ABSTRACT
The stability and the dynamic response of the offshore wind turbines can be affected by the erosion of the foundation, referred to as scour. As an alternative to applying costly scour protection, monitoring of scour development is often performed based on a shift in the fundamental natural frequency of the structure. This paper considers scour detection as an outlier detection problem using features that do not require system identification. A state of the structure is classified as healthy or damaged based on three damage sensitive features extracted from the response signals: auto-regressive model coefficients, unique entries of the response covariance matrix and transmissibility functions. The examination is performed using a numerical monopile model supported by Winkler springs and loaded by random excitation. The results show that, among the presented features, the auto-regressive model coefficients display the highest sensitivity to scour, as the use of this feature allows for detection of scour corresponding to 3% of the embedded monopile depth even in the presence of noise.
