Optimal Replacement for a Partially Observable System Subject to Dependent Failure Modes

ABSTRACT Condition‐based maintenance typically considers the failure mode (i.e., soft failure) that evolves from deterioration. In many real applications, however, another failure mode, also known as hard failure, occurs occasionally and stops a system immediately. Moreover, the time to hard failure usually depends on age and deterioration. It is therefore of great importance to combine both age and deterioration information to make preventive maintenance decisions, considering dependent failure modes. Such a problem becomes highly complicated when deterioration can only be partially observed. This paper develops a replacement model for a partially observable system subject to dependent hard and soft failures. The optimization problem is addressed within the partially observable semi‐Markov decision process framework. Under certain conditions, the optimal policy has a monotonically decreasing control‐limit form, distinguishing the proposed policy from the traditional multivariate Bayesian control chart. An algorithm is developed based on a discretization approach to find the optimal solution. The obtained adaptive Bayesian control chart provides new insights into maintenance implementation. The results of numerical examples illustrate the superiority of the proposed policy.