Home > Seminars > Distributed Supervisory Control of Large-Scale Discrete-Event Systems - From Correct Behavior to Optimal Performance

Distributed Supervisory Control of Large-Scale Discrete-Event Systems - From Correct Behavior to Optimal Performance


6/26/2015 at 11:00AM


6/26/2015 at 12:00PM


258 Fitzpatrick Hall


College of Engineering close button

Hai Lin

Hai Lin

VIEW FULL PROFILE Email: hlin1@nd.edu
Phone: 574-631-3177
Website: http://nd.edu/~hlin1/
Office: 265 Fitzpatrick Hall
Curriculum Vitae


College of Engineering Associate Professor
Wireless Institute Associate Professor
Dr. Lin's teaching and research interests are in the multidisciplinary study of the problems at the intersections of control, communication, computation and life sciences. His current research thrust is on cyber-physical systems, multi-robot cooperative tasking, advanced manufacturing systems, ...
Click for more information about Hai
Add to calendar:
iCal vCal
Since the Ramadge-Wonham (RW) supervisory control paradigm first appeared in the control literature in 1982, it has found many applications in manufacturing systems, material handling systems, transportation systems, communication networks and software systems. One of the main challenges of RW supervisor synthesis is to achieve nonblockingness in a target system, i.e., to ensure the target system to have a possible trajectory to reach a final state regardless of its current state. The difficulty arises when the target system has a large number of states, often resulted from synchronous product of many relatively small local components. To overcome this computational difficulty, many approaches have been proposed recently. For example, Ryan Leduc introduced the concept of interface invariance in his hierarchical interface-based supervisory control approach. Some large nonblocking control problems such as the AIP system with the size of about 10^21 states have been solved successfully. Ma Chuan proposed a BDD-based supervisor synthesis approach for state feedback control based on the concept of state tree structures. It has been shown that a system with 10^24 states can be readily handled. A breakthrough has appeared recently when several distributed synthesis approaches are introduced, which utilize either deterministic or nondeterministic automata to model a target system, and apply different model abstraction techniques to effectively avoid the state explosion phenomenon during synthesis. Such divide-and-conquer strategies have been shown effective for many industrial scale applications, and in many circumstances even lead to the least restrictive controlled system behaviors, which are typically considered achievable only through centralized control.        

Due to the introduction of distributed synthesis techniques, the long-standing complexity hurdle for supervisor synthesis aiming for behavioral correctness has been removed successfully. But another daunting challenge appears, which requires supervisory control theorists to develop approaches that not only ensure behavioral correctness, but also performance optimality – the latter is perhaps more appealing to industry. The main computational difficulty lies in the process of finding a few “good” trajectories among a vast number of possible trajectories according to some given performance metrics such as minimum makespan or maximum throughput. To describe those performance metrics, several automata models have been used, e.g., weighted automata, timed automata, extended finite-state automata and time-weighted automata. It turns out that performance optimality based on additive metrics such as fuel consumptions, traveling distances and operational costs etc. can be relatively easy to handle, but performance optimality based on concurrency-related metrics such as makespan is extremely difficult to achieve because the principle of optimality usually does not hold, making it hard for a designer to use local design to ensure globally good performance. I will briefly describe our current ongoing effort in this direction, mainly on distributed time-weighted synthesis to ensure good makespan of a given set of tasks within a large scale discrete-event system. Our work shows the feasibility of applying a distributed strategy in performance oriented supervisory control, which opens a door for future research in dealing with large scale complex systems.

Seminar Speaker:

Rong Su

Rong Su

Nanyang Technological University

Dr. Rong Su obtained his BE degree from University of Science and Technology of China in 1997, and MASc and PhD degrees from University of Toronto in 2000 and 2004 respectively. After holding research positions at University of Waterloo and Technical University of Eindhoven, he joined Nanyang Technological University in 2010. Dr Su’s research interests include discrete-event modeling, supervisory control, model-based fault diagnosis, operational planning and scheduling in complex systems, green building and intelligent transportation systems. He has more than 80 journal and conference papers and book chapters on the relevant subjects, and 2 patents. So far he has been involved in 8 ongoing research projects funded by Singapore National Research Foundation, Singapore Agency for Science, Technology and Research, Singapore Ministry of Education, Singapore Economic Development Board, and Civil Aviation Authority of Singapore, where he plays the role of either PI or Co-PI with a total funding of more than S$4.5 mil. Dr Su has served in Technical Program Committees and/or Organizing Committees of many conferences such as WODES’14, ICCA’14, WCICA’14, CCDC’14, ICARCV’14, CCECC’15, ICCA’15, ICIA’15, ETFA’15, CASE’15, CASE’16 and ICARCV’16, and is an associate editor for Journal of Control and Decision (Taylor & Francis Group), and Transactions of Institute of Measurement and Control. Dr Su is a senior member of IEEE.