Research On Fault Diagnosis Methods For Gyroscope Based On Fuzzy Support Vector Machines

As an important airborne sensor, gyroscope is mainly used to accurately measure the attitude angle, angular velocity and heading of the vehicle, providing accurate flight information for flight control system. Its working condition directly affects the flight safety of the aircraft, so it is very important to guarantee and improve the reliability of the gyroscope. Aiming at the actual situation, the number of normal samples is often much larger than the number of fault samples, the paper mainly study in fault diagnosis system based on fuzzy support vector machine. The main work is as follows:First of all, several common fault model of the gyroscope are built, the wavelet packet decomposition method and empirical mode decomposition are respectively used to decompose the output signal of the gyroscope. When the gyroscope go wrong, the output signal will change, which reflected in its frequency is the energy of a certain or a few frequency bands change. So the energys of each signal component can be extracted as feature vector.Secondly, the traditional support vector machine is vulnerable to class imbalance and noise. After comparing and analyzing the impact of the imbalanced data sets on the support vector machine classifier, fuzzy support vector machine is used as the core algorithm of fault diagnosis system. The membership degree design of fuzzy support vector machine is divided into two parts, the first part is used to suppress the influence of sample quantity imbalance. The second part uses the Gauss probability density function to describe the distribution of sample set, as a basis for the importance of the sample. So considering the influence of noses, class imbalance and the distribution characteristic of samples set, the designed classifier has stronger robustness.Once more, due to the non-singularity of the fault type of the gyroscope, the recognition and localization of the normal state and fault type is not a two classification problem. In this paper, a new methode is proposed to sovle the exiting of unrecognizable region in one-against-one multi classification algorithm. The method makes use of the characteristics of one-against-one method and the data relationship between the unknown samples with each class to circularly screen, so as to get the final decision classification.Finally, according to the improved fuzzy support vector machine algorithm and a oneagainst-one multi classification method, using Libsvm software package, the program is written to simulate and validate the effectiveness and accuracy of the gyroscope fault diagnosis system on MATLAB environment. The result of the experiment are analyzed and summarized.