On the sensitivity of the control structure selection problem in large-scale multivariable systems

In complex manufacturing processes, control structure selection poses a serious challenge for the control engineer. Such systems are typically interacting, large scale and may comprise many loops. Choosing a correct control structure is not often a straightforward process, especially, if as it often is the case, one is faced with the challenge of having to integrate additional proposed control systems into an already existing plant-wide control system. Current approaches to the control structure selection problem rely either on the control engineer's experience and intuition or on solutions of difficult and non-convex combinatorial optimisation problems. A set of new computationally inexpensive graphical analysis tools based on basis pursuit regularisation methods are presented that provide significant insight into the control structure selection problem. The tools presented build a centralised controller by incrementally adding one loop at a time to the structure of a baseline decentralised controller. There are no constraints as to where new loops can be added which results in a rich class of sparse controllers. With the aid of an example, it is shown that how these tools are used directly to assess the local structural sensitivity of the problem and to identify key and important connections within the controller.