Opportunistic maintenance of complex systems using dynamic Bayesian networks

With the development of industry, the complexity of systems has increased even more. Unexpected downtimes can create huge costs for companies. Therefore, choosing the appropriate maintenance strategy has become much more important to prevent profit losses. The components in the system and their dependencies can create difficulties in evaluating these strategies. Therefore, it is very important to model the system correctly for the maintenance to be applied to be effective. Bayesian networks (BNs) are highly effective in modeling complex systems and the dependencies between their components. Dynamic Bayesian Networks (DBNs) are the extended versions of BNs by adding a temporal dimension to BNs to represent the dynamic nature of the work. In this study, we presented an opportunistic maintenance framework using DBNs for CNC machines in a production facility. A CNC machine is a partially observable system that consists of many components and has dependencies between its components. We examined different CNC machines and chose the most complex one to model the maintenance problem. We divided the system into ten subsystems, and we chose the Axis System, which was the most functionally important of these subsystems, to study the maintenance problem. We identified cause-and-effect relations in the system using HAZOP analysis. Then, we used DBNs to model the aging of components and the causal relations in the system, and to calculate the probabilistic inferences. We proposed an opportunistic maintenance approach under both corrective and proactive maintenance strategies for this complex system of eleven components with stochastic dependencies among them. We developed two opportunistic maintenance policies to be used in both corrective and proactive maintenance conditions. We tackled the maintenance problem with two different objectives. One is to minimize the total cost while the other is to minimize the total downtime duration. We ran the methods with different parameters in the corrective and proactive maintenance strategies. We compared these two methods with another maintenance methodology that did not use an opportunistic approach. Finally, we determined the conditions where the proposed opportunistic maintenance strategies operated better. The results revealed that opportunistic approaches showed promising performance in situations where a planned or unplanned breakdown led to high downtime costs. Therefore, since the downtime cost in corrective maintenance was higher than that of proactive maintenance, opportunistic policies gave better results in these cases. When the downtime cost in proactive maintenance was increased, the performance of opportunistic policies in proactive maintenance also improved. These findings were also supported by the results of the scenario analysis.