On the synthesis of switched output feedback controllers for linear, time-invariant systems

Abstract

The theory of switching systems has seen many advances in the past decade. Its beginnings were founded primarily due to the physical limitations in devices to implement control such as relays, but today there exists a strong interest in the development of switching systems where switching is introduced as a means of increasing performance. With the newer set of problems that arise from this viewpoint comes the need for many new tools for analysis and design. Analysis tools which include, for instance, the celebrated work on multiple Lyapunov functions are extensive. Tools for the design of switched systems also exist, but, in many cases, the method of designing stabilizing switching laws is often a separate process from the method which is used to determine the set of vector fields between which switching takes place. For instance, one typical method of designing switching controllers for linear, time-invariant (LTI) systems is to first design a set of stabilizing LTI controllers using standard LTI methods, and then design a switching law to increase performance. While such design algorithms can lead to increases in performance, they often impose restrictions that do not allow the designer to take full advantage of the switching architecture being considered.

(cont.) For instance, if one switches between controllers that are individually stabilizing (without any switching), then, effectively, one is forced to switch only between stable systems and, hence, cannot take advantage of the potential benefits of switching between unstable systems in a stable way. It is, therefore, natural to wonder whether design algorithms can be developed which simultaneously design both the set of controllers to be switched and a stabilizing switching law. The work investigated here attempts to take a small step in the above direction. We consider a simple switching architecture that implements switched proportional gain control for second order LTI systems. Examination of this particular structure is motivated by its mathematical simplicity for ease of analysis (and, hence, as a means of gaining insight into the problem-at-large), but, as we will see, the design techniques investigated here can be extended to a larger class of (higher order, potentially non-linear and/or time-varying) systems using standard tools from robust control. The overall problem we investigate is the ability to create algorithms to simultaneously determine a set of switching gains and an associated switching law for a particular plant and performance objective.

(cont.) After determining a set of necessary and sufficient conditions for a given second order plant to be stabilizable via the given switching architecture, we synthesize an algorithm for constructing controllers for which the corresponding closed-loop system dynamics are finite L, gain stable. Also, in an effort to demonstrate that the the given structure can, in fact. he used to increase performance. we consider a step-tracking design problem for a class of plants, where we use overshoot and settling time of the output step response to measure performance. We compare the results obtained using our switching architecture to the performance that can be obtained via two other LTI controller architectures to illustrate some of the performance benefits.