Complex process networks, consisting of interconnections of numerous reaction, separation and heat exchange units, represent a key feature of modern chemical and energy plants. Controlling such networks effectively, especially in the context of transitions between different steady states, is a challenging problem. The intricate dynamics and model complexity that such networks exhibit, make conventional decentralized control design inadequate and fully centralized control design approaches impractical. Despite some progress in analyzing simple networks and designing distributed or hierarchical controllers for such networks, a rigorous yet scalable model simplification and controller design framework for complex networks is currently lacking. The goals of the this research are therefore: (i) to develop scalable methods for model reduction and decomposition of complex process networks, (ii) to develop concrete control algorithms for transition control of such networks, and (iii) to apply the developed methods to representative systems from the process and energy industries, including complex thermally coupled distillation trains, cryogenic systems and integrated reformer - fuel cell systems.
Intellectual Merit: The main novelty of this work is the introduction of graph theory as a framework for analyzing the structural properties of such complex networks, responsible for their ensemble behavior. Specifically, their modular structure lends itself to a graph theoretic analysis, whereby weak and strong connections between process units arising from time scale separation or other connectivity properties can be identified from structural information; these can be used either for model reduction (when a reduced model does exist, e.g. owing to a slow, low-order network dynamics) or model decomposition in the more general case where the ensemble behavior is captured through the aggregate behavior of some distinct, loosely coupled community structures within the network. For both cases, scalability to large scale networks will be enabled by using powerful graph-theoretic algorithms for automating the model simplification and possibly the controller design procedure. A software tool that will achieve this will be built within an object oriented programming framework.
Broader Impact: Controlling complex, integrated plants is a critical link to the economic viability, and the energy and environmental sustainability of the chemical and energy supply chains. This research will develop computational tools that will enable the development of such control methods in an automated and scalable fashion. This analysis framework can also be applied to complex networks from other disciplines, such as complex reaction pathways, ecological networks and social networks. The research will provide a setting for the effective training of graduate students in fundamental research cutting across mathematics and control, with a timely and important application component. The students will also interact with industrial partners through summer internships. The research results will be broadly disseminated through publications and presentations, and through their integration into the teaching of process control. The software that will be developed will further enhance the infrastructure for research and education.