POMDP based Maintenance Optimization of Offshore Wind Substructures including Monitoring

Abstract

Sequential decision making under uncertainty is a complex task limited normally by computational requirements. A novel methodology is proposed in this paper to identify the optimal maintenance strategy of a structural component by using a point-based Partially Observable Markov Decision Process (POMDP). The framework integrates a dynamic bayesian network to track the deterioration over time with a POMDP model for the generation of a dynamic policy. The methodology is applied to an example quantifying whether a monitoring scheme is cost effective. This complex decision problem comprised of 200 damage states is solved accurately within 60 seconds of computational time.