279621 Reduction and Control of Tightly-Integrated Plants Using Graph-Theoretic Methods

Thursday, November 1, 2012: 1:35 PM
324 (Convention Center )
Seongmin Heo1, Sujit S. Jogwar2 and Prodromos Daoutidis1, (1)Chemical Engineering and Materials Science, University of Minnesota, Minneapolis, MN, (2)Praxair, Inc., Tonawanda, NY

Energy integration is ubiquitous in modern chemical plants. Even though it leads to significant cost savings by reducing external utility consumption, integrated process networks become harder to be operated and controlled, especially in the context of transitions between different operating conditions. To this end, the control of large scale, integrated plants has been an active area of research for the past several years.

In our previous work, we have identified two different prototype structures with energy integration: networks with large energy recycle and networks with large energy throughput. It has been shown that each prototype network exhibits unique two-time scale dynamics, and model reduction has been done to obtain the description of the dynamics of the networks in each time scale. However, in the case of complex tightly-integrated plants, networks are complex, comprising of interconnections of these prototype networks. The resulting multi-loop structure has a potential to exhibit multi-time scale dynamics.

One possible approach for the analysis of these networks is detailed mathematical analysis with a successive application of singular perturbations. However, such an analysis becomes cumbersome with increasing size and complexity of a process network. An alternative is a graph theory based approach which is a generic and scalable analysis framework.

To this end, we have developed a graph-theoretic algorithm which can be used to analyze complex process networks with segregation of the energy flows. Specifically, the developed algorithm uses information about the structure and order of magnitude of the different energy flows involved in a network to automatically generate information on i) the time scales exhibited by the network, ii) control objectives to be pursued in each time scale and manipulated variables to enforce them, and iii) form of the reduced order models in each time scale.

In this work, we demonstrate the application of the developed algorithm to a benchmark chemical plant example, the toluene HDA process. Specifically, we focus on the reaction part of a design of the process which contains one feed-effluent heat exchanger (FEHE). First, we show that the process exhibits three-time scale dynamics using both numerical simulation and the developed graph theory algorithm. Specifically, the enthalpies of the process units associated with the inner recycle loop (recycle through the FEHE) are evolving in the fast time scale, the enthalpies of the process units associated with the outer recycle loop (gas recycle) as well as the total enthalpy of the inner recycle loop are evolving in the intermediate time scale and the total network enthalpy is evolving in the slow time scale. We propose a hierarchical control structure based on the results given by the algorithm. Specifically, we employ a nonlinear (input/output linearizing) controller as the slow time scale controller to control the variables at the exit of the network such as the benzene mole fraction of the liquid outlet of the separator. We use simple PI controllers as the intermediate and the fast time scale controllers to control the variables associated with the inner recycle loop such as the inlet temperatures of the furnace and the reactor. Simulation studies illustrate the effectiveness of the proposed control structure.


Extended Abstract: File Not Uploaded
See more of this Session: Control and Estimation of Large Scale Systems
See more of this Group/Topical: Computing and Systems Technology Division