential equation is assign original sequence as the initiate value,this paper adopts the minimum error of original sequence as the initiate value of the solution of differential equation;Different form typical GM(1,1)method which has no correctability,the paper developed a re-prediction method based on the comparison of the predicted sequence and the original sequence.Therefore,the result can be rectified based on the comparison.The pool experiment results tested the validation of the proposed method.