Interpretation of offshore wind management policies identified via partially observable Markov decision processes

The installation of offshore wind turbines, profiting from available abundant and stable wind resources, has been
steadily increasing in the last decade, yet preserving offshore wind structures in a good condition throughout their
lifetime still remains a challenge. Structural components are exposed to deterioration mechanisms (e.g., fatigue,
corrosion, among others), and far offshore, inspection and maintenance (I&M) operations can be complex and
expensive. Hence the need for efficient optimal I&M planning methods has been increased in order to control the
risk of structural failures by timely allocating inspection and maintenance interventions.
Identifying optimal I&M policies demands the solution of a complex sequential decision-making problem
under uncertainty and imperfect information. Whereas time-, condition-, or heuristic-based strategies are conven-
tionally followed in the offshore wind industry in order to alleviate the aforementioned computational difficulties,
the resulting policies statically select inspection and maintenance actions and/or consist in predefined heuristic de-
cision rules, e.g., equidistant inspections, repairs after detection inspection outcomes, which are optimized by ex-
ploring a subset out of the vast policy space. Instead, optimal management strategies can be identified via partially
observable Markov decision processes (POMDPs), relying on mathematical principles conceived for planning un-
der uncertainty [1]. POMDP policies, efficiently computed through point-based solvers, provide optimal adaptive
I&M strategies that ultimately result in substantial cost benefits compared to their state-of-the-art counterparts [2], also demonstrated in offshore wind inspection and maintenance planning settings [3].
Even if recently reported results demonstrate the benefits of implementing POMDP-based adaptive policies for the management of offshore wind assets, the interpretation and execution of POMDP-based strategies by decision-makers (e.g., designers, operators, etc.) accustomed to calendar- and/or condition-based conventional I&M approaches might be initially challenging. In this work, we analyze and interpret POMDP-based policies with the objective of accelerating their practical implementation by offshore wind asset management decision-makers. Also, we showcase the inherent flexibility and adaptability properties offered by POMDP-based policies in a typical offshore wind inspection and maintenance planning setting, in which a decision-maker opts for an action other than the one suggested in the optimal POMDP policy.