Wind turbine fault data prediction

Abstract

Disclosed is a method, performed by an electronic device, for providing fault data associated with a component of a wind turbine, thus enabling fault prediction and detection. The method comprises obtaining wind turbine data having a spectral element. The wind turbine data is associated with the component. The method comprises obtaining references associated with the component. The method comprises overlaying the references onto the wind turbine data. The method comprises determining the fault data by applying a recognition model to the wind turbine data with overlayed references.

Description

[0001] WIND TURBINE FAULT DATA PREDICTION

[0002] The present disclosure pertains to the field of wind turbine monitoring. The present disclosure relates to a method for providing fault data associated with a wind turbine component and related electronic device.

[0003] BACKGROUND

[0004] A wind turbine has a control and monitoring system (CMS), that is designed to improve the efficiency and effectiveness of the operations of the wind turbine. One way efficiency is improved by the CMS is through the detection of failures of components of the wind turbine and the associated systems. A wind turbine’s CMS may indicate to a user or operator of the system about a component fault, and therefore intervention such as repair may be needed to restore the wind turbine to proper functionality.

[0005] SUMMARY

[0006] A classifier can be used to categorize certain patterns as fault or no fault. However, there is a number of non-faulty patterns that get erroneously categorized as faults (such as false positive faults).

[0007] It may be seen as an object of the present disclosure to improve the reliability of the detection and prediction of failures of or within one or more components of wind turbines.

[0008] Accordingly, it would be a benefit to provide an electronic device and a method that may detect and predict failures within a wind turbine and / or a wind turbine system. Accordingly, it would be a benefit to provide an electronic device and a method for controlling an operation of a wind turbine, which mitigates, alleviates, or addresses the existing shortcomings and detects and predicts faults or failures within the wind turbine, e.g., via the CMS.

[0009] Disclosed is a method, performed by an electronic device, for providing fault data associated with a component of a wind turbine, thus enabling fault detection and prediction. Optionally, the method comprises obtaining wind turbine data having a spectral element. Optionally, the wind turbine data is associated with the component. Optionally, the method comprises obtaining references associated with the component, such as predetermined references, such as a static predetermined references. The method comprises overlaying the references onto the wind turbine data. The method comprises determining the fault data by applying a recognition model to the wind turbine data with overlayed references.

[0010] The references are not obtained or derived from the obtained wind turbine data but are obtained as described in examples herein. The references may be static references or predetermined static references. For example, each reference may be defined by a value, a distribution of values, a graphical representation or in other ways. The references remain unchanged over a given period of time, for example the references may remain the same over periods of days, months or years or may remain the same as long as the component remains unchanged, i.e. the component is not replaced or repaired.

[0011] Specifically, the step of obtaining wind turbine data comprises repeatedly obtaining wind turbine data such as during operation of the wind turbine. The step of obtaining the references may be performed only a single time or the references obtained are the same each time they are obtained from a storage or database. The steps of overlaying the references onto the wind turbine data and the determining of the fault data may similarly be performed repeatedly.

[0012] The wind turbine data represents an actual state of the wind turbine. The wind turbine data may be obtained sequentially, e.g. at a given sampling or refreshing rate. In this way, the health of the wind turbine component is continuously monitored based on the continuous determination of fault data based on the continuously monitored wind turbine data and the predetermined references which remain fixed.

[0013] Disclosed is an electronic device comprising a memory circuitry, a processor circuitry, and an interface, wherein the electronic device is configured to perform any of the methods according to the disclosed methods.

[0014] Disclosed is a wind turbine comprising the electronic device disclosed herein.

[0015] Disclosed is a computer readable storage medium storing one or more programs, the one or more programs comprising instructions, which when executed by an electronic device cause the electronic device to perform any of the methods according to the disclosed methods. Advantageously, the application of references such as spectral references on the wind turbine data improves the reliability and accuracy of the determined fault data.

[0016] Particularly, since the references are associated with a component, e.g. associated with known spectral elements of the component, the references may improve the capability of the recognition model to detect and / or predict faults.

