Multi-model approaches to robust control system design for air vehicles

Abstract

We usually do not know the exact mathematical model of aerodynamic mechanics over a large flight regime of air vehicles. Also, the effect of uncertainty may become highly nonlinear in a high-angle-of-attack region. To meet tight design specifications over this agility, a possible way is to employ one of the existing robust control techniques. However, in the lack of wide application experience, it remains a challenging problem to incorporate physical intuition into robust control system design. This thesis addresses the problem of designing a robust controller based on multi-model approaches to deal with aerodynamic uncertainty. The plant model used in controller design is characterized by a set of possible plants due to modeling error or complexity. The stability and performance of a designed controller can be measured in terms of the worst-case of a multi-model`s responses. Selection of proper models for the multi-model representation is handled by managing gap metrics between pairs of plants. The computation of gap metrics is used to limit the number of elements for a multi-model by measuring the stability margin between plants in respect to feedback control design. The problem of designing a robust controller can be transformed into the problem of finding a single controller that stabilizes a finite number of possible plants over the flight regime. In this context, two design methods are proposed by exploiting capabilities of the multi-model concept. In the first approach, a co-evolutionary algorithm is applied to find optimal gains of a fixed classical controller configuration. Competing specifications and constraints for the multi-model are represented by a cost function with inequality constraints. In the second approach, multi-objective control synthesis based on linear matrix inequalities is extended to controller design of the multi-model with modern control theory. Since various types of performance criteria for each model can be derived in the stac...