Robust Incremental Sensor-Based Control for Flexible Aircraft

La recherche d’une meilleure efficacité énergétique incite l’industrie aéronautique à utiliser des matériaux plus légers, mais plus flexibles. Cela accentue les interactions dynamiques entre les déformations élastiques des surfaces portantes et les forces aérodynamiques qu’elles génèrent, entraînant l’apparition de phénomènes aéroservoélastiques (ASE) indésirables qui perturbent les commandes de vol. L’approche courante repose sur le séquencement de gains avec des filtres coupe-bande, ce qui nécessite des modèles aérodynamiques précis et des réglages fins, engendrant des coûts croissants pour le développement des lois de commande. Dans ce contexte, cette thèse vise à établir un processus de conception des systèmes de commande de vol pour avions flexibles, en proposant l’inversion incrémentale de la dynamique non linéaire (Incremental Nonlinear Dynamic Inversion, INDI) comme alternative au séquencement de gains, et en exploitant la synthèse H∞ structurée pour optimiser de manière robuste et automatique les nombreux réglages requis. La commande INDI est d’abord analysée en profondeur afin d’en raffiner ses hypothèses simplificatrices en tests quantitatifs d’applicabilité, facilitant ainsi son utilisation en pratique. Cette analyse révèle que plusieurs hypothèses ne sont que suffisantes, mais non nécessaires. En particulier, l’architecture classique de INDI peut être reformulée sous une forme plus complexe intégrant un prédicteur de Smith. À la lumière de ces investigations, des tests quantitatifs sont proposés, accompagnés de suggestions de techniques de mitigation en cas d’échec. Ces contributions renforcent la maturité de la technologie étudiée et étendent son applicabilité à une plus grande classe de systèmes.

Abstract

In its quest to improve fuel efficiency, the aviation industry has increasingly turned to lightweight and flexible materials. This has led to greater dynamic interactions between the elastic deformations of the lifting surfaces and the aerodynamic forces they generate, giving rise to undesirable aeroservoelastic (ASE) phenomena that interfere with the aircraft’s controls. Given the common practice of using a gain-scheduled controller with specially designed notch filters, accurate aerodynamic models and precise tuning are required, leading to ever-increasing costs for control design development. In this context, this thesis aims to formulate an alternative control design process for flexible aircraft, centered around the Incremental Nonlinear Dynamic Inversion (INDI) approach as a replacement for gain scheduling, and the use of Structured H∞ Synthesis instead of manual fine-tuning. First, the INDI control scheme itself is investigated to refine its qualitative assumptions into quantitative applicability tests, which would significantly facilitate its practical use. This analysis reveals that several assumptions are sufficient but not necessary, as the classic INDI architecture can be framed as a simplification of a more complex architecture involving a Smith Predictor. Through these investigations, quantitative tests are proposed along with suggested mitigation techniques in case of failure, contributing to the improvement of the technology’s maturity and extending its applicability to a broader class of systems. Following the improvements in INDI control, the proposed design process is applied to a rigid aircraft. Quantitative tests are conducted to assess INDI applicability, and proposed mitigation techniques are implemented to compensate for the aircraft’s realistic (non-identical and non-minimum phase) actuators. Their performance is demonstrated through comparison with an unmodified version. The Structured H∞ Synthesis is employed to robustly tune the unique linear controller stage of the INDI control law across the aircraft’s entire flight envelope, thereby compensating for performance losses caused by the applied mitigations. Requirement fulfillment is evaluated through worst-case analysis and Monte Carlo campaigns, highlighting the benefits of both approaches and demonstrating how verification efforts could be enhanced by leveraging worst-case analysis tools.