Research on gyro fault diagnosis method based on wavelet packet decomposition and multi-class least squares support vector machine

Abstract

In order to diagnose the faulty gyro, a method of fault diagnosis for the genus gyro based on wavelet packet decomposition and least squares support vector machine is proposed. Four kinds of gyro fault signal models are established by studying the characteristics of the output signals in the gyro fault state. The wavelet packet decomposition is used to perform time–frequency domain decomposition of gyro signals in various states, and the wavelet packet energy entropy value in each frequency band is calculated as a feature vector for characterizing the gyro fault state. The “one-to-many” multi-class least-squares support vector machine is used to identify the feature vectors in one state, and the parameters of LSSVM are optimized by using firefly algorithm and tenfold cross-validation method, respectively. The results show that the two methods are The diagnostic accuracy rate of the gyro fault data is more than 90%, and the calculation speed is fast, and the fault diagnosis of the gyro is successfully realized.