Abstract :
[en] 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.
