(383c) Sensor Placement for Maximizing Process Efficiency: An Algorithm and Its Application | AIChE

(383c) Sensor Placement for Maximizing Process Efficiency: An Algorithm and Its Application


Paul, P. - Presenter, West Virginia University
Bhattacharyya, D., West Virginia University
Turton, R., West Virginia University
Zitney, S., National Energy Technology Laboratory

Process and energy plants are facing increasingly higher targets for efficiency. On the other hand, environmental emission standards are becoming tighter. Under these constraints, the process measurement network can play a key role. However, in a large plant, numerous state variables cannot be measured and of the thousands that can be measured, many may have low precision, reliability, or signal-to-noise ratio. Therefore, a number of key process state variables must be estimated satisfactorily to maximize plant efficiency while satisfying environmental and operational constraints. In this work, a sensor placement (SP) algorithm is developed for maximizing process efficiency without violating desired state estimation errors in key variables within a given budget.

The SP algorithm in this work is developed assuming an estimator-based control system where an optimal Kalman filter is used to estimate the states in presence of measurement and process noise. Due to the feedback loop in the control system, the resulting mixed integer nonlinear programming (MINLP) problem can be very difficult to converge for any arbitrary set of integer variables (i.e., set of sensors). Therefore, a sequential infeasible-path optimization algorithm is developed in which a ‘tearing’ approach is used to solve the feedback loops. Convergence of the tear loops is achieved using successive substitution, including acceleration methods. The MINLP problem is solved by a Genetic Algorithm (GA) subject to the linear and nonlinear equality and inequality constraints.

The developed SP algorithm is applied to an acid gas removal (AGR) unit as part of an integrated gasification combined cycle (IGCC) power plant with pre-combustion CO­2 capture. The nonlinear dynamic process model of the AGR unit developed in Aspen Plus Dynamics® is linearized around the nominal operating conditions and the resulting linear state-space model with more than 1400 states is used in the SP algorithm. In addition, more than 160 candidate sensor locations are considered. In this combinatorial optimization problem, thousands of equality constraints must be satisfied for each combination of sensors. For solving this large-scale problem, the GA-based optimization computations are parallelized using the Distributed Computing Server (DCS®) and the Parallel Computing® toolbox from Mathworks®.  In this presentation, we will present the development of the SP algorithm, its application to an AGR unit.  We will show how the AGR process efficiency is affected as the budget for the sensor network is changed. The impact of the sensor budget on the estimation accuracy of the key process variables in the AGR unit will also be presented.