Manifold learning based rescheduling decision mechanism for recessive disturbances in RFID-driven job shops

In actual manufacturing processes, some unexpected disturbances, called as recessive disturbances (e.g., job set-up time variation and arrival time deviation), would gradually make the original production schedule obsolete. It is hard for production managers to perceive their presences. Thus, the impact of recessive disturbances can not be eliminated by rescheduling in time. On account of this, a rescheduling decision mechanism for recessive disturbances in RFID-driven job shops is proposed in this article, and a manifold learning method, which reduces the response time of manufacturing system, is applied in the mechanism to preprocess manufacturing data. The rescheduling decision mechanism is expected to answer the questions of whether to reschedule, when to reschedule, and which rescheduling method to be used. Firstly, RFID devices acquire the actual process completion time of all work in process (WIPs) at every WIP machining process completion time. Secondly, recessive disturbances are quantified to time accumulation error (TAE) which represents the difference between actual process completion time and planned process completion time. Lastly, according to the TAE and production managers’ experience, the rescheduling decision mechanism selects a proper rescheduling method to update or repair the original production schedule. The realization algorithms of rescheduling decision mechanism includes: (1) supervised locally linear embedding. (2) General regression neural network. (3) Least square-support vector Machine. Finally, a numerical experiment is used to demonstrate the implementation procedures of the rescheduling decision mechanism.