| In engineering applications like power systems and robotics, the systems consist of many interconnected subsystems. In some situations, due to fault or maintenance issues, these subsystems are disconnected and again connected (structurally perturbed) to carry out the required job. Moreover, during operation, when a disturbance affects the plant and causes variation in the process variables, namely, frequency of generators or angle of robot arms, most probably the operator adjusts the controller parameters to overcome this disturbance. In this situation, one can ask two practically important questions: "Is it possible to maintain overall stability under structural perturbations (connective stability) or when the controllers are detuned capriciously? How much of the performance loss of the system is due to these scenarios?";During the last three decades, many research efforts designed a controller for each subsystem in a plant, this is called decentralized design. The controllers were synthesized for stability under load fluctuations and for other objectives. However, due to complexity, there are few clear and effective design techniques for connective stability and/or for stability under arbitrary control loop detuning. Furthermore, every system possesses some degree of nonlinearity and structured controller design for nonlinear systems is an active area of investigation.;The approach of this thesis is to utilize mathematical tools, namely, structured singular values, properties of norms, different optimization techniques, transformation methods, and matrix operations to cast the design of controllers into a convex optimization framework (in some cases, quasi-convex) which is readily solvable by available numerical software. These algorithms are clear and computationally efficient. An additional goal is to develop new conditions for reducing the performance loss of a decentralized control system. The work here verifies the theoretical design algorithms developed in this thesis as applied to Syncrude control problems. This plant is highly nonlinear and the controllers presently working cannot overcome problems such as: (a) due to arbitrary tuning of the controllers and sudden load variations, the 900-pound header pressure undergoes oscillations which the controllers are incapable to damp out quickly and (b) controlling the load fluctuations requires large quantities of natural gases, a major economic concern. |
