Sensor and actuator placement in dynamic processes is an engineering
task that has fallen between the cracks of contemporary chemical engineering
design fields. Steady state process design, which focuses heavily on capital
costs, typically does not address process dynamic. On the other hand, modern
control system design, which addresses dynamic performance, typically assumes
that all process hardware has already been selected. However, researchers
from both fields agree that inappropriate selection of these hareware devices
will have a significant impact on overall process performance.
In the current project a new dynamic performance criteria, based on
the theory of covariance analysis, will be used to identify sensor and
actuator configurations that achieve pre-specified bounds on closed-loop
performance at a minimum capital cost. The main advantage of the proposed
dynamic performance criteria is that the performance bounds, which need
to be specified by the process engineer, are with respect to commonly used
statistical process control parameters and are thus readily available for
most processes. The proposed solution method applies Linear Matrix Inequality
(LMI) techniques to an original Mixed Integer NonLinear Program (MINLP),
resulting in a computationally efficient Mixed Integer Convex Program (MICP).
As an illustration of the method, an example will be presented addressing
the problem of sensor and actuator placement in a Non-Isothermal Plug Flow
Reactor. |