A novel data-driven approach to optimizing replacement policy

Parallel systems are a commonly used structure in reliability engineering. A common characteristic of such systems is that the failure of a component may not cause its system to fail. As such, the failure may not be immediately detected and the random (disruption) time at which the number of failed components reaches a certain predefined number may therefore be unknown. For such systems, scheduling maintenance policy is a difficult task, which is tackled in this paper. The paper assumes that times between inspections conform to a modulated Poisson process. This assumption allows the frequency of inspection responds to the variation of the disruption state. The paper then estimates the disruption time on the basis of inspection point process observations in the framework of filtering theorem. The paper develops a unified cost structure to jointly optimise inspection frequency and replacement time for the system when the lifetime distribution of a component follows the Pareto or exponential distribution. Numerical results are provided to show the application of the proposed model.