Reliable detection of subsurface defects in thick composite materials is critical for ensuring structural integrity in industrial applications such as wind turbine blades, aerospace components, and marine structures. This paper addresses dataset scarcity in AI-aided damage detection for thick composites using infrared thermography through a transfer learning framework leveraging finite element simulation data. Experimental datasets were obtained by conducting step-heating thermography experiments on glass-fiber-reinforced polymer (GFRP) and epoxy resin plates with artificial subsurface defects. Transient thermal analyses were performed on finite element models to mimic the actual step-heating thermography process, resulting in a large simulated dataset containing thermal videos representing the plate's surface thermal behavior during the heating-cooling process. Principal component thermography was used to extract features from both simulated and experimental thermal videos, compressing damage-related information in the raw data and enhancing the most informative features. Noise analysis on the experimental data revealed key differences compared to the simulated dataset. A U-Net architecture for image segmentation was implemented within the transfer learning framework, first pre-trained on simulated data and then fine-tuned with experimental data. The results revealed fundamental features shared across domains and demonstrated improved damage detectability in thick composite plates, especially for defects deeper than 15mm. This approach demonstrates the potential of transfer learning to improve damage detection in industrial applications involving thick composite structures, such as wind turbine blades.

This paper introduces the extreme theory of functional connections (X-TFC), a physics-informed machine learning algorithm, and tailors it to estimate the remaining useful life (RUL) of wind turbine gearbox bearings experiencing fatigue crack growth. Unlike purely data-driven methods, X-TFC embeds a physics model, based on Head’s theory in this work, into its training objective. The core of X-TFC is a random-projection single-layer neural network trained via an extreme learning machine, which requires only limited damage progression data and solves for output weights with a least-squares optimization algorithm. A composite loss function balances the network’s fit to observed degradation data against the residuals of the governing crack growth differential equation, ensuring the learned damage trajectory remains physically plausible. When applied to a vibration-based health-index (HI) dataset measured during the growth of a crack on the inner ring of a high-speed bearing in a wind turbine gearbox, X-TFC achieves near-zero prediction bias. Even when trained on only the first 10 %–20 % of the damage progression data, with sufficient physics weighting its predictions remain monotonic and smooth, delivering high prognosability and trendability. To quantify the epistemic uncertainty, we employ a Monte Carlo ensemble of independently initialized X-TFC models trained on noise-perturbed data, which yields confidence intervals around each RUL estimate and captures both model-parameter and epistemic uncertainty. In addition to a vibration-based HI, we demonstrate that the proposed framework can be directly applied to a supervisory control and data acquisition (SCADA) data-based HI measured during similar wind turbine gearbox bearing crack faults, preserving its accuracy and interpretability. This extension shows the versatility of our approach, which is applicable to bearings of multiple gearbox manufacturers, models, and ratings using only SCADA data. By integrating domain knowledge with machine learning, X-TFC offers a rapid, reliable tool for crack prognostics. Its adaptability to other bearing failure modes, such as pitch bearing ring cracks, positions X-TFC as a powerful enabler of data-driven, physics-informed asset management in the wind energy sector and beyond.

This paper studies the sensitivity of drivetrain condition monitoring system (CMS) signals to blade damage, exploring how these signals, particularly vibration, can serve as a potential tool for detection and tracking damage progression. This is achieved using a decoupled simulation approach, combining an aeroelastic solver with a drivetrain model. First, aeroelastic simulations are performed in OpenFAST, where the low-speed shaft (LSS) forces, moments, and tower top position vector are extracted and transferred to the drivetrain model. The drivetrain is modelled using the multi-body simulation environment SIMPACK. Blade damage is introduced in OpenFAST by reducing stiffness in the flap-wise or edgewise direction. The reference DTU-10MW onshore wind turbine is used as a test case. First, the impact of blade damage on LSS shear forces is analysed. Then the drivetrain response is assessed using virtual velocity sensors placed at the main bearing, rear bearing and gearbox housing. Results indicate that damage occurring in the blade mid-span region shows higher sensitivity compared to tip and root locations. A positive correlation is observed between LSS shear force and bearings side-side velocity, with higher forces leading to increased vibration. Additionally, the trend suggests that higher stiffness reduction results in higher velocity, indicating damage progression.

State-of-the-art Deep Learning (DL) methods based on Supervisory Control and Data Acquisition (SCADA) system data for the detection and prognosis of wind turbine faults require large amounts of failure data for successful training and generalisation, which are generally not available. This limitation prevents benefiting from the superior performance of these methods, especially in SCADA-based failure prognosis. Data augmentation approaches have been proposed in the literature for generating failure data instances within a SCADA sequence to reduce the imbalance between healthy and faulty state data points, which is relevant to fault detection tasks. However, the successful implementation of DL-based failure prognosis methods requires the availability of multiple run-to-failure SCADA sequences. This paper proposes a data-driven method for generating synthetic run-to-failure SCADA sequences with custom operational and environmental conditions and progression of degradation. An Artificial Neural Network (ANN) is trained with signals that represent these factors to reconstruct the SCADA signals. Then, it is used to generate synthetic SCADA datasets based on data available from a wind turbine that experienced a gearbox failure. Synthetic data sets generated are evaluated on the basis of the similarity of their signal distributions, the temporal dynamics within each signal, and the temporal dynamics among different SCADA signals with those in similar field datasets. The results show that the generated synthetic datasets are consistent with their field counterparts, with a comparatively lower diversity in their dynamic behaviour in time.

Digitalisation is one of the key drivers for reducing the costs and risks of wind energy. When considering whether to embark on a digitalisation initiative, two key questions arise. The first is what business or operational opportunities might feasibly be addressed and the second is which of the many potential aspects of digitalisation are relevant to those opportunities. In this work, we show how these questions can be answered with a use-case-driven approach, based around a survey aiming to collect and collate the main”pain points” (or everyday challenges) of people in the wind energy sector. Although the relatively low number of participants of the survey (46) means that the results should only be used indicatively, it is still possible to make some general recommendations for priorities for digitalisation efforts in the wind energy sector. Firstly, digitalisation efforts should focus both on supporting people carrying out cross-lifecycle tasks, in particular sharing data, managing data, undertaking general data analyses and accessing data. Tools to do this should deal with varying data formats and naming conventions, make metadata more accessible, define data and metadata standards, make more data publicly available and improve the quality of data. Secondly, efforts should also focus on supporting people in the wind farm operational phase, in particular with failure detection, fault diagnosis, failure rate modelling and predictive maintenance. Solutions to do this should focus on accessible and validated tools for fault detection, cloud or other data pipeline solutions for SCADA data and tools for exhaustive data documentation. Finally, digitalisation efforts should focus on better communicating and helping people become aware of existing solutions and tools, as well as on helping people to exert a stronger influence on possible solutions.

Effective and timely health monitoring of wind turbine gearboxes and generators is essential to reduce the costs of operations and maintenance activities, especially offshore. This paper presents a scalable and lightweight convolutional neural network (CNN) framework using high-dimensional raw condition monitoring data for the automatic detection of multiple wind turbine electromechanical faults. The proposed approach leverages the potential of combining information from a variety of signals to learn features and to discriminate the types of fault and their severity. As a result of the CNN layers used to extract features from the signals, this architecture works in the time domain and can digest high-resolution multi-sensor data streams in real-time. To overcome the inherent black-box nature of AI models, this research proposes two interpretability techniques, multidimensional scaling and layer-wise relevance propagation, to analyse the proposed model's inner-working and identify the signal features relevant for fault classification. Experimental results show high performance and classification accuracies above 99.9% for all fault cases tested, demonstrating the efficacy of the proposed fault-detection system.

In-service turbine monitoring is essential for maximizing the wind energy contribution to the global energy budget. Measurement of turbine shaft torque under transient wind conditions is fundamental to develop reliable condition monitoring techniques. Contact based measurements bring their own disadvantages and non-contactless measurements have many potential advantages. However, their performance needs to be validated against standard methods. This paper focuses on the development of an enhanced transient Feature Selective Validation (FSV) techniques to undertake this analysis with an emphasis on transient data processing. The nature of FSV makes it a natural technique to consider for this problem space. Open questions have existed as to how transients should be dealt with in FSV. This paper overcomes the limitations of previous approaches for step-function transient comparison and presents analytical methods to ensure the transient feature itself is considered, irrespective of how much pre- and post- transient data happens to be included.

This paper presents a simplified automated fault detection scheme for wind turbine induction generators with rotor electrical asymmetries. Fault indicators developed in previous works have made use of the presence of significant spectral peaks in the upper sidebands of the supply frequency harmonics; however, the specific location of these peaks may shift depending on the wind turbine speed. As wind turbines tend to operate under variable speed conditions, it may be difficult to predict where these fault-related peaks will occur. To accommodate for variable speeds and resulting shifting frequency peak locations, previous works have introduced methods to identify or track the relevant frequencies, which necessitates an additional set of processing algorithms to locate these fault-related peaks prior to any fault analysis. In this work, a simplified method is proposed to instead bypass the issue of variable speed (and shifting frequency peaks) by introducing a set of bandpass filters that encompass the ranges in which the peaks are expected to occur. These filters are designed to capture the fault-related spectral information to train a classifier for automatic fault detection, regardless of the specific location of the peaks. Initial experimental results show that this approach is robust against variable speeds and further shows good generalizability in being able to detect faults at speeds and conditions that were not presented during training. After training and tuning the proposed fault detection system, the system was tested on “unseen” data and yielded a high classification accuracy of 97.4%, demonstrating the efficacy of the proposed approach.

In MW-sized wind turbines, the most widely-used generator is the wound rotor induction machine, with a partially-rated voltage source converter connected to the rotor. This generator is a significant cause of wind turbine fault modes. In this paper, a harmonic time-stepped generator model is applied to derive wound rotor induction generator electrical & mechanical signals for fault measurement, and propose simple closed-form analytical expressions to describe them. Predictions are then validated with tests on a 30 kW induction generator test rig. Results show that generator rotor unbalance produces substantial increases in the side-bands of supply frequency and slotting harmonic frequencies in the spectra of current, power, speed, mechanical torque and vibration measurements. It is believed that this is the first occasion in which such comprehensive approach has been presented for this type of machine, with healthy & faulty conditions at varying loads and rotor faults. Clear recommendations of the relative merits of various electrical & mechanical signals for detecting rotor faults are given, and reliable fault indicators are identified for incorporation into wind turbine condition monitoring systems. Finally, the paper proposes that fault detectability and reliability could be improved by data fusion of some of these electrical & mechanical signals.

Non-intrusive, reliable and precise torque measurement is critical to dynamic performance monitoring, control and condition monitoring of rotating mechanical systems. This paper presents a novel, contactless torque measurement system consisting of two shaft-mounted zebra tapes and two optical sensors mounted on stationary rigid supports. Unlike conventional torque measurement methods, the proposed system does not require costly embedded sensors or shaft-mounted electronics. Moreover, its non-intrusive nature, adaptable design, simple installation and low cost make it suitable for a large variety of advanced engineering applications. Torque measurement is achieved by estimating the shaft twist angle through analysis of zebra tape pulse train time shifts. This paper presents and compares two signal processing methods for torque measurement: rising edge detection and cross-correlation. The performance of the proposed system has been proven experimentally under both static and variable conditions and both processing approaches show good agreement with reference measurements from an in-line, invasive torque transducer. Measurement uncertainty has been estimated according to the ISO GUM (Guide to the expression of uncertainty in measurement). Type A analysis of experimental data has provided an expanded uncertainty relative to the system full-scale torque of ±0.30% and ±0.86% for the rising edge and cross-correlation approaches, respectively. Statistical simulations performed by the Monte Carlo method have provided, in the worst case, an expanded uncertainty of ±1.19%.