Design of Nonlinear Sensor Networks for Process Plants

Abstract

In this article we extend and generalize the ideas of two previous articles devoted to linear sensor networks to nonlinear systems. In those previous articles the use of cutsets and a decomposition procedure were proposed and proved efficient to solve large scale linear problems. In this article we show that a similar procedure, now based on a variable elimination scheme, can be also used efficiently for medium size nonlinear problems, but its computational efficiency for realistic large scale problems is not satisfactory. We also propose an alternative technique for the case of tree enumeration using instruments instead of equations that is very efficient for heavily instrumented flowsheets.

New efficient breadth‐first/level traversal tree search method for the design and upgrade of sensor networks

“…If tight bounds are not obtained (in this context, when current bests are found in the right side of the tree), the level traversal tree search basically has to explore the whole level of tree, which makes it impossible to solve large-scale problems efficiently using this method. The large-scale problem with medium level of specification shown in Nguyen and Bagajewicz, 14 the TE example case study 2, exposes such limitation of the level traversal tree search. We attempted to solve the TE example using level traversal tree search, but the solutions provided by this method are worse than the solution obtained by the equation-based method with decomposition described in Nguyen and Bagajewicz 14 even after several days running.…”

“…The large-scale problem with medium level of specification shown in Nguyen and Bagajewicz, 14 the TE example case study 2, exposes such limitation of the level traversal tree search. We attempted to solve the TE example using level traversal tree search, but the solutions provided by this method are worse than the solution obtained by the equation-based method with decomposition described in Nguyen and Bagajewicz 14 even after several days running. For the TE example, a kind of heuristic local search or approximate method is probably the most efficient method.…”

“…The cutset-based depth-first methods are computationally efficient but applicable to linear systems only. Later, Nguyen and Bagajewicz 14 extended the cutset-based methods to solve nonlinear problems by replacing cutsets with process model equations; they also proposed a special tree search method tailored for problems with high level of specifications, called the inverted tree search. Although very efficient in certain cases, both the equation-based method and the inverted tree search fail to perform well in realistic large-scale nonlinear problems (Nguyen and Bagajewicz 14 ).…”

Software Accuracy-Based Sensor Network Design and Upgrade in Process Plants

“…16 The inverse tree method using individual variables 16 can also be used, depending on the requirements on the system. At each node, software accuracy is evaluated using one of the three options: (i) accuracy defined by maximum undetected bias, (ii) stochastic accuracy using Monte Carlo simulation, (iii) expected value of accuracy using the approximate method.…”

Globally Optimized Power Allocation in Multiple Sensor Fusion for Linear and Nonlinear Networks

“…Deployment of these networks ensures controlling instrumentation in industrial automation, sensing data remotely in a data collecting environment and providing surveillance in defense related applications [7]. These networks were initially used in military applications but their potential has recently been exposed in other areas of science and engineering such as process monitoring in industrial plants, navigational and guidance systems, radar tracking, sonar ranging [24], [1], [22], [17].…”