Control of deadlock and blocking for production systems with unreliable workstations

Over the past decade, the development of supervisory controllers for automated manufacturing systems has been an active area of research with particular attention focusing on deadlock avoidance. Most work to date assumes that system workstations do not fail. This paper considers deadlock and blocking problems in systems with one unreliable workstation, the objective being to develop supervisory control policies that allocate workstation buffer space so that failure of the workstation server does not propagate through blocking to effectively stall other portions of the system. In short, when a workstation fails, the system should automatically continue producing all part types not requiring that workstation. To accomplish this, the supervisory controller must simultaneously avoid deadlock and buffer allocation states that are not feasible initial states for the 'reduced' system resulting from workstation failure. When the workstation fails, the controller must then simultaneously avoid both deadlock and buffer allocation states that are not feasible initial states for the original system so that transition to normal operation is smooth when the failed workstation is restored. This paper shows this class of problems through several examples, identifies properties that controllers must satisfy to deal effectively with these problems and develops a polynomial control policy that satisfies these properties.