Scour identification around offshore wind turbine monopile foundations using an unsupervised anomaly detection approach

This paper presents a novel data-driven framework for detecting scour damage around monopile foundations using acceleration data collected from an offshore wind turbine (OWT) tower under healthy conditions. The framework consists of multiple steps, including data generation, preprocessing, feature engineering, unsupervised Machine Learning (ML) model training, and performance evaluation. Validation is conducted against numerically-generated data using OpenFAST, with foundation stiffness modelled through a Soil-Structure Interaction (SSI) model developed in MATLAB. The healthy data are pre-processed and segmented prior to frequency-domain feature extraction. A One-Class Support Vector Machine Algorithm (OCSVM) is subsequently trained using features obtained through an unsupervised feature selection algorithm. The model performance is evaluated under real-time operational conditions expected offshore. Results indicate the framework's effectiveness in identifying scour states, particularly those exceeding a depth of 3 m for the given foundation analysed. The effectiveness of OCSVM for scour detection is appraised in terms of its potential for real-time condition monitoring of OWT foundations. • A data-driven scour detection for offshore monopiles using tower vibration data. • Integrated OpenFAST and MATLAB SSI model to simulate scour effects on OWT dynamics. • Real-time scour detection under realistic offshore operational conditions. • Scour detection framework with only healthy data across varied conditions. • Strategic training data analysis to optimize scour monitoring of OWT foundations.