Multibody simulation (MBS) is an established method for modeling wind turbines for purposes such as load calculation. While capable of modeling wind turbines at high levels of detail, large MBS models can be computationally expensive. This makes deployment for real-time critical use cases like digital twins difficult. Machine learning models can be a faster alternative because they only capture relevant details and scale well with parallelization. In this paper, a real-time capable surrogate model is developed with the goal of capturing the operating mode and the tower oscillations of an MBS model. Firstly, several machine learning models based on different architectures are being scaled so that they can be evaluated within the available computational budget. The selected models are being trained on the MBS results. The trained models are being run in a loop with the turbine’s controller and an optional numerical tower oscillator to create the surrogate model. All tested models show good convergence behavior and high R 2 values during training. However, running the models in the surrogate environment proves more difficult as some models diverge from the expected results. The best performing model in the surrogate environment is determined using quantitative criteria.

In this paper, we address input load/generation uncertainty in the stochastic economic dispatch and unit commitment for forecasting and then scenario generation purposes. We propose a constrained multi-variate linear regression formulation to model the joint spatial-temporal distribution of forecasts of loads, and renewable energy productions such as solar cells and wind turbines at various locations. In this model, l 0 and l 1 constraints are added to simulate LASSO and Ridge regression models. In order to generate robust solutions for stochastic economic dispatch and unit commitment, we eliminate heavy-tails from forecasted errors using extreme value theory. We then employ a Monte-Carlo sampling technique to generate scenarios. Finally, we propose a new exact model for scenario reduction. For this paper, we used data published at NYISO and ERCOT websites. Our numerical results show how energy loads and renewable energies can be forecasted using our methodologies.

The airfoil DU91-W2-150 was investigated in the Low Speed Low Turbulence Tunnel at the Delft University of Technology to study unsteady aerodynamics. This experimental study tested the airfoil under a wide range of angles of attack (AoA) from 0 ◦ to 3 1 0 ◦ at three Reynolds numbers ( Re) from 2 × 1 0 5 to 8 × 1 0 5 . Pressure on the airfoil surface was measured and Particle Image Velocimetry (PIV) measurements were conducted to capture the flow field in the wake. By examining the force coefficient and comparing the wake contours, it shows that an upwind concave surface provides a higher load compared to a convex surface upwind case, highlighting the critical role of surface shape in aerodynamics. When comparing separation at specific locations along the chord for all three Res, it is observed that as Re increases, separation tends to occur at lower angles of attack, both for positive stall and negative stall. The examination of the aerodynamic force variation indicates that, during reverse flow, fluctuations are more pronounced compared to forward flow. This is owing to separation occurring at the aerodynamic leading edge (geometric trailing edge) in reverse flow. In terms of vortex shedding frequency, the study found a nearly constant normalized Strouhal number ( St) of 0.16 across various Res and AoAs in fully separated regions, indicating a consistent pattern under these conditions. However, a slight increase in St, between 0.16 and 0.20, was observed for AoAs exceeding 180 degrees, possibly due to the convex curvature of the airfoil in the upwind direction. In conclusion, this research not only corroborates previous findings for small AoAs, but also adds new data on the aerodynamic behavior of the DU91-W2-150 airfoil under large AoAs, offering various perspectives on the effects of surface curvature, Re, and flow conditions on key aerodynamic parameters.

Offshore wind has an important role in achieving net zero targets and is a vital contributor to addressing the energy crisis. Continued expansion in offshore wind is driving industry and governments across the world to explore deeper and more exposed sites. Standard wind measurements using meteorological masts are more challenging, or even impractical at such sites. As a result, floating LIDAR and physics-based mesoscale models are increasingly being deployed as cost-effective solutions for resource assessment. This study investigates the performance of a meso-microscale coupled Weather Research and Forecasting model by comparing against using 15-months of in-situ wind data collected from two floating LiDAR systems deployed in the Celtic Sea, Southwest UK. Wind speed measurements from model and measurement were compared directly. Vertical wind speed gradient estimates, chosen to quantify how the two data sets predicted wind shear, were also compared, along with estimates of energy generation, as predicted by the two data sets. At approximate turbine hub height (150 m), the model predicted mean wind speeds are 2.38% and 2.81% lower than the measurements and 4.96% and 4.51% lower at the P95 wind speed, for the two measurement locations. However, the differences are greater at wind speeds faster than the turbine cut-out speed. The inverse is true at for windspeeds, below the turbine cut-in speed, where model predicted wind speeds are greater than the measurements. As these high and low wind speeds do not contribute to energy generation, the differences between model-predicted and measurement-predicted energy production are smaller, with model estimates lower by 0.56% and 1.32% for the two measurement sites. Model calculated wind speed gradients were notably different from the measurement-calculated wind speed gradients. In particular, the 90 th percentile value is 35% lower than that predicted by measurements. Attempts to account for this when calculating energy generation suggested that the model estimates were lower than measurements by a much lower margin of 0.6% and 1.37%for the two floating LiDAR locations. This work identifies differences between numerical model predictions and floating Lidar measurements for a proposed offshore wind site. Of the parameters considered, the energy production estimates show the least difference. This helps inform studies looking to accurately evaluate wind resource at more exposed sites. However, usage of wind data that specifically relies on accuracy for large or small wind speeds, such as engineering design, or weather window calculations will be required to consider the impact of the observed differences on their results.