[0017] It is an advantage of the present disclosure that the disclosed electronic device and method enable a more accurate detection and prediction of failures of or within one or more components of a wind turbine. This leads to a more robust detection of failures, in that the occurrence of false positive detections of failures, such as events or data incorrectly detected as failures, is reduced. In other words, it is an advantage of the disclosure that the references improve the accuracy and reliability of fault detection and / or prediction by the neural network model, as the misclassified faults are reduced or eliminated. In other words, the disclosure reduces “false positive” detection of a fault of a component.

[0018] This may result in an improved ability to pre-emptively replace or prepare for a replacement or repair of a soon-to-fail component. The present disclosure may also reduce the downtime or offline time of one or more wind turbines. Through early detection, repair costs can be reduced, which may result in significant savings in terms of time and resources. The disclosed technique may also result in a cost-reduction by allowing repair technologies and service technicians to be more effectively distributed over a number of wind turbines. For example, the disclosed may allow for a more accurate prioritization of repair needs and / or may allow for a more effective use of travel or repair time, allowing all failed or soon-to-fail components in a particular range to be repaired.

[0019] BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The above and other features and advantages of the present disclosure will become readily apparent to those skilled in the art by the following detailed description of exemplary embodiments thereof with reference to the attached drawings, in which:

[0021] Fig. 1 is a diagram illustrating schematically a wind turbine farm, according to the disclosure,

[0022] Fig. 2 is a flow-chart illustrating an exemplary method, performed by an electronic device, providing fault data associated with a component according to this disclosure, and Fig. 3 is a block diagram illustrating an exemplary electronic device according to this disclosure.

[0023] Fig. 4 is an example graphical representation of vibration patterns with references overlayed, according to the disclosure.

[0024] Fig. 5 is an example illustration of a comparison between a model’s prediction on a vibration pattern that resembles a failure vibration pattern with and without references, according to the disclosure.

[0025] DETAILED DESCRIPTION

[0026] Various exemplary embodiments and details are described hereinafter, with reference to the figures when relevant. It should be noted that the figures are only intended to facilitate the description of the embodiments. They are not intended as an exhaustive description of the disclosure or as a limitation on the scope of the disclosure. In addition, an illustrated embodiment needs not have all the aspects or advantages shown. An aspect or an advantage described in conjunction with a particular embodiment is not necessarily limited to that embodiment and can be practiced in any other embodiments even if not so illustrated, or if not so explicitly described.

[0027] The disclosure provides fault data associated with a component of a wind turbine by allowing pre-processing wind turbine data of the component to superimpose and / or overlay references associated with the component. The wind turbine data has a spectral element, or characteristic to allow for references to be superimposed / overlayed before applying a recognition model (such as a neural network model) to identify a “true” fault. For example, the wind turbine data is vibration data and / or acoustic data. A recognition model is applied to the wind turbine data with superimposed / overlayed references.

[0028] In one or more examples, the disclosed technique detects emerging fault patterns in the wind turbine data (such as CMS data) using recognition techniques (such as advanced image recognition techniques). In one or more examples, the disclosed technique comprises, inter alia, the following: first, plotting and saving one or more patterns to facilitate model training; In one or more examples, references (such as markers, such as order-tracked markers) are superimposed on the plots to enable a recognition model (such as convolutional neural networks (CNN)) to accurately identify specific patterns within wind turbine data (such as images). In one or more examples, the recognition models are trained using, for example, a model architecture (such as a CNN architecture, such as EfficientNet B7 architecture) that uniformly scales all dimensions of depth, width, resolution using a compound coefficient. In one or more examples, as part of the model training process, a fine-tuning of the last two layers of the CNN is prioritized to improve performance. In one or more examples, this approach enables the disclosed technique to accurately predict a fault, and / or a specific fault, such as a targeted fault. In one or more examples, it may be appreciated that the strategic incorporation and use of references in the training plots, as described herein, notably improves accuracy, resulting in precise and reliable results of detection and prediction of faults. In one or more examples, the disclosed technique also eliminates instances where similar patterns are misclassified as failures by a recognition model trained without the wind turbine data overlayed with references.

