Supervisory control of distributed systems using Petri nets and network unfolding

Distributed systems are large systems consisting of many concurrent components. Guaranteeing the desirable operation of such systems has long been recognized as an extremely challenging issue. Traditional methods based on exhaustive state space searching were hindered by the exponential-size of the state space and the uncontrollability of system executions. This dissertation presents novel results that use supervisory control theories to synthesize supervisory plug-ins for distributed systems. A plug-in can be implemented as a software patch that disables certain system interactions that lead to undesirable system executions. This dissertation selects Petri net as the distributed system model and uses network unfolding to facilitate supervisor synthesis. Unfolding is a partial order method that generates a compact state space and identifies strings of fundamental system executions (base configurations). This dissertation demonstrates that supervisory plug-ins can be synthesized by identifying critical system states and disabling interactions between base configurations that lead to these states. It also proves that the plug-ins synthesized are optimal in the sense that they are maximally-permissive.