The Application Of Wavelet And Multi-kernel SVM In UAV Sernsors Fault Diagnosis

A variety of sensors in unmanned aerial vehicle(UAV) are an important part. Sensor technology,signal processing techniques and development of computer technology have laid a solid foundationfor diagnosis techniques. To avoid occurrence of fatal accidents and improve the safety and reliabilityof the aircraft, it can discovery faults by means of fault prediction and diagnosis. But the UAV sensorsfault diagnosis is a typical small sample learning problems. Fault of Sensors has certain suddenness,and shall not repeat or simulation. So the available fault sample size is often quite limited. Hence, thispaper studied the small sample of the fault diagnosis system. In view of the small sample problem,this paper studies a kind of system of sensors fault diagnosis that based on wavelet packet andmulti-kernel support vector machine. In this paper, the main work is as follows:Firstly, this paper analysis and classify the usual fault of rate gyro in Unmanned aerial vehiclerate gyro, summed up the common types of fault, and summarized the methods of fault diagnosis,concluded the limitations of traditional fault diagnosis methods. At the same time, it studies thecurrent commonly used fault diagnosis methods. When the traditional fault diagnosis methods solvethe finite sample, it occurs through learning and locally optimal solution.Secondly, analysis of the signal energy distribution and the energy of the wavelet packettechnique is utilized to extract the signal feature. The sensors failure occur and the energy distributionof signals in different frequency changes. This paper uses wavelet package decomposition to extractthe energy of signal in different frequency. It uses energy characteristics to train support vectormachine classifier, achieve the purpose of fault diagnosis and judging the type of fault.Thirdly, the design of Fault diagnosis system includes single-kernel SVM and multi-kernelsupport vector machine. In practical applications, the unreasonable parameters of support vectormachine can lead to appear larger error and accuracy lower. Diagnosis system cannot be used. In theprocess of SVM mathematical derivation, parameter determines the complexity of the algorithm andthe position determines the optimal classification. So this article uses cross validation method tooptimize the parameters of SVM. In this paper, according to the shortcoming of a single kernel SVMlacking of explanatory for complex sample data.So multi-kernel learning method was proposed, theidea is to linear combination of many kernels instead of the single kernel. In the design ofmulti-kernel function, in order to make the kernel function is more able to adapt to the actual sampledata, using real number coding genetic algorithm optimization of linear combination weight coefficient.Finally, in the Matlab/Simulink environment, it studies the death fault, shock fault, multiplicativefault, bias fault of rate gyro in VAV to verify the rationality and correctness of the design scheme.The simulation results show that: the paper designed the fault diagnosis system based on thewavelet packet and multi-kernel support vector machine can effectively extract the fault features andidentify the faults. Comparing with the single kernel support vector machine, diagnostic accuracyincreased by10%. It embodies the advantages of multi-kernel support vector machine. At the sametime proved that the support vector machine can effectively solve the problem of small sample.