Limited Memory Influence Diagram Framework for Modeling Condition-Based Maintenance for a System Subjected to Degradation

Limited memory influence diagrams are useful frameworks for representing and analyzing decision-making under uncertainty and offer the advantage of modeling decision problems with large and complex domains. This paper develops a limited memory influence diagram framework for finding the optimal maintenance strategy for a deteriorating system in the context of condition-based maintenance. The degradation process of the system is modeled by a discretized gamma process. A dynamic Bayesian network framework is applied to describe and update the system state over time based on its degradation process. Inspections are performed at regular intervals corresponding to the time between time slices in the dynamic Bayesian network framework. At each inspection, the system may undertake preventive maintenance activities based on the results from system condition monitoring. The dynamic Bayesian network is extended to a limited memory influence diagram, which additionally includes decision nodes and utility nodes. The decision nodes indicate maintenance actions and the utility nodes indicate the maintenance costs associated with maintenance activities. The decision on repair actions is based on inspection results, and the system is renewed after a given number of inspections. The optimal maintenance strategy is determined so that the expected utility over all possible strategies is maximized. This paper proposes a comparison of optimized maintenance strategies for the system for different inspection time intervals. Numerical examples are given to illustrate the results.

Keywords: Limited Memory Influence Diagram, Dynamic Bayesian Network, Preventive Maintenance, ConditionBased Maintenance, Maintenance Optimization.