Due to the nonlinearity, non-stationarity and noise interference of the rolling bearing fault signals of wind turbines, the extraction of features is challenging. This paper proposes a fault diagnosis method for wind turbine bearings based on the improved dung beetle optimizer (IDBO) optimized Long Short-Term Memory (LSTM) network. Firstly, Levy flight strategy and T-distribution perturbation strategy are integrated into the traditional dung beetle optimizer (DBO) to optimize the algorithm. The optimized IDBO algorithm solves the problems of low convergence accuracy and easy entrapment into local optima in the traditional dung beetle algorithm. Secondly, the IDBO algorithm is combined with the long short-term memory method to build the IDBO-LSTM fault diagnosis model. The advantage of fewer parameters in this algorithm enables the model to reduce the overall number of parameters and computational complexity while ensuring the accuracy of temporal feature extraction. Finally, this pa-per takes the bearing open dataset of Case Western Reserve University as an example to compare the proposed method with other models through optimization tests. The experimental results show that the IDBO-LSTM model outperforms other models, with a fault diagnosis accuracy rate of 98.3%. This demonstrates the superiority of the proposed improved dung beetle optimizer and effectively improves the accuracy of fault diagnosis for rolling bearings of wind turbines.

not-yet-known not-yet-known not-yet-known unknown For the current visual detection methods of wind turbine blade defects, their detection models are usually excessively large, making them difficult to achieve a balance between model accuracy and inference speed. To tackle this problem, this paper introduces a lightweight wind turbine blade defect detection network, GCB-YOLO, which attempts to maintain a high detection accuracy and simultaneously achieve a rapid detection speed. At the beginning, a GhostNet network is employed to replace a portion of the YOLOv5s backbone network responsible for feature extraction. This replacement serves to reduce the network’s parameter size and computational load, thereby achieving compression of the feature extraction network. Subsequently, a CA (Channel Attention) mechanism is incorporated into the backbone network, which enhances its ability to focus on small-sized defects. Finally, the neck network PANet is substituted with a Bifpn network, bolstering its ability to discern small-sized defects. A series of validation experiments were conducted using an image dataset gathered from real wind farms. The result showed that the GCB-YOLO exhibited a reduction of 46.2% of model parameter number compared to that of the YOLOv5s. The improved model only has a 7.5MB volume. Hence, in GPU computation mode, the image detection speed reached 115.3 frames per second. More importantly, the proposed method achieved an mAP@0.5 of 94.72%, simplifying the deployment on edge computing devices and simultaneously meeting the real-time defect detection requirement with a sustained high level of detection accuracy.

The monitoring of wind turbine blade root load (BRL) has always been a valuable yet highly challenging problem. BRL monitoring is often measured directly using strain sensors, which is expensive to install and poses safety risks. Thus, it is necessary to explore an indirect measurement method that establishes a mapping relationship between BRL and easily measurable data. This paper presents a non-intrusive BRL monitoring method based on the vibrational signals at bearings (including main bearing, gearbox bearing, and generator bearing) and a new proposed AE-LSTM-Attention neural network. The proposed neural network incorporates autoencoder (AE) pre-training process, long short-term memory (LSTM) neural network, and channel attention mechanism to overcome difficulties in extracting and fusing long-term, high-frequency, and multi-channel signal features. Additionally, considering the wind turbine transmission mechanism, an optimal LSTM step number selection criterion is proposed. The wind turbine’s physical feature is innovatively bridged with the neural network training parameters, e.g., setting the transmission ratio of the gearbox as the number of LSTM steps, by which the most cost-effective prediction performance can be achieved. Experiments were conducted using data from actual operating wind turbines, and the results show that the proposed method can successfully achieve non-intrusive monitoring of BRL characteristics and outperform traditional methods in terms of predictive performance. The mapping relationship between wind turbine vibration signals and BRL characteristics can be established efficiently and accurately.

