This thesis proposes a comprehensive methodology to determine the optimal condition-based maintenance (CBM) policy for partially observable deteriorating systems. Traditional solution methodologies often fail to solve larger-sized systems due to the curse of dimensionality. To overcome this, the study integrates the Newsvendor inventory model into an observed-state Linear Programming (LP) formulation, where the system's underlying dynamics are formulated as a Markov Decision Process (MDP). This integration removes the spare part quantity and reduces the decision space only into the decision of timing of the intervention. This structural breakthrough enables the exact optimization of larger-sized problems, scaling the mathematical optimization up to 20 components. In the literature, a system-wide deterioration probability and a system-wide sensor that gives partial information about the underlying deterioration is generally assumed. In addition to the literature, the thesis mathematically quantifies the value of information by explicitly assessing the financial impact of utilizing component and deterioration specific deterioration probabilities, as well as deploying component-level sensors. Furthermore, a robust discrete-event simulation is built to find the optimal policy by evaluating threshold-based rules. Validated against the exact LP model, this simulation empowers the resolution of systems even larger than the 20-component limit. Crucially, the simulation is extended to solve cases that remain analytically challenging to represent with traditional exact methodologies. One of these cases studied in the thesis is systems that can deteriorate multiple levels at a time, an addition to the literature that generally assumes systems that can generate a single step in a time.

Keywords

Condition Based Maintenance, Partially Observable Systems, Value of Information, Reliability Engineering and Simulation Modeling.