Finite-Time Consensus and Readjustment Three-Stage Filter for Predictive Schedules in FMS

Enabling manufacturing systems to adapt quickly to production needs is crucial for industries to gain a competitive advantage. Modern manufacturing systems must be sufficiently flexible and intelligent to react promptly to market demand and ensure that manufacturing operations are deadlock-free. This study proposes a distributed multi-agent system (MAS) for flexible manufacturing systems (FMS) with order-controlled production that avoids deadlocks through a readjustment three-stage filter. Production schedules generated by a classical predictive scheduling algorithm are modified by three readjustment filters to avoid potential deadlocks and are used by MAS agents to drive production. The first filter sequentially groups certain events from schedules according to the buffer type of their respective production resources. The second filter parallelizes certain events associated with null-buffer resources and breaks the contiguity of other specific events to optimize production. The third filter identifies overlap by mapping the events that cause deadlocks and readjusting their positions in the production schedules. The readjusted schedules are sent to the MAS agents who cooperate in executing production tasks using event-based consensus control with predefined times and message exchanges via robot operating system topics. A physical factory simulation platform is modified to operate as an FMS and used to evaluate the proposed approaches. The FMS simulates a production environment with resource sharing and product parallelism. The experiments prove that the rescheduled predictive schedules are fully executed in the FMS without employing computationally onerous rescheduling methods for deadlock situations involving non-preemption and circular waiting. This study details the operations of consensus control and the algorithms used in the readjustment filters of the predictive schedules. At the end of the paper, information is provided on the MAS load scalability and the time complexity of the proposed agent model and the readjustment filters.