Distribution-aware federated learning for wind turbine fault diagnosis in edge-cloud under data imbalance

Because of the cost, it is unrealistic to sample the failure state for a long time, which makes the data collected from the scenario of engineering usually extremely imbalanced. However, imbalanced training data pose a negative effect on the fault diagnosis algorithms based on the data driven. When the data are extremely imbalanced, this problem becomes more challenging. Furthermore, to reduce the deployment cost, in industrial practice, it is often required that the parameters and computation of the deployed diagnosis model should be within a certain range, which puts forward the requirement of lightweight for diagnosis model. Therefore, in this article, a novel lightweight framework for fault diagnosis with extremely imbalanced data, called HS-KDNet, is proposed. Soft labels generated by knowledge distillation can represent the similarity between categories, i.e., through to learn the soft labels, the information about all categories are considered in each update of the parameters, not only the information about the current samples. Consequently, unlike traditional data re-balancing strategies based on generating pseudo samples, we utilized knowledge distillation to suppress the adverse effects of imbalanced data for the first time. On two classical bearing datasets, the effectiveness and superiority of the proposed HS-KDNet were demonstrated, and the experimental results shown that, except for HS-KDNet, knowledge distillation can significantly inhibit the adverse effects of imbalanced data on other simple models.

Collisions between birds and wind turbines can be significant problem in wind farms. Practical deterrent methods are required to prevent these collisions. However, it is improbable that a single deterrent method would work for all bird species in a given area. An automatic bird identification system is needed in order to develop bird species level deterrent methods. This system is the first and necessary part of the entirety that is eventually able to, monitor bird movements, identify bird species, and launch deterrent measures. The system consists of a radar system for detection of the birds, a digital single-lens reflex camera with telephoto lens for capturing images, a motorized video head for steering the camera, and convolutional neural networks trained on the images with a deep learning algorithm for image classification. We utilized imbalanced data because the distribution of the captured images is naturally imbalanced. We applied distribution of the training data set to estimate the actual distribution of the bird species in the test area. Species identification is based on the image classifier that is a hybrid of hierarchical and cascade models. The main idea is to train classifiers on bird species groups, in which the species resembles more each other than any other species outside the group in terms of morphology (coloration and shape). The results of this study show that the developed image classifier model has sufficient performance to identify bird species in a test area. The proposed system produced very good results, when the hybrid hierarchical model was applied to the imbalanced data sets.

Wind Turbines and Adverse Health Effects: A Second Opinion.

There is a big wind energy potential in supplying the power in an island and most of the islands are off-grid. Due to the limited area in island(s), there is need to find appropriate layout / location for wind turbines suited to the local wind conditions. In this paper, we have considered the wind resources data of an island in Trøndelag region of the Northern Norway, situated on the coastal line. The wind resources data of this island have been analysed for wake losses and turbulence on wind turbines for determining appropriate locations of wind turbines in this island. These analyses are very important for understanding the fatigue and mechanical stress on the wind turbines. In this work, semi empirical wake model has been used for wake losses analysis with wind speed and turbine spacings. The Jensen wake model used for the wake loss analysis due to its high degree of accuracy and the Frandsen model for characterizing the turbulent loading. The variations of the losses in the wind energy production of the down-wind turbine relative to the up-wind turbine and, the down-stream turbulence have been analysed for various turbine distances. The special emphasis has been taken for the case of wind turbine spacing, leading to the turbulence conditions for satisfying the IEC 61400-1 conditions to find the wind turbine layout in this island. The energy production of down-wind turbines has been decreased from 2 to 20% due to the lower wind speeds as they are located behind up-wind turbine, resulting in decreasing the overall energy production of the wind farm. Also, the higher wake losses have contributed to the effective turbulence, which has reduced the overall energy production from the wind farm. In this case study, the required distance for wind turbines have been changed to 6 rotor diameters for increasing the energy gain. From the results, it has been estimated that the marginal change in wake losses by moving the down-stream wind turbine by one rotor diameter distance has been in the range of 0.5 to 1% only and it is insignificant. In the full-length paper, the wake effects with wind speed variations and the wind turbine locations will be reported for reducing the wake losses on the down-stream wind turbine. The Frandsen model has been used for analysing turbulence loading on the down-stream wind turbine as per IEC 61400-1 criteria. In larger wind farms, the high turbulence from the up-stream wind turbines increases the fatigues on the turbines of the wind farm. In this work, we have used the effective turbulence criteria at a certain distance between up-stream and down-stream turbines for minimizing the fatigue load level. The sensitivity analysis on wake and turbulence analysis will be reported in the full-length paper. Results from this work will be useful for finding wind farm layouts in an island for utilizing effectively the wind energy resources and electrification using wind power plants.

