Preventive maintenance of wind turbine blades is critical for the reliable operation of renewable energy systems, yet inspection remains costly, slow, and highly dependent on human visual assessment or manual review of large volumes of drone imagery. This work presents an automated system for detecting surface defects in wind turbine blades using machine learning techniques. The proposed system implements a reproducible pipeline that covers data preparation, feature extraction, classification, and interpretation of results, using a dataset of more than 1,300 images acquired during real inspections. After data curation and balancing, the images are resized and normalized to reduce computational cost and then processed with a Histogram of Oriented Gradients (HOG) to capture local edge and texture patterns associated with erosion, cracks, and fractures. These features are used to train a Support Vector Machine (SVM), achieving an accuracy close to 90\% in distinguishing between healthy and damaged blades.

The accurate prediction of mooring and structural tension is essential for efficient structural health monitoring (SHM) of floating offshore wind turbines (FOWTs), which are exposed to extreme harsh and dynamic sea conditions. This paper presents a data-driven framework inspired by the Deep Operator Network (DeepONet) architecture to directly anticipate time-dependent mooring tension responses from time-varying platform motion. A novel cycle-consistency constraint is integrated into the traditional forward network to improve the model's physical consistency, stability, and robustness. Within this framework, the forward network acquires the mapping from the three translatory platform's motion and temporal inputs to the appropriate tension responses, while an auxiliary inverse network is developed to rebuild the original motion sequences from the predicted tensions. The cycle-consistency loss guarantees aggrement between the original and reconstructed signals, thus regularizing the training process and guiding the network towards physically consistent predictions. This approach is particularly advantageous for sparse or noisy datasets, as conventional data-driven models frequently experience overfitting and provide non-physical predictions. The framework is validated using OpenFAST's numerically simulated FOWT datasets generated under realistic environmental loading conditions. The sensing configuration consists of platform-mounted motion sensors measuring the three translational degrees of freedom (surge, sway, and heave) together with tension sensors installed at the fairlead locations of the mooring lines. Results demonstrate that the proposed cycle-consistent architecture improves prediction accuracy and stability compared to independently trained forward networks, particularly in sparse and noisy data regimes. This framework signifies a significant advancement in physics-informed, data-efficient SHM methods for FOWTs. It also provides a basis for the development of sophisticated digital twins and facilitates real-time monitoring and control of FOWTs.

Recent advances in Physics-Informed Neural Networks (PINNs) have opened new possibilities for integrating structural dynamics and data-driven learning in Structural Health Monitoring (SHM). This work presents a physics-informed framework for input load estimation and virtual sensing of offshore wind turbine support structures, where the governing dynamics of the system are embedded directly into the learning process. Unlike purely data-driven models that require extensive labeled datasets, the proposed approach leverages known physical relationships among displacement, velocity, acceleration, and external loads to enhance interpretability and generalization. The method adopts an encoder–decoder neural architecture that maps measured accelerations and strains to a reduced-order modal space before decoding the corresponding dynamic responses and reconstructing the applied loads through embedded structural dynamics relationships. Physical consistency is enforced through the equations of motion and differential constraints between displacement, velocity, and acceleration, while automatic differentiation ensures temporal consistency without requiring explicit load data during training. This hybrid approach captures the temporal and spatial evolution of loads even with limited or noisy measurements. The framework is first validated on numerical simulations of an offshore wind turbine, accurately recovering unmeasured input loads and structural responses across diverse operating conditions. It is then demonstrated using experimental vibration data, confirming its robustness to sensor noise and sparse instrumentation. Results show that the proposed physics-informed strategy can recover complex loading patterns and provide virtual measurements that are otherwise inaccessible in practice. Overall, study advances the use of PINNs for inverse input load estimation problem in SHM, offering a computationally efficient and generalizable tool for condition monitoring and fatigue assessment of large-scale energy infrastructure.

Strain gauges provide the most direct route to fatigue estimation because they allow computing the Damage Equivalent Moment through cycle counting. Their cost and the need for regular maintenance make them expensive to operate in offshore environments. As a result, strain gauges are installed on only a few fleet leaders, leaving most turbines without ground truth on accumulated fatigue. Attempts to compensate for this lack of strain data have relied on learning a mapping from 10 minute SCADA aggregates to DEM and extrapolating it to the rest of the farm, but SCADA channels are often incomplete, and their low temporal resolution removes the fast loading cycles that dominate fatigue. Proxies such as the standard deviation of power are used to approximate fluctuations in mechanical loading, but the information they provide is coarse. High frequency acceleration preserves the full vibration response and captures mechanical loading fluctuations that 10 minute SCADA cannot resolve, making it a viable surrogate when strain is absent.

The objective of this contribution is then to learn a mapping from accessible high frequency acceleration to DEM and to extrapolate it from the few SG-instrumented turbines to the full farm. The method has two stages. First, an autoencoder is pretrained in an unsupervised manner on triaxial acceleration from forty-four turbines of a single offshore wind farm. Each 10-minute record is split into shorter windows, so the encoder learns local dynamic patterns at patch level. Then, domain adversarial neural network regularization suppresses turbine specific signatures and aligns the latent space across assets, producing features that remain stable across the fleet. After pretraining, the encoder is frozen. Second, for each 10-minute interval, the windows are embedded with the frozen encoder to form a sequence of latent vectors. This sequence is passed to an LSTM trained only on the strain instrumented turbines, and the final hidden state for each 10-minutes of acceleration is fed to a small neural network for regression. One instrumented turbine is held out to test transferability.

The trained Encoder-LSTM reaches an R2 of about 0.97 on the held-out turbine, outperforming SCADA based and hand-crafted acceleration feature baselines that remain near 0.95. The latent space aligns with wind speed, pitch angle and other operational variables, confirming that the encoder extracts the relevant physical behaviour. The performance gain comes from the higher temporal resolution of the vibration data, which preserves dynamic load cycles that SCADA aggregates remove.

All results presented here are obtained from accelerometer data alone, without any need for SCADA, metocean, or additional measurements. This gives a single source workflow that avoids data synchronization and missing channel issues. A small set of strain-equipped turbines provides the ground truth needed to train the mapping, and high frequency acceleration is increasingly available as the default sensing setup across modern wind farms. The result is a farm-wide fatigue estimation framework that converts sparse strain information into full coverage and moves population-based Structural Health Monitoring closer to practical deployment.