Data-driven Virtual Monitoring and Life-cycle Management of Offshore Wind Support Structures

With the rapid advancement of offshore wind energy, efficiently managing inspection and maintenance (I&M) of wind turbine support structures has become increasingly important. Various deterioration mechanisms in the harsh marine environment accelerate the structural degradation, inducing a risk of failure which might result in substantial economic losses. Estimation of such deterioration processes involves uncertainties which often hurdle decision-making in life-cycle management planning. Collecting additional data, e.g., through structural health monitoring, can reduce uncertainties in the estimation of deterioration mechanisms, enabling more rational and informed maintenance decisions. However, collecting continuous information through monitoring systems also incurs significant sensor installation and maintenance costs. Addressing these concerns, this thesis is dedicated towards the development of a life-cycle management framework for offshore wind structures by leveraging digital twin technology with the objective of optimally allocating inspection, monitoring, and maintenance actions. The life-cycle management planning is formally formulated as a decentralized partially observable Markov decision process (POMDP), which is a principled framework for decision-making under uncertainty. In this work, maintenance decisions are informed not only by inspection and monitoring data, but also by a probabilistic digital twin. Particularly, the concept of probabilistic digital twins for virtual monitoring is presented relying on Bayesian neural networks and state-of-the-art learning algorithms, and implemented in both numerical and real-world case studies. Featuring high-dimensional state, action, and observation spaces, the formulated POMDP is solved via multi-agent reinforcement learning (MARL) algorithms, advising decisions on when and where to inspect, monitor, or maintain. The outcomes of this research not only showcase the potential of digital twins as virtual sensors, but also quantify their added benefit in life-cycle management planning. While the primary application of this research is in the context of offshore wind turbine support structures, the insights and developed methodologies can be adapted to a wide spectrum of engineering systems or infrastructures, marking a significant shift towards data-driven life-cycle management.