(756b) Stochastic Programming Approach Vs. Estimator-Based Approach for Sensor Network Design for Maximizing Efficiency

The measurement technology plays a key role in achieving efficient operation of the process plants without violation of the environmental and operational constraints. Inaccuracies in the measurements of the controlled variables can lead to inefficient operation (or even unsafe operation under extreme conditions) of the plant. Inaccuracies in the measurements of other variables that are used for monitoring purposes can lead to undesirable conditions and can damage equipment and reduce their life. However, it may be difficult, if not impossible to measure a number of variables or the measurements may be noisy, inaccurate, or unreliable. Furthermore, the measurement technology for measuring certain variables may be costly or the measurement is available after undesired delay. It may be possible to estimate these variables accurately by optimally located sensors. For a large-scale process plant, it is difficult to decide which variables should be measured for maximizing the plant efficiency while satisfying desired estimation accuracy in other variables. With this motivation, the focus of the current work is on development of sensor placement (SP) algorithms to obtain the numbers, locations, and types of sensors for a large-scale process with the estimator-based control system.

Two SP algorithms are developed and investigated. In one algorithm (Paul et al., 2015, 2016), dynamics in the process efficiency loss due to the estimator-based control system that receives measurements from a candidate sensor network are explicitly accounted for. For a large-scale process with large number of candidate sensor locations, this approach leads to a computationally expensive mixed integer nonlinear programming problem. A number of novel approaches is developed by modifying the SP algorithm, problem formulations, and computational approaches to make the solution tractable. In another algorithm, the estimation error is accounted for in terms of probability distributions and therefore, a stochastic programming approach is used to solve the SP problem (Sen et al., 2016). A novel algorithm called BONUS to solve the problem (Diwekar, 2015).

The developed SP algorithms are implemented in an acid gas removal (AGR) unit as part of an integrated gasification combined cycle (IGCC) power plant with pre-combustion CO­₂ capture. The high-fidelity plant model is developed in Aspen Engineering Suite. The model has several thousand states and hundreds of candidate sensor locations. In this presentation, we will compare and contrast these two SP algorithms by evaluating the efficiency loss of the optimal sensor network synthesized by each of these algorithms along with their computational performance.

References:

1. Paul et al., Sensor Network Design for Maximizing Process Efficiency: An Algorithm and Its Application, AIChE Journal, V. 61, pp 461-476, 2015.
2. Paul P, Bhattacharyya D, Turton R, Zitney S “Dynamic Model-Based Sensor Network Design Algorithm for System Efficiency Maximization”, Computers & Chemical Engineering, 89, 27-40, 2016
3. Diwekar U. and A. David, BONUS Algorithm for Large Scale Stochastic Nonlinear Programming Problems, Springer, 2015.
4. Sen P., K. Sen, and U. Diwekar, A multi-objective optimization approach to optimal sensor location problem in IGCC power plants, Applied Energy 181, pp 527–539, 2016.