An analysis and comparison of two methods for UAV actuator fault detection and isolation

[摘要] ENGLISH ABSTRACT: Fault detection and isolation (FDI) is an important aspect of effective fault tolerant controlarchitectures. The Electronic System Laboratory at Stellenbosch University identified theneed to study viable methods of FDI. In this research two FDI methods for actuator failureson the Meraka Modular UAV are investigated.The Meraka Modular UAV is an unmanned aircraft that was developed by the CSIR. Asimple six degree of freedom non-linear mathematical model is developed that presents aplatform on which the two FDI methods are formulated. The theoretical model is used ina simulation environment to extensively test and compare the performance of the proposedFDI methods in different types of flight conditions.The first method investigated is a multiple model adaptive estimator (MMAE), which incorporatesa bank of Kalman filters. Each Kalman filter in the MMAE is conditioned foreach expected actuator fault scenario. The limitations of using linear Kalman filters are explainedand they are replaced by extended Kalman filters, whose associated advantages anddisadvantages are discussed. Each filter in the bank of Kalman filters produces a residualvector and residual covariance matrix. This information is subjected to a Bayes classifier todetermine the fault scenario which will have the highest likelihood of being active.The second method that is studied incorporates the parity space approach for FDI. Theparity space consists of the parity relations that quantify all the analytical redundanciesavailable between the sensors' outputs and actuator inputs of a system. A transformationmatrix is then optimised to transform these parity relations into residuals that are speciallysensitive to specific actuator faults. Actuator faults cause the parity space residuals' varianceto increase. A cumulative summation procedure is used to determine when the residuals'variance has changed sufficiently to indicate an actuator fault. A pseudoinverse actuatorestimation scheme is used to extract the actuator deflections from the parity relations.The FDI performance is tested by deliberately failing specific actuators of the Meraka ModularUAV in-flight. The flight test data is then used to analyse and compare the performanceof the two FDI methods investigated in the research. It is found that, for the specificMeraka Modular UAV, the FDI performs as expected with disturbance effects and actuatorexcitation influencing the FDI effectiveness. The research shows that the bank of Kalmanfilters creates less false alarms whereas the parity space FDI is more sensitive to faults. Itis illustrated that FDI can be improved with active actuator excitation and process noiseestimation techniques, delivering promising results.

[发布机构] Stellenbosch University