Chemical Engineering Seminar Series: Prof. Sujit Jogwar

Community behavior of control systems.

Integrated chemical plants with material recycle and energy integration are a rule rather than exception in modern manufacturing complexes due to their sustainability (economic and environmental) benefits. However, the tight interconnections introduced by such integration lead to challenges in operation and control. In this context, distributed control offers a practical trade-off between globally (plant-wide) valid but computationally expensive centralized and locally effective decentralized architectures. Even though the performance of distributed control strongly depends on the selected decomposition (classification of controller variables), optimal synthesis of distributed structure has not received much attention.

Decomposition of a large, interconnected network into a number of communities (with strong intra-community interactions and weak inter-community interactions) has seen tremendous success in the analysis of such networks observed in social, ecological, biological and information systems. Motivated by this, our research contribution attempts at developing a novel strategy for distributed architecture synthesis by identifying communities of control relevant variables. Accordingly, key issues addressed in this research include abstraction of the control problem into a meaningful graph and establishing a connection between properties of distributed architecture (inter-controller communication, input-output strength, robustness, etc.) and community detection on the abstracted graph. Our framework has many advantages (e.g. possibility of non-square controllers, provision to ensure controllability and observability, scalability to large networks, etc.) over the existing approaches. The effectiveness of the framework is illustrated via several industrially relevant examples through rigorous simulations and lab-scale experiments. This work has also opened up a path for the synthesis of distributed architectures in allied areas like estimation, fault detection and optimization.