Hierarchical / Distributed Control of Complex Process Networks
Event details
| Date | 23.06.2017 |
| Hour | 10:15 › 11:15 |
| Speaker | Prof. Prodromos Daoutidis, University of Minnesota, USA |
| Location | |
| Category | Conferences - Seminars |
Abstract :
The talk will focus on the control of integrated large-scale plants, a classic open problem in process control.
The first part of the talk will focus on the dynamics and control of networks with large rates of material and / or energy recovery and recycle, compared to input/output flows. Such networks exhibit dynamics over multiple time scales, with individual units evolving in a fast time scale with weak connections, which become significant over slower time scales giving rise to a slow evolution of the entire process network. A model reduction method based on singular perturbations will be described which allows obtaining a hierarchy of low-order nonlinear models valid in the different time scales. A graph reduction analogue of this method which can be fully automated will also be described. This multi-time-scale analysis lends itself naturally to a hierarchical control framework, whereby network-level control objectives can be effectively addressed at a slow, supervisory level.
The second part of the talk will address generic integrated process networks with no material or energy flow segregation. A natural paradigm for addressing this problem is the one of distributed control, in which coordinated controllers tackle operational objectives of different sections of the plant. A key underlying problem is the optimal decomposition of the integrated system into the distributed control architecture. A new approach to this problem inspired from network science will be described. It relies on identifying “communities” of system variables whose members interact strongly among them, yet are weakly coupled to the rest of the network members. A modularity measure defined on suitable graphs is used to quantify strength of interactions; maximization of modularity leads to optimal decompositions. Such decompositions will be shown to lead to significant reduction in the computational cost of distributed optimization-based control while retaining satisfactory performance compared to centralized control. The role of communities in sparsity-promoting control of networks will also be elucidated.
Bio :
Prodromos Daoutidis is a College of Science and Engineering Distinguished Professor and Executive Officer in the Department of Chemical Engineering and Materials Science at the University of Minnesota. He received a Diploma degree in Chemical Engineering (1987) from the Aristotle University of Thessaloniki, M.S.E. degrees in Chemical Engineering (1988) and Electrical Engineering: Systems (1991) from the University of Michigan, and a Ph.D. degree in Chemical Engineering (1991) from the University of Michigan. He has been on the faculty at Minnesota since 1992, having served as Director of Graduate Studies in Chemical Engineering (1998-2004) and Chair of the Physical Sciences Policy and Review Council (2000-03), while he has also held a position as Professor at the Aristotle University of Thessaloniki (2004-06). He is the recipient of several research and teaching awards and recognitions, including the NSF CAREER Award, the PSE Model Based Innovation Prize, the Best Paper Prize from the Journal of Process Control, the Ted Peterson Award of the CAST Division of AIChE, the George Taylor Career Development Award, the McKnight Land Grant Professorship, the Ray D. Johnson / Mayon Plastics Professorship and the Shell Chair at the University of Minnesota. He has also been a Humphrey Institute Policy Fellow. He has served as Program Coordinator in Areas 10B and 10D of the CAST Division of AIChE, and as AIChE Director in AACC, and is currently serving as CAST Programming Chair. He is the Associate Editor for Process Systems Engineering in the AIChE Journal, and an Associate Editor in the Journal of Process Control. He has co-authored 5 books, 250 refereed papers, and has supervised to completion 26 PhD students and post-docs, 10 of which have gone into academic positions. His research interests are in design and control of energy systems, process and plant-wide control, control of nonlinear and distributed parameter systems, model reduction, dynamics and control of chemical and biological systems, and control of advanced materials processing.
The talk will focus on the control of integrated large-scale plants, a classic open problem in process control.
The first part of the talk will focus on the dynamics and control of networks with large rates of material and / or energy recovery and recycle, compared to input/output flows. Such networks exhibit dynamics over multiple time scales, with individual units evolving in a fast time scale with weak connections, which become significant over slower time scales giving rise to a slow evolution of the entire process network. A model reduction method based on singular perturbations will be described which allows obtaining a hierarchy of low-order nonlinear models valid in the different time scales. A graph reduction analogue of this method which can be fully automated will also be described. This multi-time-scale analysis lends itself naturally to a hierarchical control framework, whereby network-level control objectives can be effectively addressed at a slow, supervisory level.
The second part of the talk will address generic integrated process networks with no material or energy flow segregation. A natural paradigm for addressing this problem is the one of distributed control, in which coordinated controllers tackle operational objectives of different sections of the plant. A key underlying problem is the optimal decomposition of the integrated system into the distributed control architecture. A new approach to this problem inspired from network science will be described. It relies on identifying “communities” of system variables whose members interact strongly among them, yet are weakly coupled to the rest of the network members. A modularity measure defined on suitable graphs is used to quantify strength of interactions; maximization of modularity leads to optimal decompositions. Such decompositions will be shown to lead to significant reduction in the computational cost of distributed optimization-based control while retaining satisfactory performance compared to centralized control. The role of communities in sparsity-promoting control of networks will also be elucidated.
Bio :
Prodromos Daoutidis is a College of Science and Engineering Distinguished Professor and Executive Officer in the Department of Chemical Engineering and Materials Science at the University of Minnesota. He received a Diploma degree in Chemical Engineering (1987) from the Aristotle University of Thessaloniki, M.S.E. degrees in Chemical Engineering (1988) and Electrical Engineering: Systems (1991) from the University of Michigan, and a Ph.D. degree in Chemical Engineering (1991) from the University of Michigan. He has been on the faculty at Minnesota since 1992, having served as Director of Graduate Studies in Chemical Engineering (1998-2004) and Chair of the Physical Sciences Policy and Review Council (2000-03), while he has also held a position as Professor at the Aristotle University of Thessaloniki (2004-06). He is the recipient of several research and teaching awards and recognitions, including the NSF CAREER Award, the PSE Model Based Innovation Prize, the Best Paper Prize from the Journal of Process Control, the Ted Peterson Award of the CAST Division of AIChE, the George Taylor Career Development Award, the McKnight Land Grant Professorship, the Ray D. Johnson / Mayon Plastics Professorship and the Shell Chair at the University of Minnesota. He has also been a Humphrey Institute Policy Fellow. He has served as Program Coordinator in Areas 10B and 10D of the CAST Division of AIChE, and as AIChE Director in AACC, and is currently serving as CAST Programming Chair. He is the Associate Editor for Process Systems Engineering in the AIChE Journal, and an Associate Editor in the Journal of Process Control. He has co-authored 5 books, 250 refereed papers, and has supervised to completion 26 PhD students and post-docs, 10 of which have gone into academic positions. His research interests are in design and control of energy systems, process and plant-wide control, control of nonlinear and distributed parameter systems, model reduction, dynamics and control of chemical and biological systems, and control of advanced materials processing.
Practical information
- General public
- Free