Online system identification for fault tolerant control of unmanned aerial vehicles

Appel, Jean-Paul (2013-03)

Thesis (MScEng)--Stellenbosch University, 2013.

Thesis

ENGLISH ABSTRACT: In this thesis the strategy for performing System Identification on an aircraft is presented. The ultimate aim of this document is to outline the steps required for successful aircraft parameter estimation within a Fault Tolerant Control Framework. A brief derivation of the classical 6 degree-of-freedom aircraft model is firstly presented. The derivation gives insight into the aircraft dynamics that are to be used to estimate the aircraft parameters, and provides a basis for the methods provided in this thesis. Different techniques of System Identification were evaluated, resulting in the choice of the Regression method to be used. This method, based on the Least-Squares method, is chosen because of its simplicity of use and because it does not require as much computational time as the other methods presented. Regression methods, including a recursive algorithm, are then applied to aircraft parameter estimation and practical considerations such as Identifiability and corrupted measurements are highlighted. The determination of unknown measurements required for System Identification of aircraft parameters is then discussed. Methods for both estimating and measuring the Angle-of-Attack (AoA) and angular accelerations are presented. The design and calibration of an AoA probe for AoA measurements, as well as a novel method that uses distributed sensors to determine the angular accelerations is also presented. The techniques presented in this thesis are then tested on a non-linear aircraft model. Through simulation it was shown that for the given sensor setup, the methods do not provide sufficiently accurate parameter estimates. When using the Regression method, obtaining measurements of the angle-of-attack solely through estimation causes problems in the estimation of the aerodynamic lift coefficients. Flight tests were performed and the data was analyzed. Similar issues as experienced with estimation done on the non-linear aircraft simulation, was found. Recommendations with regards to how to conduct future flight tests for system identification is proposed and possible sources of errors are highlighted.

AFRIKAANSE OPSOMMING: In hierdie tesis word die strategie vir die uitvoering van Stelsel Identifikasie op 'n vliegtuig aangebied. Die uiteindelike doel van hierdie document is om die stappe wat nodig is vir 'n suksesvolle vliegtuig parameter beraming, binne 'n Fout Tolerante Beheer Raamwerk, uit eente sit. 'n Kort afleiding van die klassieke 6 graad-van-vryheid vliegtuig model word eerstens aangebied. Die afleiding gee insig in die vliegtuig dinamika wat gebruik moet word om die vliegtuig parameters te beraam, en bied 'n basis vir die metodes wat in hierdie tesis verskyn. Verskillende tegnieke van Stelsel Identifikasie is geëvalueer, wat lei tot gebruik van die regressie-metode. Hierdie metode is gekies as gevolg van sy eenvoudigheid en omdat dit nie soveel berekening tyd as die ander metodes vereis nie. Regressie metodes, insluitend 'n rekursiewe algoritme, word dan toegepas op vliegtuig parameter beraming en praktiese orwegings soos identifiseerbaarheid en korrupte metings word uitgelig. Die bepaling van onbekende afmetings wat benodig is, word vir Stelsel Identifisering van die vliegtuig parameters bespreek. Metodes om die invalshoek en hoekige versnellings te meet en beraam, word aangebied. Die ontwerp en kalibrasie van 'n invalshoek sensor vir invalshoek metings, sowel as 'n nuwe metode wat gebruik maak van verspreide sensore om die hoekversnellings te bepaal, word ook aangebied. Die tegnieke wat in hierdie tesis aangebied is, word dan op 'n nie-lineêre vliegtuig model getoets. Deur simulasie is dit getoon dat die metodes vir die gegewe sensor opstelling nie voldoende akkurate parameters beraam nie. Dit is ook bewys dat met die gebruik van die Regressie metode, die vekryging van metings van die invalshoek slegs deur skatting, probleme in die beraming van die aerodinamiese lug koëffisiente veroorsaak. Die tegnieke wat in hierdie tesis verskyn, word dan op werklike vlug data toegepas.Vlugtoetse is uitgevoer en die data is ontleed. Aanbeveling met betrekking tot hoe om toekomstige vlug toetse vir Stelsel Identifikasiete word voorgestel, en moontlike bronne van foute word uitgelig.

Please refer to this item in SUNScholar by using the following persistent URL: http://hdl.handle.net/10019.1/80125
This item appears in the following collections: