Damage identification for jacket-supported offshore wind turbines with limited measurements

Damage may shorten the service lives of jacket-supported offshore wind turbines (JOWTs) or cause structural failure. Therefore, efficient damage identification based on structural health monitoring (SHM) data is crucial. However, data-driven approaches are often constrained by insufficient training data due to the limited measurements in real structures. To address this issue, this study proposes a damage identification scheme named DA-SHMnet for JOWTs by integrating a convolutional neural network SHMnet and a domain adaptation (DA) algorithm. An updated finite element model of JOWT is used to simulate responses under various damage scenarios, with both labeled simulation and unlabeled monitoring data for training. The study validates the proposed method through a challenging case with three tasks, specifically using acceleration responses at different locations, with various training sample sizes, and under different noise levels. The results demonstrate that DA-SHMnet can effectively transfer feature knowledge from simulation data to actual structures. With limited measurements, DA-SHMnet outperforms the original SHMnet, improving accuracy by an average of 10.8 % in different acceleration response locations, 2 ∼ 4.9 % with varying sample sizes, and by an average of 0.5 % under noise levels of 20 ∼ 100 %. The study may provide a new pathway towards damage identification with limited monitoring data.