[0029] Fig. 1 is a diagram illustrating an example wind farm system 1 comprising a plurality of wind turbines 20, 26.

[0030] A wind turbine (such as wind turbine 20, 26) comprises one or more components, such as one or more mechanical components, one or more electrical components, one or more structural components. For example, the one or more components of the wind turbine may include one or more of: a gear box, bearings including journal and roller bearings (such as a main shaft bearing and / or gear box bearings), and a generator.

[0031] A wind turbine may comprise one or more sensors of a monitoring system, such as of a condition monitoring system (CMS), or a CMS may be shared between one or more wind turbines.

[0032] Fig. 1 shows a wind turbine 20, an electronic device 30. A controller may form part of wind turbine 20 or may be external to wind turbine 20. The controller may be seen as a controller configured to control operations of the wind turbine. The controller may be configured to perform the disclosed technique.

[0033] Fig. 1 shows two wind turbines, but it is to be understood that the present disclosure equally applies to a group of wind turbines arranged as a wind farm, and that the illustrated wind turbine may be a representative wind turbine of a wind farm. The electronic device 30 may be communicatively coupled with the wind turbine 20, and / or the electronic device 30 may communicate through one or more servers 40. The server 40 may be communicatively coupled with the wind turbine 20. In one or more examples, the server 40 may include a controller configured to perform the disclosed method.

[0034] Fig. 2 shows a flow diagram of an example method 100, performed by an electronic device, for providing fault data associated with a component of a wind turbine according to the disclosure. The method 100 may be seen as a method for early prediction of a fault and / or for determining a fault. The electronic device is the electronic device disclosed herein, such as electronic device 300 of Fig. 3

[0035] With reference to Fig. 2, the method 100 comprises obtaining S102 wind turbine data having a spectral element. In one or more examples, the wind turbine data is associated with the component. The wind turbine data may be seen as data obtained from or about the functionality and / or operation of one or more wind turbines, from internal and / or external sources, such as from one or more sensors, such as from a CMS system. The wind turbine data is for example in the spectral domain in that the wind turbine data comprises one or more data elements in the spectral domain, such as one or more spectral elements. A spectral element may be seen as an element of any data with a frequency or wavelength. Herein, wind turbine data having a spectral element is discussed and may be seen as wind turbine data in a spectral domain, for example, the spectral element (of the wind turbine data) is in the spectral domain. In other words, for example, wind turbine data exhibiting a behaviour observable in a spectrum of frequencies, thereby having spectral properties. For example, the electronic device may achieve obtaining wind turbine data having a spectral element by sampling data from CMS and / or sensor(s) in a time domain and converting the time domain data into the spectral domain. In one or more example methods, the spectral element comprises a frequency element indicative of a frequency of vibration and / or sound in a spectrum of frequencies. A frequency element may be seen as an element of the spectral data that indicates a frequency, e.g., a frequency of the pattern, e.g., a number of cycles or vibrations during a particular unit of time. In one or more examples, and as described herein, the spectral element may comprise a frequency of vibration and / or of sound in a spectrum of frequencies. In one or more example methods, the wind turbine data comprises vibration data associated with the component and / or acoustic data associated with the component. Vibration data may be seen as data from any structural vibration. For example, data from vibrations of a component of the drive train, the rotor, the blades , generator, or another structural element that has a spectral element. Acoustic data may be seen as data collected from and associated with sound waves, for example, data collected through acoustic monitoring techniques, such as microphone(s). In one or more examples, the acoustic data has a spectral element.

[0036] A component may be seen as a component of a wind turbine and / or wind turbine system. For example, a component may be a mechanical and / structural and / or electrical component. For example, a component may be one or more of a gear box, bearings (including journal and roller bearings, for example, the main shaft bearings or gear box bearings of the wind turbine) and a generator. For example, wind turbine data associated with a component can include vibration data associated with the gear box. In one or more examples, the wind turbine data may be data detected and / or measured and / or collected by one or more sensors, for example, one or more sensors in the wind turbine itself. In one or more examples, the wind turbine data may be detected and / or collected by a sensor and / or a sensor set external to the wind turbine.