A coupled medium-fidelity drivetrain model is developed and implemented in OpenFAST for a 10-MW land-based reference turbine. The implementation is verified against a fully coupled multibody wind turbine model, including a detailed drivetrain. The new model can simultaneously and accurately estimate main bearing loads and represent elastic bending of the drivetrain. It has low computational cost, useful for early design phases, sensitivity analyses and complex systems like wind farms (where computational expense must be expended elsewhere). Here, the model is extended to a monopile offshore wind turbine and used to investigate sensitivity of predicted main bearing rolling contact fatigue to different synthetic turbulence models. Large eddy simulations (LES) intentionally targeting stable, neutral, and unstable atmospheric conditions at below-, near- and above-rated wind speeds, were used as a reference. The turbulence models recommended by IEC, the Mann spectral tensor model and the Kaimal spectral model with exponential coherence, were fitted to the LES data. Additionally, a constrained turbulence generator, PyConTurb, based on LES data, was applied in the aero-hydro-servo-elastic simulations. Taking PyConTurb as the baseline, the Kaimal model significantly underestimates fatigue of the downwind main bearing, with between 10 and 40% less damage. The Mann model also underestimates the downwind main bearing fatigue with up to 30%. The upwind main bearing damage is driven by mean loads, and differences between models are less significant, although the trends are similar. Reasons for these discrepancies are investigated and attributed to differences in spatial and temporal variations among the turbulence models.

This article presents an analytical model to estimate wind velocity uncertainties in ground-based (dual) Doppler lidar measurements. The model follows the principles of uncertainty propagation as recommended by the Guide to the Expression of Uncertainty in Measurements. Key-input quantities of the measuring model considered uncertain include elevation and azimuth angles and focus distance, or range. The uncertainty model also accounts for bias and random errors originating from hardware components and data processing techniques. Uncertainty correlations within a single lidar and between instruments in a dual-lidar system are addressed. The measurement model assumes perfect spatio-temporal synchronisation between the lidar instruments while probing a non-turbulent wind inflow described by a vertical shear model. Results from the analytical solution are verified using Monte Carlo simulations, obtaining very good agreement from the comparison. The pattern of the uncertainty distributions is predominantly influenced by the relative positioning of the measuring system(s) and the intended measurement point(s). The magnitude of the uncertainty distributions is principally determined by the intrinsic uncertainty of the lidar, modulated by the set of relevant uncertainty correlations.

Wind turbines operating in unsteady and complex environments often encounter asymmetric and unsteady vortex flows on the leeward side of the blades, leading to airflow separation and dynamic stall. These issues significantly impact the aerodynamic performance and energy efficiency of wind turbines. However, there is a lack of research regarding the regulatory relationship between airfoil parameters, trailing-edge flap oscillation angles, and aerodynamic parameters. To address this research gap, this study employs a multi-block overlapping grid technique and a custom-programmed method to simulate active flow control on wind turbine blades by coordinating the motion of the main wing and trailing-edge flap during pitching oscillations. Aerodynamic parameters of airfoils were computed to investigate the impact of trailing-edge flap deflection direction, deflection amplitude, and other parameters on the wind turbine blade thrust, torque, and aerodynamic characteristics under deep stall conditions. An optimized aerodynamic performance curve of the blade was produced. The results show that when a deep stall occurred and the wind speed was low, the lift coefficient of the airfoil can be increased by approximately 23% with the amplitude αamp1=40° and the same deflection direction; moreover, the tangential force and torque value on the blade concurrently increased. Adjusting the motion of the main wing and trailing-edge flap reduced the thrust and load fluctuation of the blade by changing the flap amplitude value. Additionally, the operation efficiency of the wind turbine improved. These results can provide a reference for the active flow control and application of wind turbine flaps.

