Dynamic Predictive Maintenance Strategy based on a Time Transformation Method

Abstract

An efficient and cost-effective maintenance strategy is key to ensuring availability and performance of industrial assets. Conventional time-based scheduled maintenance often fails to achieve this goal, especially as it is static: it does not take into account asset condition evolution nor evolution in operational conditions over time. Here, we propose a predictive maintenance strategy which is dynamic, based on asset remaining useful life (RUL), and risk-informed. Building on a recently introduced time transformation that facilitates RUL estimation, this work develops a simulation framework that enables implementation, testing, and optimization of a dynamic maintenance strategy for a fleet of identical assets. RUL estimates are continuously updated, and the timing of the next maintenance intervention is optimized such that the probability of failure before that time does not exceed a specified bound. Alternatively, a warning threshold and the probability bound can be selected to minimize the total expected cost of preventive maintenance actions and failures (the latter including corrective maintenance and failure consequences). The proposed dynamic strategy is benchmarked against conventional (static) scheduled maintenance strategies. The cost of the scheduled preventive maintenance policy is compared with that of the predictive maintenance policy. The study also examines the effects of imperfect maintenance. A simplified yet realistic example is presented to demonstrate the applicability of the method.