An Extended Predictive Maintenance Model under Covariate Distribution Parameters with Uncertainty

Predictive maintenance is the latest maintenance strategy, and it has been widely used in many areas. For most applications, condition monitoring may not directly reflect the degree of degradation. Therefore, both condition monitoring and life data can be considered comprehensively to estimate the failure hazards as the degradation state. And proportional hazard model is one of the popular methods. The proportional hazard model considers condition monitoring data as internal covariates, and it is generally assumed that distribution parameters of the covariates are determined for the predictive maintenance model. However, the distribution parameters may be of uncertainty due to the influence of component operating conditions and sensor detection accuracy fluctuations in practical applications. Therefore, this paper proposes an extended predictive maintenance model under the covariate distribution parameters with uncertainty. For the model, the Bayesian updating method is utilized to update the time-fixed covariate distribution parameters. The Monte Carlo simulation method is used to estimate approximately the expected cost rate for the extended predictive maintenance model under various prognosis scenarios. For the simulation studies, the influence of uncertain distribution parameters for covariate is analyzed on the optimal predictive maintenance policy. The results show that the proposed model can reduce the expected maintenance cost.