Skip to Main content Skip to Navigation

Méthodologies de réglage automatique temps-réel de lois de pilotage

Abstract : This thesis deals with the tuning of flight control laws for civilian carrier aircraft, during its development phase. Due to the limited representativeness of the models used for the initial tuning, flight control laws may require several flight test iterations to achieve satisfactory performance. It is a long and costly process and the aircraft manufacturer thus wants to shorten it. As we have noticed that flight test data are less used than models, we have designed methodologies allowing a better use of these data for a control law tuning purpose. An aircraft being a parameter varying system, we must keep in mind that these data are valid in a small region of the flight domain only. As a consequence, the modifications made to the control laws, interpolated using parameters of the flight domain, must be passed on solely in this region. We considered two cases : an offline retuning work (i.e. performed in design office) and an online tuning (i.e. directly on the aircraft during a test flight). Both the notions of interpolation and local data made us choose a local learning method in order to interpolate the control laws. This method is a form of neural networks known as “mixture of experts”. Its structure will allow us to divide the flight domain into several sub-domains, to which we will allocate a local model. Therefore, we designed a generic methodology allowing to generate data to be interpolated, to define the mixture of experts structure and to perform a learning task on it. The data generation is a significant part of this methodology and considering the interpolation of control laws, it is not trivial. We thus propose an inversion based synthesis method for feedforward control laws. The mixture of experts structuring consists in choosing an optimal meshing of the flight domain, selecting the type of local model and managing the weightings applied to each local model. The methodology has been tested on two applications, pointed out by Airbus as reflecting the issue at stake. The first one deals with the estimation of the aircraft aerodynamic sideslip and the second one deals with a feedforward control law compensating for the side-effects yielded by the airbrakes extension. These examples show how the proposed methodology helps systematizing a tuning process which has been until then tedious.
Document type :
Complete list of metadatas
Contributor : Josiane Dartron <>
Submitted on : Tuesday, November 10, 2009 - 9:51:52 AM
Last modification on : Monday, December 14, 2020 - 12:38:04 PM
Long-term archiving on: : Thursday, June 17, 2010 - 7:53:16 PM


  • HAL Id : tel-00430820, version 1



Lilian Ronceray. Méthodologies de réglage automatique temps-réel de lois de pilotage. Automatique / Robotique. ISAE, 2009. Français. ⟨tel-00430820⟩



Record views


Files downloads