A versatile and widely used tool in atmospheric research and operational forecasting is the Weather Research and Forecasting (WRF) model. The WRF model utilizes advanced algorithms and physics parameterizations to simulate complicated atmospheric phenomena. However, this model relies on a configuration that encompasses spin-up time, different Planetary Boundary Layer (PBL), and Land Surface Model (LSM) schemes to describe the physics near the surface. This work aims to optimize the configuration of the WRF model by evaluating the impact of the spin-up period, PBL, and LSM schemes to obtain the best representation of the wind field to calculate the power production of a wind farm located in the complex terrain of La Rumorosa, Baja California, Mexico. This site was chosen because of the availability of data, the challenging conditions of the terrain complexity, and severe changes in the wind direction. Results indicate that one day spin-up is the best option to predict the wind speed for one month or even one year. Moreover, the best statistical metrics are obtained when the Mellor-Yamada-Nakanishi–Niino Level 2.5 (MYNN2) PBL scheme is coupled with Noah-Multiphysics LSM. The best set-up for the WRF model was used to simulate the performance of a wind turbine in the farm with excellent results.

With an increasing share of the electricity grid relying on wind to provide generating capacity and energy, there is an expanding global need for historically accurate high-resolution wind data. Conventional downscaling methods for generating these data based on numerical weather prediction have a high computational burden and require extensive tuning for historical accuracy. In this work, we present a novel deep learning-based spatiotemporal downscaling method, using generative adversarial networks (GANs), for generating historically accurate high-resolution wind resource data from the European Centre for Medium-Range Weather Forecasting Reanalysis version 5 data (ERA5). We show that by training a GAN model with ERA5 low-resolution input and Wind Integration National Dataset Toolkit (WTK; [[1]](#ref-0001)) data as the high-resolution target, we achieved results comparable in historical accuracy and spatiotemporal variability to conventional dynamical downscaling. This GAN-based downscaling method additionally reduces computational costs over dynamical downscaling by two orders of magnitude. GANs are trained on data sampled from the contiguous United States (CONUS), selected to provide a diverse sampling of terrain conditions, and validated on data held out from training, as well as observational data. This cross-validation shows low error and high correlations with observations and excellent agreement with holdout data across distributions of physical metrics. We additionally downscaled the members of the European Centre for Medium-Range Weather Forecasting Ensemble of Data Assimilations (EDA) for 2012–2015 and 2019–2023 to estimate uncertainty over the period for which we have observational data. We applied this approach to downscale 30-km hourly ERA5 data to 2-km 5-minute wind data for January 2000 through December 2023 at multiple hub heights over Ukraine, Moldova, and part of Romania. The geographic extent was motivated by the urgent need for planners in Ukraine to rebuild and decentralize the grid, which has been severely damaged by the conflict between Russia and Ukraine. Comparisons against observational data from the Meteorological Assimilation Data Ingest System (MADIS) and multiple wind farms show comparable performance to the CONUS validation. This 24-year data record is the first member of the super resolution for renewable energy resource data with wind from reanalysis data dataset (Sup3rWind).

Real-time, proportional deployment of an Active Gurney Flap (AGF) located near the trailing edge of the wind turbine blade on the pressure surface is used to modulate aerodynamic performance. Wind tunnel experiments are conducted using an AGF model based on the geometry of a commonly used wind turbine blade over a range of Re numbers from 160 ,000 to 414 ,000, angle of attack (AOA) from - 1 0 ◦ to 1 5 ◦ , and AGF deployment positions from 0 ◦ to 1 3 5 ◦ . The torque required to raise and lower the AGF was measured and found to be approximately C T = 0 . 1 , independent of the deployment rate and the Re number. Aerodynamic force measurements demonstrate that the deployment of the AGF is capable of improving the lift and pitch moment performance with an increased lift-to-drag ratio. Aerodynamic changes are proportional to the AGF deployment angle, with a maximum effect observed at 7 0 ◦ to the chord. The aerodynamic changes occur instantly as the flap is deployed, although sudden changes in aerodynamic loading induce mechanical vibration of the wind tunnel model on the supporting sting.

