(149e) Distributed Control of Integrated Process Systems – an Experimental Study

Distributed architecture has gained significant attention in recent years in the context of estimation, optimization and control of large scale systems. It proposes valuable trade-off between performance and practical implementation. In the context of distributed control, especially distributed model predictive control (DMPC), significant work has been done in terms of problem formulation, communication protocols, convergence and stability analysis, and performance comparison with centralized and decentralized control schemes. The performance of DMPC strongly depends on the underlying distributed architecture (decomposition of the problem into sub-problems). Such a decomposition is non-trivial in the case of integrated process systems and has been a focus of research in the last few years.

Initial works on distributed architecture synthesis relied on heuristics (such as topology [1] or variable type [2]) to decompose the master control problem into coupled sub-problems. Recent works have relied on graph theory to systematically decompose the master control problem while simultaneously optimizing one of the performance indicators (e.g. minimum computational time [3], response sensitivity [4] or strength of input-output response [5]) of the overall system. For an integrated system, one can therefore have a number of distributed architectures to choose from. In an attempt to come up with useful recommendations, these structures are explicitly compared via rigorous simulation studies [6]. However, no such comparison has been performed on an live system which poses a number of practical challenges like plant-model mismatch, measurement noise, transport lags and computational limitations.

In this work, distributed control of an experimental quadruple tank system is pursued. This 2 input-2 output system gives rise to 2 DMPC architectures as shown in the figure below. It is established that, depending on the operating steady state, one architectures proposes better (analytical) performance compared to the other. The proposed work aims at experimental validation of these theoretical claims as well as quantification of this performance improvement. Firstly, a state-space model of the experimental system is identified and used for the synthesis of distributed architectures based on the existing methods. The DMPC controllers are then implemented and tuned via rigorous simulations. Subsequently, they are deployed on the experimental facility and their performance is compared with a centralized MPC. The rigorous comparison of these three control schemes, in terms of closed-loop performance and computation time, under practical constraints, is presented.

References:

[1] Stewart, B. T., Venkat, A. N., Rawlings, J. B., Wright, S. J., & Pannocchia, G. (2010). Cooperative distributed model predictive control. Systems & Control Letters, 59(8), 460-469.

[2] Liu, J., Muñoz de la Peña, D., & Christofides, P. D. (2009). Distributed model predictive control of nonlinear process systems. AIChE journal, 55(5), 1171-1184.

[3] Jogwar, S. S., & Daoutidis, P. (2017). Community-based synthesis of distributed control architectures for integrated process networks. Chemical Engineering Science, 172, 434-443.

[4] Tang, W., & Daoutidis, P. (2018). Network decomposition for distributed control through community detection in input–output bipartite graphs. Journal of Process Control, 64, 7-14.

[5] Jogwar, S. S. (2019). Distributed control architecture synthesis for integrated process networks through maximization of strength of input–output impact. Journal of Process Control, 83, 77-87.

[6] Pourkargar, D. B., Almansoori, A., & Daoutidis, P. (2018). Comprehensive study of decomposition effects on distributed output tracking of an integrated process over a wide operating range. Chemical Engineering Research and Design, 134, 553-563.