Adaptive predictive maintenance optimization for continuous process manufacturing systems considering uncertain task profiles

Continuous process manufacturing systems (CPMSs) refer to manufacturing systems whose operation is unstoppable during the execution of production task. Given that in-process shutdown is unavailable for CPMSs, the only approach to assure task completion is the maintenance activities conducted within the intervals between task executions. However, due to the variation of production requirements raised by the uncertain changes in task profiles, fixed maintenance strategies are facing increasing adverse in application. Accordingly, this study proposes a novel approach to predictive maintenance (PdM) optimization that adapts to the CPMS working-condition shift associated with task profile uncertainty. Specifically, based on a solid investigation of production task profile uncertainty and its influence on CPMS operation, an adaptive PdM optimization method considering the current and future maintenance effect is proposed. With the aid of a reinforcement learning (RL) algorithm, it can dynamically adjust the maintenance policy according to the prediction of future task profile changes. The applicability of the proposal is verified with an industrial case of an insulating base CPMS.