Posters

Abstract

Monitoring of structural health has become widespread in the wind industry and is of vital importance to reduce the cost of maintenance, extend the lifetime of the structures and ultimately reduce the environmental impact and cost of energy. For floating turbines, lowering costs is crucial to ensure the feasibility and viability of the concept as a means of capturing energy. Monitoring the structural health of a wind turbine can help extend life and reduce costs for floating wind. As part of a pilot project, 4Subsea has installed two sensors on a floating wind turbine in the North Sea. One motion sensor capable of observing accelerations and rotations in/about three axes, as well as four strain sensors placed equidistantly around the circumference of the tower base. Initial trials showed a very strong, linear correlation between strain, pitch and wave height, leading to the idea that it may be possible to calculate fatigue damage based on sea state parameters alone. To achieve this task, we have trained two kinds of Recurrent Neural Networks (RNN) on five years’ worth of data. A significant effort went into gathering data, cleaning, and ensure that the dataset is split into test and train sets without data leakage between the two. In addition to optimizing feature selection and network structure, we have tested with different weather sources and included turbine operational data (SCADA) as features to the RNN. The results show that it is indeed possible to obtain a very good estimate of fatigue damage based on sea state and/or SCADA data alone. This demonstrates that the structural health of a floating wind farm can be monitored more effectively by reducing the number of turbines requiring monitoring or the duration of the monitoring periods.