Fault Diagnosis For Gyroscope Based On Incremental Fuzzy Support Vector Machine

Gyroscope(gyro) is an important sensor of measuring attitude information for unmanned aerial vehicle(UAV), whose working stability and reliability directly affects the flight safety of aircraft. The paper, according to the characteristics of the gyro faults, puts forward a fault diagnosis system for gyro based on incremental fuzzy support vector machine. The main works of this paper are as follows:First of all, the paper attempts to solve the problem of unbalanced sample and noise interference for gyro output. On one hand, fault samples of gyro are limited, so there is imbalance problem between normal samples and fault samples, On the other hand, gyro is vulnerable to interference of airborne environmental, and gyro output often contains a lot of noise signals, which easily affects the accuracy of fault diagnosis. To this end, the problem of unbalanced sample and noise is solved by designing an improved fuzzy support vector machine, whose fuzzy membership function is designed by the combination of unbalanced sample characteristic factor and denoising fuzzy factor.Secondly, in order to realize the collecting samples and learning algorithm of synchronization, the paper proposes an incremental learning algorithm based on fuzzy support vector machine. When new samples are added to training, the initial samples filtered by traditional incremental learning algorithm tend to have a lot of redundancy. Two conditions that non support vectors may be converted to support vectors are obtained by the analysis: one is distributed in the class boundary; the other is close to heterogeneous class. A kind of double weighting function based on distance between samples and the center of homogeneous and heterogeneous is proposed to screening important information of the samples. Then combined with the KKT conditions, it realizes the online incremental learning process of fuzzy support vector machine.Thirdly, the paper builds a gyro fault diagnosis system based on fuzzy support vector machine and incremental learning algorithm, whose performances are verified by digital simulation in Matlab environment. Rate gyro is used as research object of experiments, it, according to the characteristics of the fault, achieve four kinds of fault s of total, bias, drift and periodic interference of through the simulation. The gyro signal is extracted by wavelet packet transform to obtain energy characteristics with the use of input of support vector machine classifier, then the validity and rationality of fuzzy support vector machine and incremental learning algorithm is respectively verified by experiments.Finally, the paper builds a hardware platform for gyro fault diagnosis system, which takes TMS320F28335 as the core of system. It transplants the classification model trained to DSP based on the simulation of fuzzy support vector machine. Furthermore, feature extraction, normalization and classification of signals are realized in DSP, which has completed the hardware realization of gyro fault diagnosis system.The experimental results of digital simulation and DSP hardware simulation show that the gyro fault diagnosis system not only improves diagnostic accuracy of fault samples, and reduces the false alarm rate, but also meets the real-time requirements, which has a certain practical application value.