On the use of machine learning in damage identification of offshore jacket wind turbine: a virtual case study

Abstract

This study presents a high-fidelity, simulation-based Finite Element – Operational Modal Analysis – Machine Learning (FE–OMA–ML) framework for damage diagnosis of a jacket-type offshore wind turbine support structure. Finite-element-simulated acceleration responses are processed using Enhanced Frequency-Domain Decomposition (EFDD) to identify modal frequencies and damping ratios, which are then used as machine-learning input features. A progressive stiffness-degradation dataset covering 13 structural conditions and multiple sensor configurations is constructed to represent the gradual nature of in-service deterioration. The proposed approach is evaluated in terms of exact component-level classification, binary damage detection, layer-level localisation, ensemble-vote-based ranking, response-noise robustness, and sensor-layout optimisation. The reliability of the extracted modal features is further examined through FE–EFDD consistency checks of natural frequencies and mode shapes. The results show that the workflow retains useful hierarchical and ranked diagnostic information across different degradation levels, sensor layouts, and noisy-response conditions. However, exact component localisation becomes more sensitive as data complexity and measurement noise increase. The analysis also supports identifying efficient sensor configurations that preserve useful diagnostic performance. The proposed framework therefore provides a structured numerical strategy for evaluating modal-feature-based damage diagnosis of offshore jacket support structures.