Summary
Main Objective
The objective of DISC is the design of supervisors and fault detectors exploiting the concurrency and the modularity of the plant model. Coordinated controllers should preferably be designed using only local plant models, and requiring only limited information exchange between local controllers. The diagnosers should be capable of handling the asynchronous nature of distributed systems.
We plan to use several techniques to reduce the computational complexity of solving the above mentioned problem for distributed plants: modularity in the modelling and control design phases; decentralized control with communicating controllers; modular state estimation, distributed diagnosis and modular fault detection based on the design of partially decentralized observers; fluidisation of some discrete-event dynamics to reduce state-space cardinality.
The expected outcome of this project are: new methodologies for modular control design and diagnosis of complex distributed plants and new tools for the modelling, simulation and supervisory control design that will be part of an integrated software platform.
To achieve these goals it is necessary that the component models, used for design, be very flexible. The common platform will allow us to use heterogeneous models, such as timed or fluid Petri nets, timed automata etc., within the same design problem. To successfully achieve the project the partners of DISC will appropriately combine these experiences for model reduction/abstraction.
To validate the methods and the platform that they will develop the partners of DISC will consider three case studies that address industrially relevant problems in three different cases: road traffic control; high throughput screening in the chemical and pharmaceutical industries; automotive safety monitoring systems. Each of these examples involves modelling the plant as a network of interacting timed discrete-event components.
Technical Approach
One of the most obvious consequences of the increasing complexity of man-made systems, is the emergence of more and more autonomous modes of operation for complex distributed plants. In fact, as the complexity of these systems grows, the risk of a human operator error increases and a correct behaviour can only be ensured by a supervisory control system.
Supervisory control theory was introduced by Ramadge and Wonham to solve a general design problem: given a plant and a specification that describes a legal behaviour, is it possible to construct a supervisor, i.e., a control agent, that observes the current state of the plant and selects actuators that can control the execution of some controllable events in order to meet the specification? Synthesis procedures have been obtained for supervisors but the computational cost of the design of these supervisory controllers is often considered prohibitive due to the exponential ex-plosion of the state space.
Technological and economical limitations often imply that supervisors must operate with limited state information: sensors cannot detect every event in every component; bandwidth limitations restrict the information exchange between different agents. All this makes the synthesis problem even harder, and often requires that the two related problems of state estimation and fault diagnosis should be simultaneously addressed during the supervisory design phase. Note that state estimation and fault diagnosis suffer from the same curse of dimensionality as supervisory design.
A new generation of technologies and tools for supervisory control and diagnosis of wide infra-structures or manufacturing organisations is clearly necessary. One main difficulty is to offer methods and tools that engineers can use without years of new education. This means that one should not try to change their current techniques for the local handling of sensors, modes of operation, actuators and the like. We therefore plan to provide computer assisted methods for solving the system-level problems, relying on discrete models such as automata and Petri nets that many control and computer engineers are already accustomed to. We have good reasons to believe that this is the right level on which methods and efficient tools must be developed so as to produce optimal solutions to hard combinatorial or optimisation problems. This may also allow adapting controllers in a modular way in case changes occur in these infrastructures, but still guaranteeing safety of the system.
Key Issues
The above mentioned needs clearly motivate the choices that were made in the preparation of DISC.
a) There is a significant methodological part of the project to pave the way for additional new technological advances in the field. A major focus of the project is on decentralized control for distributed plants.
b) The project devotes simultaneously attention to the two inter-related aspects of "supervisory control" and "fault diagnosis".
c) The models chosen are discrete-event driven and asynchronous models and the reference models that will be used in the project are automata and Petri nets.
d) An important part of the project is a workpackage devoted to the development of a software platform and of tools that will allow an engineer (and not necessarily a researcher) to apply the developed methodologies not only to academic problems, but also to problems of industrial relevance.
e) The workpackage devoted to the dissemination of the results does not neglect the communication aspect, and several workshops and minicourses are foreseen.
Expected Impact
The expected outcome of this project are: new methodologies for modular control design and diagnosis of complex distributed plants and new tools for the modelling, simulation and supervisory control design.
Such tools will embed powerful algorithms from well mastered parts of control theory and computer science that end users don't have to be aware of, and yield as a result adequate controllers in a readable format (automata, Petri nets...) that engineers can experiment with and later implement. We believe this is the only way to obtain correct and efficient controllers for complex plants in a reasonable time.
Web: http://www.disc-project.eu