Predictive maintenance policy based on process data

a r t i c l e i n f o For the 'under maintained' and 'over maintained' problems of traditional preventive maintenance, a new predictive maintenance policy is developed based on process data in this article to overcome these disadvantages. This predictive maintenance method utilizes results of probabilistic fault prediction, which reveals evolvement of the system's degradation for a gradually deteriorating system caused by incipient fault. Reliability is calculated through the fault probability deduced from the probabilistic fault prediction method, but not through prior failure rate function which is difficult to be obtained. Moreover, the deterioration mode of the system is determined by the change rate of the calculated reliability, and several predictive maintenance rules are introduced. The superiority of the proposed method is illustrated by applying it to the Tennessee Eastman process. Compared with traditional preventive maintenance strategies, the presented predictive maintenance strategy shows its adaptability and effectiveness to the gradually deteriorating system. The annual cost of maintenance has been reported to go up to 15% for manufacturing companies, 20%–30% for chemical industries [1], 40% for iron and steel industries [2]. Thus developing new maintenance technologies and arranging proper maintenance scheduling has become more and more important to enhance production and economic efficiency. Despite this economic factor, the maintenance of equipment always has a major impact on system reliability, availability and security. The evolvement of maintenance technology has experienced three different types, i.e. corrective maintenance (CM), preventive maintenance (PvM) and predictive maintenance (PdM). CM, the earliest maintenance technology, means repairing a system only after a breakdown or an obvious fault. PvM means performing repair, service, or replacement for a component or system at a fixed period to prevent a breakdown. PdM decides whether or not to do system maintenance based on the condition of the system. CM and PvM are the traditional maintenance policies, which may cause low reliability or high maintenance cost. While PdM utilizes appropriate condition monitoring and maintenance management technologies, which can greatly increase the efficiency and profitability of industrial production [3]. Although PdM is an effective approach to promote system reliability, the implementation of this new PdM is not an easy task for uncertainties and less of fault data in practical processes. Most of the present predictive maintenance approaches are based on a classical assumption, that is, the system failure can be explained by a stochastic deterioration process [4–7], which is consistent with many …