A Research On Fault Diagnosis Technology Of Attitude Sensor

With technological progress and social development, the progress of scienceand technology and social development, the space industry becomes a symbol ofnational strength, is catching more and more people’s attention. Satellite attitudesystem as a satellite critical subsystems, because of the special environmentand the factor of the space, the failure probability is very high. And is often fatal.in order to effectively reduce and eliminate potential safety problems, ensurethe normal work of the satellite in orbit, improve the reliability of the spacecraft,we studied several appropriate fault diagnosis methods. The research work of thisthesis is embodied in the following aspects:First of all, After so much research, we introduce the related concepts,methods and research of fault diagnosis, analyze the present situation of researchat home and abroad.Secondly, according to the satellite attitude control principle blockdiagram,we set up the simulation model to obtain the sample data of faultdiagnosis. And also analyze the mathematical models of key components, faulttype characteristics and wher the difficulty lies. As discussed below, laying thegroundwork for fault diagnosis algorithm analysis.Then, aiming at the gyro sensor of single redundancy, according to themappingrelationship dynamics equations and gyro angular speed, using star-sensitive attitude quaternion output to get an estimate of gyro angular velocity,and then using the Kalman filter, comparing the estimated value and the truevalue to gain residuals,and taking sequential probability ratio method, and finallydetermining the fault with likelihood ratio,and also improving this method. Dueto the project requirements. Because of the needs of the project, according to thecharacteristics of gyro sensor with multi axis angular velocity measurement data, using the partial least squares method of the multivariate statistical method.The methods focus on input interpretation prediction effect, it can remove noiseand other interference better,and the PLS models have strong robustness. and canovercome both of linear dimension reduction. We can monitor process throughthe SPE statistical indicators, and put forward the index of fault detection factorbased weight, to achieve a variety of combinations of fault diagnosis fault typeand location.Finally, because of the wavelet’s good time domain and frequencydomain localization characteristic advantage,it can detect the mutations ofnormal signal well,and realize locating the fault precisely, we apply it tothe faultdiagnosis of star sensor without the need for modeling,and combiningthe Characteristics of the information entropy which can characterize thesignal’s energy trends,combinating the two reasonably can receive the gooddiagnosis effect.