Wind energy is one of the main renewable energy sources in the current energy transition. Due to ever more and ever larger wind turbines (WT), the requirements for WT operation become more complex. Model predictive control (MPC) for WTs shows the potential to handle these requirements and conflicting control objectives in a single optimization-based controller. Recent research has widely investigated MPC for WT in simulation, but mostly lacks experimental validation. This work aims to experimentally validate MPC on a full-scale WT under real conditions. To this end, we combine an Extendend Kalman Filter for nonlinear state estimation with robust linear time-varying MPC. We evaluate the proposed control algorithm in terms of time-domain performance and power curve in simulation. However, the main contribution of this work is the experimental validation on a 3MW WT in Northern Germany with a total duration of 3h continuous full access of the controller. We were able to demonstrate stable operation of the proposed MPC in the upper partial load regime, transition regime and lower full load regime, at measured wind speeds between 4.76m/s and 13.06m/s, inside and outside the wake shadow of another WT. The power curve determined in simulation shows comparable results to a reference feedback controller. The MPC formulation combines several control objectives in a single optimization problem, yet the tuning effort still remains complex. In future work, we plan to reduce the complexity of the control loop based on this experimentally validated MPC. We provide our experimental data at https://github.com/rwth-irt/MPC_WT_experiment.

A main bearing (MB) preforms the critical role of supporting the rotor and whole drive train in a wind turbine. It endures very heavy and complex load coming from the rotor, meanwhile, high reliability in whole operation life is required for the high cost of MB replacement. Taking a spherical roller bearing (SRB) MB as an example, the paper analyzed complex load characters of a SRB MB configuration and discussed how to simplify the complex load in MB design works, then the key aspects influencing the MB performance were discussed, including raceway performance and shrink fit. Further more, thresholds of raceway contact pressure occurrence, fitting contact pressure and hoop stress were presented based on the state of art experiences. The paper gives an overview of SRB MB design for field operation performance.

As a commodity, wind energy is typically traded in ex-ante time frames and is dependent on forecasts to secure physical positions up to 36 hours after gate closure for trading in Day-Ahead markets. Wind energy is traded in discrete quantities, however it is generated from an intermittent and variable resource. Deterministic forecasts are preferred for energy trading as the most compatible solution to provide a defined forecast quantity. However, deterministic forecasts cannot capture the stochastic nature of the underlying power source and are therefore sub-optimal. Ensemble based forecasts have the potential to reduce forecast error by accounting for uncertainties not captured in deterministic models. However, ensemble forecasts are not always available at the vertical levels at which wind turbines operate. Therefore, a method is needed to apply ensemble information to turbine hub heights for energy forecasting purposes. This paper presents a novel machine learning based method that translates the perturbations from a localised Numerical Weather Prediction model’s 10m wind speed component to an ensemble energy forecast at 100m. The extrapolated ensemble based forecast has improved the forecast accuracy by 9% when compared to the deterministic output. The findings will have important implications for future energy trading, transmission system operation and meteorological forecasting.

Computationally efficient frequency-domain models can play a very important role in facilitating conceptual design optimization of floating wind turbines (FWTs). However, achieving sufficient accuracy in such models is challenging due to the nonlinear variation of the aerodynamic loads, particularly the interaction between the floating platform motions and the controller. Building on previously proposed approaches from the literature, this work implements and improves upon three methods to evaluate the influence of rotor dynamics on FWTs dynamics in frequency domain. The investigated methods rely on: coupled fixed-nacelle simulations in turbulent wind; decay tests in steady wind; and linearized analytical expressions of the steady state aerodynamic loads. The main objective is to assess the suitability of these methods for future optimization of the floating platform and the mooring system. The various techniques are compared through a case study of three semi-submersible FWTs with increasing rotor size. While all approaches have good accuracy below-rated wind speed, only the decay test approach provide good estimates of the wind-induced global responses across all tested conditions.

Low level jets (LLJs) describe conditions in which the wind speed reaches a local maximum with respect to altitude near the surface, and have been observed intermittently in the US Mid-Atlantic offshore environment. LLJs pose unique operating conditions for future wind turbines operating in the region, presenting negative shear and locally strong veer, but they are not typically considered in existing turbine standards. This work builds upon recent research that explains the formation and evolution of U.S. Mid-Atlantic LLJs through a simple analytical governing equation. We generate several LLJ inflow conditions with varying jet characteristics based on this analytical model and create monotonically-sheared (MS) analogues with constant veer in order to assess the impacts of the LLJ on turbine performance and loading. Using aeroelastic simulations with these inflow conditions on the IEA 15MW reference turbine, we find that the LLJ leads to a greater range of tower top pitching and yawing moments, which could contribute to larger accumulated structural fatigue in components compared to monotonically-sheared inflow. These preliminary results demonstrate a path toward a unified set of test cases for low-level wind maxima that can inform International Electrotechnical Commission standards related to offshore wind turbine design.