Community acceptance still remains a challenge for wind energy projects. The most popular explanation for local opposition, the Not in My Backyard effect, has received fierce criticism in the past decade. Critics argue that opposition is not merely a matter of selfishness or ignorance, but that moral, ecological and aesthetic values play an important role. In order to better take such values into account, a more bottom-up, participatory decision process is usually proposed. Research on this topic focusses on either stakeholder motivations/attitudes, or their behavior during project implementation. This paper proposes a third research focus, namely the ‘objects’ which elicit certain behavioral responses and attitudes—the wind turbine and parks. More concretely, this paper explores Value Sensitive Design (VSD) as way to arrive at wind turbines and parks that better embed or reflect key values. After a critical discussion of the notion of acceptance versus acceptability and support, the paper discusses existing literature on ecology and aesthetics in relation to wind turbine/park design, which could serve as ‘building blocks’ of a more integral VSD approach of the topic. It also discusses the challenge of demarcating wind park projects as VSD projects. A further challenge is that VSD has been applied mainly at the level of technical artifacts, whereas wind parks can best be conceptualized as socio-technical system. This new application would therefore expand the current practice of VSD, and may as a consequence also lead to interesting new insights for the VSD community. The paper concludes that such an outcome-oriented approach of wind turbines and park is worth exploring further, as a supplement to rather than a replacement of the process-oriented approach that is promoted by the current literature on community acceptance of wind parks.

During recent years, wind energy has moved from an emerging technology to a nearly competitive technology. This fact, coupled with an increasing global focus on environmental concern and a political desire of a certain level of diversification in the energy supply, ensures wind energy an important role in the future electricity market. For this challenge to be met in a cost-efficient way, a substantial part of new wind turbine installations is foreseen to be erected in big onshore or offshore wind farms. This fact makes the production, loading and reliability of turbines operating under such conditions of particular interest.

In the context of Industry 4.0, machine learning algorithms have been commonly used to monitor the health state of wind turbine gearboxes to avoid catastrophic failure and reduce maintenance costs. However, due to the lack of a certain category of data (i.e., healthy or faulty) and the various working conditions of wind turbines, many existing methods may not provide reliable results in practical industrial applications. To solve this problem, we create an industrial internet of things (IIoT) platform, through which a machine learning-based adaptive fault detection method for wind turbine gearboxes is proposed. The features are extracted and adapted to fine-tune the pre-trained model on newly arriving samples from different wind turbines, components, or failure modes. The adaptation performance is evaluated with accuracy, false alarm rate, and fault detection rate. Case studies are then performed using highfrequency vibration signals acquired from two megawatts (MW) onshore wind turbines. The results show that the proposed adaptive method significantly improves the fault detection performance when class distribution is not balanced, and can be easily applied to the fault diagnosis of large numbers of wind turbines. This, integrated with the IIoT platform that alleviates the shortage of computational and storage capacity in wind farms and requires less user involvement, allows for a more effective condition monitoring system.

The aerodynamic interaction between wind turbines grouped in wind farms results in wake-induced power loss and fatigue loads of wind turbines. To mitigate these, wind farm control should be able to account for those interactions, typically using model-based approaches. Such model-based control approaches benefit from computationally fast, linear models and therefore, in this work, we introduce the Dynamic Flow Predictor. It is a fast, control-oriented, dynamic, linear model of wind farm flow and operation that provides predictions of wind speed and turbine power. The model estimates wind turbine aerodynamic interaction using a linearized engineering wake model in combination with a delay process. The Dynamic Flow Predictor was tested on a two-turbine array to illustrate its main characteristics and on a large-scale wind farm, comparable to modern offshore wind farms, to illustrate its scalability and accuracy in a more realistic scale. The simulations were performed in SimWindFarm with wind turbines represented using the NREL 5 MW model. The results showed the suitability, accuracy, and computational speed of the modeling approach. In the study on the large-scale wind farm, rotor effective wind speed was estimated with a root-mean-square error ranging between 0.8% and 4.1%. In the same study, the computation time per iteration of the model was, on average, 2.1 × 10 − 5 s. It is therefore concluded that the presented modeling approach is well suited for use in wind farm control.

