Detection of oscillatory actuator failures in passenger airliners

[摘要] ENGLISH ABSTRACT: This project investigates and develops techniques to detect oscillatory failure cases (OFCs) inaircraft control surface actuators. Oscillatory failures induce additional loads on the structure ofthe aircraft, requiring additional structural support to withstand these loads, increasing the overallmass of the aircraft. If oscillatory failures can be detected and pacified quickly, then the additionalstructural support would not be required, and the mass of the aircraft can be reduced, resultingin improved fuel efficiency and aircraft performance.Oscillatory failure case (OFC) detection is performed by evaluating the difference (residual) betweenthe measured behaviour of the real actuator and the simulated behaviour of a fault-free analyticallyredundant actuator model running in parallel with the real actuator. An OFC detection systemmust generate a residual signal using the analytically redundant actuator model, and evaluatethe residual signal to determine whether an oscillatory failure is present. The challenge for theresidual evaluation stage is to distinguish between the components of the residual signal resultingfrom modelling uncertainty and sensor noise, and the components resulting from an actual oscillatoryfailure case. The OFC detection system must detect oscillatory failures within a maximumallowable detection time, but must not produce false alarms.Five different oscillatory failure detection techniques are investigated and developed, namely oscillationcounting, integrated absolute error (IAE), discrete Fourier transform (DFT), multi-windowFourier transform (MWFT), and phase-locked loop (PLL) detection. Oscillation counting is anexisting OFC detection technique that was developed by Goupil [1] and is currently in service onthe Airbus A380 passenger airliner. The other four techniques are new OFC detection techniquesthat are developed in this project.A simulation framework is created to serve as a testbed for the training and testing of the differentOFC detection techniques. The simulation framework contains models for the physical actuator,the analytically redundant actuator, the oscillatory failures (both liquid and solid failures), theflight control system, and the aircraft longitudinal dynamics. The simulation models the aircraft'sresponse to an oscillatory failure, since it affects the performance of the OFC detection.The five OFC detection techniques are trained and rigorously tested using training and testingdata generated with the simulation framework. The detection thresholds for each technique are'trained on fault-free data to determine the lowest detection thresholds that do not producefalse alarms. The detection techniques are then tested using testing data to determine the smallestamplitude oscillatory failure that each technique can detect within the specified maximum allowabledetection time. The number of false alarms for each technique is also determined.The results show that DFT, MWFT, and the PLL outperform oscillation counting and IAE bydetecting smaller amplitude oscillatory failures and with shorter detection times, with MWFTproviding the most promising results. However, oscillation counting and IAE are the most computationallyefficient techniques, while DFT, MWFT, and PLL are more computationally expensive.Overall, the multi-window Fourier transform (MWFT) technique is the recommended approach forOFC detection, offering the best detection performance with only a small increase in computationalcomplexity.

[发布机构] Stellenbosch University