Distributed Model-based Control for Gas Turbine Engines

Controlling a gas turbine engine is a fascinating problem. As one of the most complex systems developed, it relies on thermodynamics, fluid mechanics, materials science as well as electrical, control and systems engineering. The evolution of gas turbine engines is marked with an increase in the number of actuators. Naturally, this increase in actuation capability has also been followed by the improvement of other technologies such as advanced high-temperature and lighter materials, improving the efficiency of the aero engines by extending their physical limits. An improvement in the way to control the engine has to be undertaken in order for these technological improvements to be fully harnessed. This starts with the selection of a novel control system architecture and is followed by the design of new control techniques. Model-based control methods relying on distributed architectures have been studied in the past for their ability to handle constraints and to provide optimal control strategies. Applying them to gas turbine engines is interesting for three main reasons. First of all, distributed control architectures provide greater modularity during the design than centralized control architectures. Secondly, they can reduce the life cycle costs linked to both the fuel burnt and the maintenance by bringing optimal control decisions. Finally, distributing the control actions can increase flight safety through improved robustness as well as fault tolerance. This thesis is concerned with the optimal selection of a distributed control system architecture that minimizes the number of subsystem to subsystem interactions. The control system architecture problem is formulated as a binary integer linear programming problem where cuts are added to remove the uncontrollable partitions obtained. Then a supervised-distributed control technique is presented whereby a supervisory agent optimizes the joint communication and system performance metrics periodically. This online optimal technique is cast as a semi-definite programming problem including a bilinear matrix equality and solved using an alternate convex search. Finally, an extension of this online optimal control technique is presented for non-linear systems modelled by linear parameter-varying models