Decomposing Process Network into units for Distributed Monitoring and Advanced Control

Process networks generally comprise interconnected operating units coupled together. To ensure the operating safety and efficiency and to achieve higher profits, advanced control systems have been developed and widely applied. In particular, distributed controls have attracted considerable attention from both academia and industry. The first key step to develop a distributed control system is to decompose the entire process network into smaller units. It is worth mentioning that different decompositions of process networks can affect the control performance significantly, thus subsystem decomposition has been recognized as a fundamental and challenging problem in the control systems engineering field. While researchers in this field have proposed a few subsystem decomposition methods for distributed control, most of them require that all the state information should be able to be measured by sensors in an online fashion, which is generally difficult to realize in practice. To this end, researchers have been looking for an alternative solution and have identified the incorporation of distributed state estimation and distributed control in a single integrated design as a promising solution.

In a newly published research paper in AIChE Journal, Dr. Xunyuan Yin (Postdoctoral fellow) and Professor Jinfeng Liu from the University of Alberta developed a systematic approach that can be used to divide nonlinear process networks into smaller units for simultaneous distributed estimation and control. The solution to this problem was based on a community structure detection approach that resorts to modularity maximization, and it has proved to be very effective and efficient.

The proposed subsystem decomposition procedure involves several key stages: initial vertices aggregation, adjacency matrix construction, subsystem configuration based on the modularity maximization, and then subsystem configuration considering the smallest modularity decrease. A network is first established by treating the process states, sensor measurements and control inputs as the nodes of the network. Additionally, mutual connections between the nodes are characterized based on the bidirectional edges. Then, the authors proposed to initialize the subsystem structure and construct an adjacency matrix suitable for simultaneous monitoring and control. The measure of modularity is taken advantage of to evaluate the quality of possible community structures. Then, a fast-folding algorithm is adopted for maximizing the modularity to find an optimal structure among different subsystem configurations. Whereas the modularity value could take a value between 0 and 1, larger values indicate more appropriate community structures. Based on the optimal subsystem structure recommended by this new method, satisfying monitoring and control performance can be achieved while the load of communication among local agents can be much reduced.

One of the advantages of this method is that users can choose a desired number of subsystems when considering decomposing a large process into smaller units. Another major advantage is that the proposed method can deal with processes of very large scales efficiently by applying graph-theory-based concepts and algorithms; these are very favorable from an application point of view.

The effectiveness and applicability of the proposed approach were illustrated using two large industrial processes with different complexities. Specifically, the approach that Dr. Yin and Dr. Yin proposed is applied to a rector-separator process and a large-scale wastewater treatment process to establish subsystem structures that can be directly used for developing monitoring and control systems in a distributed architecture. Through the use of a folding algorithm, it allows finding the optimal subsystem structure while it can more appropriately handle the constraints on the subsystem structure. The study by Dr. Yin and Dr. Liu is the first to address subsystem decomposition for simultaneous distributed estimation and control, which will advance the design and development of high-performance monitoring and control systems for large-scale industrial processes.