Sensor-Based Shutdown Event Detection in Monopile-Supported Offshore Wind Turbines Using Machine Learning

Abstract

Offshore wind turbines (OWTs) present significant challenges for structural health monitoring (SHM) because of varying environmental and operational conditions. This paper proposes a machine learning (ML) approach to detect emergency shutdown events using sensor data located at the top of the foundation. Shutdowns, caused by rapid blade feathering, induce transient stresses, which can provide valuable insights into the structural integrity of the structure when measured. An integrated finite element model is used to simulate OWT load states under varying conditions, generating synthetic data for supervised learning. A k-Nearest Neighbors classifier is trained to distinguish between nominal, shutdown, and parked states. Results suggest the classifier performs well across different environmental scenarios, including unseen conditions. While low wind speeds show reduced accuracy, this is not critical as the method reliably identifies transience in significant shutdown events, achieving its primary objective. The results demonstrate the feasibility of using ML for event detection in OWTs from foundationlevel accelerometers and inclinometers, with potential applications in SHM and digital twin updating approaches.