Research On Fault Diagnosis Of AUV Thruster Based On Grey Qualitative Simulation

The autonomous underwater vehicle, as an important carrier of human exploration andexploitation of marine resources, gets more attention and is in rapid development with theaccelerated process of ocean development. Autonomous Underwater Vehicle (AUV), whichis unmanned and uncabled, works in the marine environment, and its security is one of theimportant issues in its research and practical process. Condition monitoring and faultdiagnosis is the basis and key technologies of the AUV security, and research on them hasimportant scientific significance and application value to improve the intelligence level ofthe AUV and speed up its practical course.This paper aims at the problems of AUV fault diagnosis that there are big errors via themathematical model method and low propeller fault degree identification accuracy by faultdiagnose decision tree-based method. This paper studied from the following three aspects ofAUV: construction of gray qualitative model; running of pattern recognition; as well as thepropeller fault degree identification.This paper studies on the AUV gray qualitative modeling methods. On one hand, sincethe complex marine work environment of the AUV, the nonlinear characteristics of itself andrandom disturbance from the sea currents make it difficult to establish precise analyticalmodel purely based on quantitative methods. On the other hand, pure qualitative simulationmethod ignored useful quantitative information of the AUV system in modeling process,which makes it more difficult to filter the predicted condition singular behavior branch.Therefore, the fault diagnosis of AUV based on pure quantitative methods or purequalitative methods usually appear with the big model error as well as misdiagnosis. Tosolve the upper problems, this paper combines qualitative with quantitative information, andpresents an AUV modeling method based on the gray qualitative simulation. In the method,through analyzing the relationship between variables in the system running process, thepossible successor state of the system is derived. Through probability gray number, higherderivative and the duration, the singular behavior branches of the successor state are filtered,and AUV gray qualitative model under normal and typical fault modes are constructed.Based on the experimental data acquired under these modes, the analysis on the consistency between the experimental data and that got by gray qualitative model is conducted, and thegray qualitative modeling method of AUV in this article is verified.This paper studies on the AUV running pattern recognition method. To solve theproblem that large amount of AUV observed variables decrease the online diagnosticaccuracy, this paper proposes a running pattern recognition method based on gray relationalanalysis. During this method, associated calculation on the online observation sequence withthe sequence of the gray qualitative model is conducted, and the running state of the AUV isidentified via discriminated association degree. For the weak ability of resistance to externalinterference in the AUV pattern recognition process based on gray relational analysismethod, a weighted average gray relational comprehensive evaluation method is proposed,in which the number of same qualitative states are integrated as weighted values into theAUV running the correlation calculation process, in order to increase system dynamicinformation. Corresponding correlation degree and comprehensive evaluation indexcalculation of qualitative pool experimental data with the gray qualitative model isconducted. By evaluating relation degree between actual operation and each modelcorresponding variables, the analytical method based on gray relational the validity of theAUV pattern recognition is verified. By comparing the actual operation with the modelevaluation index, the effectiveness of the AUV pattern recognition method based on theweighted average gray relational comprehensive evaluation is verified.This paper studies on AUV propeller fault degree identification methods. Since thereare too many kinds of observed variables, selection of observing variables, which containsfault degree, and modeling, have a big affect on fault degree recognition. To solve theproblem, this paper presents a selection method of the observing variables containing faultdegree information in the case of false, and gets characteristic variables associated with thefault degree via selection of the observing variables. For the lower identification accuracy ofpropeller fault degree identification via fault-based decision tree method, this paper presentsa propeller fault degree identification method combining three-dimensional surface fittingwith characteristic variables fractal box dimension calculation. During this method, viafractal box dimension analysis on characteristic variables of the fault that right mainpropeller lacking of force in the uniform direct running process, the of fault of in the force speed-fault degree-characteristics of variable fractal box dimension fault degreethree-dimensional model is constructed. During the uniform direct running process of theexperimental prototype at the speed of0.3m/s in the pool, the fault simulation of the rightpropeller contributing75%force, the experimental data is acquired, and characteristics arecalculated via variable fractal box dimension method. By substituting the results into theconstructed three-dimensional model of the fault degree, and comparing the recognitionresults with the setting fault degree, the validity of propeller fault degree identificationmethod of this paper is finally verified.