Markovian Decision Process Modeling Approach for Intervention Planning of Partially Observable Systems Prone to Failures

Abstract

In this work we consider a system which gradually deteriorates over time. The system is fully functional in the beginning. Over time the system eventually becomes malfunctional. Once malfunctional the system must be replaced with a (new) fully functional system. There is a cost associated with this system replacement. However there is an option of repair/correction of partially deteriorated system at a lower cost. Once replaced or repaired/corrected the system is as good as new. The information about the deterioration level of the system is monitored through signals which provide only partial information. These signals are based on classification of intelligent sensors for deterioration monitoring. Signals are received as green yellow or red. The green signal indicates a system in a condition from fully functional to a predefined level of partially deteriorated system, the yellow signal indicates a system in a condition from the predefined level of partially deteriorated system to malfunctional system, finally the red signal indicates a malfunctional system. We model this system as a discrete time Markovian decision process and solve it through Linear Programming. Our work herein comprises model development and extensive numerical studies for impact of system parameters on the maintenance decisions and costs.