Using a Reflective Real-Time Operating System to Implement a Just-in-Time Scheduling Policy for a Flexible Manufacturing Workcell1

Research on the automated control of a flexible manufacturing workcell has generally assumed that there is a low-level computer system to control the machines, without addressing the precise functionality of this low-level system and how it might uniquely support the manufacture of products. This work addresses how Spring, a real-time operating system, can be used to implement a Just-In-Time (JIT) manufacturing policy for a flexible manufacturing workcell. The JIT policy is characterized by the right amount of manufacturing material being at the right place at the right time. We argue that two schedulers are desired to control a workcell because of two sets of scheduling requirements and responsibilities. The High-Level Scheduler (HLS) schedules machine operations need to construct products as close to their deadlines as possible. Spring schedules the fine-level computational tasks that comprise machine operations according to the Worst-Case Execution Time (WCET) of the tasks such that the individual tasks are executed before their deadline. The reflective properties of Spring—retaining at run-time the worst-case execution time of tasks, their resource requirements, etc.—enable Spring to provide dynamic feedback to the HLS, which the HLS uses to make scheduling decisions. Simulation studies show that the two-level scheduling system outperforms both (a) a single scheduler based on worst-case execution times, and (b) a single scheduler based on average-case execution times. Using only the worst-case execution times can lead to underutilization of resources and relatively few products manufactured, while using only the average-case execution times can lead to multiple products missing their deadlines through a ripple effect. The contribution of this work is showing how the reflective properties of the real-time operating system allow the HLS to make important decisions in a fast-changing environment. 1This work is funded by the National Science Foundation under grant IRI-9208920.