Symbolic Supervisory Control of Timed Discrete Event Systems

Abstract: With the increasing complexity of computer systems, it is crucial to have efficient design of correct and well-functioning hardware and software systems. To this end, it is often desired to control the behavior of systems to possess some desired properties. A specific class of systems is called discrete event systems (DES). DES deal with `discrete' quantities, e.g., ``number of robots in a manufacturing cell'', and their processes are driven by instantaneous `events', e.g., ``start of a machine''. In this thesis, the focus is on DES and an extension of such systems, which also considers the time points at which the events may occur, called \emph{timed DES (TDES)}. Real-time applications such as communication networks, manufacturing facilities, or the execution of a computer program, can be considered into TDES. Having a DES or TDES, with some given specifications, by utilizing a well-known mathematical framework, called supervisory control theory (SCT), it is possible to automatically generate a supervisor that restricts the system's behavior towards the specifications, only when it is necessary. Applying the SCT to large and complex systems, typically follows with some issues, concerning computational complexity and modeling aspects, which is tackled in this thesis. We model DES by extended finite automata (EFAs), state transition models that contain discrete-valued variables. TDES are modeled by an augmentation of EFAs, called timed EFAs (TEFAs), which contain a set of discrete-valued clocks. Based on EFAs or TEFAs, the supervisor can be symbolically computed, using binary decision diagrams (BDDs), data structures that could, in many cases, lead to smaller representation of the state space. For complex systems, the computed supervisor may consist of many states, causing representation and implementation difficulties. To tackle this, based on the states of the supervisor, we symbolically compute logical constraints that will be attached to the original models to restrict the system's behavior. Consequently, we present a framework, where given a set of EFAs or TEFAs, the supervisor is computed using BDDs, and represented in a modular manner based on the computed logical constraints. The framework has been developed, implemented, and applied to industrial case studies.