Analyzing Plant-wide Control Structures for Industrial Processes

Abstract

Generally, any industrial process needs to be controlled, in some extend, to meet certain requirements related to safety, robustness, stability, and operating cost. However, a closed-loop policy requires several decisions before it will be considered as a potential and feasible control strategy. In fact, selections such as the controlled variables (CVs), the manipulated variables (MVs), the input-output pairing between these sets, the controller structure, policy, order, tuning, and technology are common in process control. Plant-wide control (PWC) is a very important and complex area to address the previously mentioned decisions. The overall performance as well as the investment/operating cost can be seriously affected if the PWC problem is not solved properly. Furthermore, this problem increases with the process dimension, which in many cases an heuristic and exhaustive treatment is not possible. In these cases it is very useful some systematic and generalized methodology to overcome these drawbacks. In this chapter, the decentralized PWC design is based on two modification for the well-known extended minimum square deviation (EMSD) approach. The overall strategy combines, in a single algorithm, most of the decisions previously discussed by using steady-state tools. Indeed, the selection of CVs, and MVs is evaluated based on the sum of square deviations (SSD) concept, parameterized by the controller order. Thus, a complete oversizing analysis is performed by considering all the potential PWC policies for a given industrial plant. On the other hand, the proposed strategy also takes into account the controller interaction degree assessment via the net load evaluation (NLE) approach. All the optimization problems presented in this chapter rely on multi-objective combinatorial representations and were efficiently solved via stochastic global search such as genetic algorithms (GA). The first PWC approach, is a sequential methodology involving SSD and NLE concepts, which allows defining the optimal solution via Pareto charts. This approach requires, a priori, a proper parametrization of all the potential orders. A second PWC methodology is suggested by integrating the selection of CVs, MVs, and the order all together in a single decision variable. This modification is very useful for systematizing all the PWC procedure. It is worth mentioning, that both modifications do not require dynamic simulations for supporting the final decision. This is particularly interesting because it opens a new relationship between PWC and fault-tolerant control (FTC) or variable structure control (VSC). All the approaches for PWC design presented in this chapter are tested on the well-known Tennessee Eastman (TE) process.