Research On Fault Diagnosis Algorithm For Small UAV Sensor And Software Development

According to the specialty of powerful maneuverability, more faults and difficultto be detected for small UAV in real time, a combining method of least squaressupport vector machine(LS_SVM) and principal component analysis(PCA) isproposed for common fault diagnosis and signal reconstruction of small UAV angularrate sensors. In this paper, we take nonlinear6-degree-of-freedom Aerosonde UAVmodel as a research subject, then add fault models and fault diagnosis models insimulink environment. The purpose of this study is to ensure diagnosing fault rapidlyand completing signal reconstruction when small UAV angular rate sensors go wrong.It can make small UAV continue to fly in a secure performance range.The LS_SVM has the specialty of high fitting accuracy and rapid diagnosis. Itcan create a predictive model of sensors’ outputs. The model from off-line training isused for online prediction. The residuals between estimates and actual values are usedto detect fault. Angular rate sensor measurements used for feedback control,navigation calculations or monitoring the status of the system, output signals ofsensors have a certain correlation. It is difficult to separate fault signals for LS_SVM.This article uses the characteristics of sensitive to system abnormal data of principalcomponent analysis (PCA) which can separate fault signals. The combining methodof LS_SVM and PCA for small UAV sensor fault diagnosis can make up for theinadequacies of the each method. It makes the entire detection system run moreefficiently and accurately.According to the diagnosis result, the LS_SVM estimates will be the outputvalues of faulty sensor instead of the actual values. Simulation results show that thismethod has a very good effect on UAV sensor fault diagnosis; the precision of signalreconstruction is very high. It can guarantee the safe flight of UAV and prove that thismethod has high reliability and stability.