Process safety, ensuring product quality, meeting environmental standards all strongly depends on the set of variables we select for measurements. This selected set of variables play a vital role in any chemical plant and the selection process itself is a difficult task due to technical infeasibility and high cost of measuring instruments. Also the number of possible combination of variables is very large. Therefore, we formulate it as an optimization problem.
Process industries try to operate close to an economically optimal point but uncertainty in the form of disturbances, modeling error and measurement errors prevents the implementation of optimal policy and therefore results in loss. This loss is defined in terms of an economic cost quantity as "l oss in operational profit". We try to choose the set of measured variables (sensors) such that the above defined loss is minimum and we operate the plant at the near optimal point. Therefore, our objective is to select those set of measured variables th at incur minimal loss in the presence of uncertainties.
Measured variables are chosen to monitor, control, detect and diagnose faults. Sensors are selected to satisfy these objectives individually but in reality only one sensor network (set of measured variables) can be implemented. Therefore, our second objective is to come with the strategy to obtain a global sensor network that satisfies all the above mentioned objectives.