Power regulation of a wind farm through flexible operation of turbines using predictive control

The purpose of this research is to advance wind farm wake modeling and improve research techniques for understanding and mitigating impacts to bats due to wind farm development. Bat fatalities associated with wind turbines are an increasing concern with the widespread development of wind energy. Until now, research has focused on using biometrical and statistical techniques involving acoustic detection of bat activity and carcass surveys to develop mitigation strategies. However, there is an urgent need to develop and improve systematic strategies for smart wind turbine curtailment to mitigate bat fatalities while minimizing power loss. Unlike the perspective of biometricians that focus on bat habitat and post construction mortality surveys to evaluate impacts, the present study is motivated by the hypothesis that wind farms modify bat activity due to turbine wakes causing variations in wind speed within and around wind farms. Based on carcass survey evidence and wind farm operational characteristics, a model is proposed to describe the extent of low wind speed regions behind turbines in wind farms, that may provide preferable pathways for bats to fly when ambient wind speed is greater than the curtailment limit, where bats typically fly. To investigate the hypothesis further and detect bat activity within wind farms, a new approach is developed utilizing Doppler radar. A calibration procedure for wind farm wake modeling using a simple analytical approach and wind turbine operational data was developed to improve the identification of low wind speed regions preferable for bat activity. The proposed procedure uses a Gaussian-based analytical wake model and wake superposition model. The wake growth rate varies across the wind farm based on the local streamwise turbulence intensity. The wake model was calibrated by implementing the proposed procedure with turbine pairs across the wind farm. The performance of the proposed procedure was validated at an onshore wind farm in central Iowa, USA. The results were compared with the industry standard wind farm wake model and have shown a higher prediction accuracy in sitewide wind speed and power prediction. This new SCADA-based calibration procedure can be used to identify potential low-wind habitat for bats in the wake of wind turbines and for wind farm operational optimization. Together, along with bat activity monitoring, this framework can provide a useful tool for real-time wind farm control and smart curtailment. In order to study bat activity and their interaction with wind turbines and to investigate the feasibility of using Doppler radar for bat detection at a large wind farm, a bat detection experiment was conducted during the fall migration period within a wind farm in Iowa, USA. An X-band polarimetric Doppler radar was deployed to scan the entire wind farm and surrounding area up to 16 km. The results show the potential of X-band radar for detecting bat activity levels at a wind farm using the change in radar reflectivity and Doppler velocity information. The results can provide guidance to develop carcass survey strategies and improve understanding of bat activity within a wind farm.

This thesis presents the experimental and numerical analysis of seismic waves that are produced by wind farms. With the aim to develop renewable energies rapidly, the number of wind turbines has been increased in recent years. Ground motions induced by their operation can be observed by seismometers several kilometers away. Hence, the seismic noise level can be significantly increased at the seismic station. Therefore, this study combines long-term experiments and numerical simulations to improve the understanding of the seismic wavefields emitted by complete wind farms and to advance the prediction of signal amplitudes. Firstly, wind-turbine induced signals that are measured at a small wind farm close to Würzburg (Germany) are correlated with the operational data of the turbines. The frequency-dependent decay of signal amplitudes with distance from the wind farm is modeled using an analytical method including the complex effects of interferences of the wavefields produced by the multiple wind turbines. Specific interference patterns significantly affect the wave propagation and therefore the signal amplitude in the far field of a wind farm. Since measurements inside the wind turbines show that the assumption of in-phase vibrating wind turbines is inappropriate, an approach to calculate representative seismic radiation patterns from multiple wind turbines, which allows the prediction of amplitudes in the far field of a complete wind farm, is proposed. In a second study, signals with a frequency of 1.15 Hz, produced by the Weilrod wind farm (north of Frankfurt, Germany) are observed at the seismological observatory TNS (Taunus), which is located at a distance of 11 km from the wind farm. The propagation of the wavefield emitted by the wind farm is numerically modeled in 3D, using the spectral element method. It is shown that topographic effects can cause local signal amplitude reductions, but also signal amplification along the travel path of the seismic wave. The comparison of simulations with and without topography reveals that the reduction and amplification are spatially linked to the shape of the topography, which could be an explanation for the relatively high signal amplitude observed at TNS. Finally, the reduction of the impact of wind turbines on seismic measurements using borehole installations is studied using 2D numerical models. Possible effects of the seismic velocity, attenuation, and layering of the subsurface are demonstrated. Results show that a borehole can be very effective in reducing the observed high-frequency signals emitted by wind turbines. However, a borehole might not be beneficial if signals with frequencies of about 1 Hz (or lower) are of interest, due significant wavelength-dependent effects. The estimations of depth-dependent amplitudes with a layered subsurface are validated with existing data from wind-turbine-induced signals measured at the top and bottom of two boreholes. The experimental analysis of measurements conducted at wind farms and the advances of modeling such signals improve the understanding of the propagation of wind-farm induced seismic wave fields. Furthermore, the methods developed in this work have a high potential of universal application to the prediction of signal amplitudes at seismometers close to wind farms with arbitrary layout and geographic location.

