Data-driven approach for imperfect maintenance model selection

To better describe the reality that repaired systems are neither in “as good as new” nor “as bad as old” conditions, many imperfect maintenance models have been developed in the past decades to mathematically characterize the maintenance quality from various points of view. However, without complete knowing the potential physics of failures and the mechanism of improvement from repair, but only with the data from system operation, it always troubles the industry participants whether the pre-assumed imperfect maintenance model is adequate enough to describe the effect of repair. Beyond the aforementioned question, another issue might arise once there are several candidate models available. The participants desire to select the most adequate model among the candidates to facilitate the future decision making. To address these two important issues, a Goodness-Of-Fit (GOF) test is proposed in this paper to check the adequacy of the pre-assumed imperfect maintenance model. Meanwhile, a Bayesian model selection method is also introduced with the aim of identifying the most adequate model among several competitive candidates. The effectiveness of the proposed methods is demonstrated by the designed numerical studies, and it shows that the methods have a great capability to identify the most adequate model from the competitive candidates.