Hydrogen farm concept: A Perspective for Turkey

Considering the increasing number of wind farms and the increasing proportion of renewable energy in the energy grid, wind farms will be more and more likely to supply ancillary services to the grid for significant periods at a time. Also, there is a trend towards overplanting offshore wind farms. These trends offers two opportunities; on the one hand, the better the ancillary services can be predicted and delivered, the more valuable they will be. On the other hand, it is an opportunity to allow the most heavily loaded wind turbines in the wind farm to operate at lower loads for a while and even out the wear-and-tear across the wind farm. Wind farm control has, understandably, been largely focused on maximizing the total power output of the wind farm, by balancing which turbines should capture the energy from the wind, wake redirection and even sinusoidal loading of the first row of wind turbines. It is also possible to take loading of the wind turbines in the farm into account, for instance using model predictive control. However, it has also been shown that fairly basic feedback control systems can already improve tracking behavior significantly. Because wind turbine controllers are in a much better position to control detailed loads on a wind turbine than the, rather low frequency, wind farm controller, the role of the wind farm controller should be assign power production to those wind turbines that are in the best position to produce at minimal cost (i.e. loads). The work presented here investigates a basic feedback control system that achieves power tracking but also redistributes loading across the wind farm to reduce the rate of damage of the whole farm and to relieve loading from the wind turbines that have accumulated the heaviest loads. The controller uses an online rainflow counting algorithm to keep track of loading and load rate and redistributes accordingly. The performance of the controller is tested using FastFarm and the interaction between the wind farm controller and the way the wind turbine controller implements the power setpoint is also investigated.

Analysis of performance is presented for wind energy conversion by a Savonius type vertical axis rotor configured for generation of electrical power. The technical approach capitalizes on the high torque property of a large Savonius rotor at rotation rates dictated by the ratio of the wind speed to the peripheral speed of the rotor vanes. This property enables generation of AC power by means of a multiplicity of generators positioned around the periphery of the rotor that are sized and have gear ratios matched to selectable ranges of wind speed. The system can produce electrical power over a range of wind speeds and capacities that are significantly beyond the practical and economical operating envelope achievable with horizontal axis wind turbines (HAWT). A performance comparison is made with conventional HAWT rated at 1.5 megawatts. Potential performance is examined for systems with rated capacities in excess of 6 megawatts with the feature of increased capacity at higher wind speeds. Adaptation of this technology to offshore sites and the implications for cost-effective performance in wind farms are discussed. Introduction This paper presents an entirely new concept, referred to as SavRot, for implementation of vertical axis Savonius rotor technology that will have significant advantages over highly developed horizontal axis wind turbines (HAWT). This new adaptation of mature technology can provide systems that are quieter (noise from both vanes and generators); pose no danger to birds in flight; are easily maintainable by ground crews; have a greater tolerance for storm conditions; capture gusty energy independent of wind direction; can have higher power generation capacity; and have lower capital and operating costs. The inherent design flexibility of SavRot technology enables a broader range of applications than are suitable for HAWT. The technology is especially attractive for offshore wind farms that need higher power generation capacity at single sites, endurance and power production in storms, lower cost of ownership, and fewer hazards to navigation. Changing conditions in the marketplace are considered insofar as they bear on the demand for and specification of more capable and less expensive equipment than now is available for wind energy conversion. An overview is provided of technological impediments to the capability of HAWT systems to respond to emerging market demands. A performance comparison is drawn between HAWT rated at 1.5 megawatts (MW) and an equivalent SavRot system. Performance potential of larger SavRot systems is illustrated for a scaled set of rotors with standard capacity ratings up to 6 MW and power generation capability greater than 10 MW at higher wind speeds. Consideration is given to the application of large SavRot systems in the offshore wind farm market. Wind Energy Market The U.S. wind energy conversion market is underpinned by the Federal mandate to have renewable sources provide 20% of national electricity consumption by 2030. Wind is the most cost effective source to meet the demand. The wind energy conversion goal for 2030 has been set at a total capacity 300,000 MW by the U.S. Department of Energy (2008). Total U.S. capacity at the end of 2006 was 11,500 MW. At the end of 2009 it was 35,000 MW (American Wind Energy Association, 2010), or about 11.7% of the 2030 goal. To meet the goal, the required annual increase of installed capacity has been estimated by the Department of Energy (2008) to level off at 16,000 MW per year beginning in 2018, at which time the plan calls for 110,000 MW to be in place. Virtually all of the growth in the U.S. wind energy market has been in the form of terrestrial wind farms. The wind farm typically is an aggregation of HAWT systems tied to a large scale electrical network that has primary power supplied by multiple conventional power plants. Wind farms have been subsidized on many levels from the inception of the wind energy conversion industry. The subsidies are an essential ingredient for counteracting low return on investment.

Rapid measurement of classification levels of primary macronutrients in durian (Durio zibethinus Murray CV. Mon Thong) leaves using FT-NIR spectrometer and comparing the effect of imbalanced and balanced data for modelling