[0037] The method 100 comprises obtaining S104 references associated with the component.

[0038] References may be seen as a set of reference points associated with a given component, for example a reference point where a peak in frequency is expected for the given component. For example, references can be generated based on kinematic markers. In one or more examples, they may be calculated based on and / or associated with the type of component (e.g., gearbox). The calculation of the frequency of the fault for each reference or set of references may differ based on arrangement, design, carrier design or other factors. Thus, the references such as spectral references may comprise spectral indicators or frequencies such as numerical values representing frequencies. The references may be seen as an indicator indicating frequencies or spectral ranges of the wind turbine data where a possible fault, a wear effect(s) and / or unintended operation of a component may result in changes in the spectral content of the wind turbine data.

[0039] The references may comprise a plurality of different spectral values, such as spectral ranges, such as a plurality of different, e.g. non-overlapping, spectral ranges. The different references may be a function of only frequency so that the overlaying provides that the references are distributed as a function of the spectral content of the wind turbine data, and / or the different references may be a function of frequency and optionally another parameter, such as an amplitude of the wind turbine data so that the references are distributed in two dimensions of the wind turbine data. Further, the different references may be a function of frequency and a plurality of other parameters, e.g. so that the references are distributed dependent on 3 variables of the wind turbine data.

[0040] In one or more examples, the markers may be graphical references associated with the mechanical component. In an embodiment, the references (e.g., graphical references) comprise markers. For example, see markers in Fig. 4 (e.g., 402 of Fig. 4).

[0041] In one or more examples, the references may be obtained one or more of a variety of ways, for example, by receiving (e.g., from a transducer output in the CMS, from a manual or virtual operator of the CMS), retrieving (e.g., from a memory or data repository internal or external to the wind turbine, for example, from a location in the cloud), generating (e.g., based on one or more inputs related to the wind turbine operation or function), and / or determining (e.g., based on a pre-programmed or accessed database or reference). In one or more examples, the reference (e.g., graphical reference) comprises a predetermined graphical reference. In one or more examples, the reference is determined and / or retrieved from a database including frequencies of fault(s) of each equipment (e.g., component). In one or more examples, the reference is based on predetermined component-specific data, for example, kinematic data of a component such as a gearbox.

[0042] The method 100 comprises overlaying S106 the references onto the wind turbine data. In one or more examples, the references may be compared and / or overlayed, and / or superimposed, over the wind turbine data (e.g., the spectral data), such as visually superimposed on a graphical plot, for example, as depicted in Figs. 4 and 5.

[0043] In one or more example methods, the references comprise one or more of: markers, and graphical references. For example, the graphical references can be seen as graphical representations of the references, e.g. in a graph, such as predetermined graphical references, and / or graphical reference determined from predetermined component specific data.

[0044] The method 100 comprises determining S108 the fault data by applying a recognition model to the wind turbine data with overlayed references. Fault data may be seen as data indicative of a probability of a fault, for example, a failure of one or more components or processes of the wind turbine(s). In one or more examples, fault data may be determined based on the wind turbine data by applying a mathematical model to the wind turbine data. In one or more examples and as described herein, the use of the overlayed references supports training the recognition model by highlighting references (such as labels, such as using labelled data where labelled data has overlayed markers / references) to the recognition model (e.g., a CNN) for an accurate fault detection (such as a location in spectral domain). In one or more example methods, the fault data is indicative of a probability of presence of a fault or indicative of a probability of absence of a fault. In one or more examples, the fault data may indicate a probability of a presence of a fault. For example, the fault data may indicate a high likelihood of a failure in one or more components. In one or more examples, the fault data may indicate a low likelihood of a failure in one or more of the components. In one or more example methods, the fault data comprise a fault parameter indicating the presence or absence of a fault, and a fault score indicating the probability of presence of the fault or the probability of absence of a fault. A fault parameter may be seen as an indicator of a presence or absence of a fault as output by the recognition model. A fault score may be seen as a numerical score indicating the probability of presence or absence of a fault, such as the confidence in the fault parameter. In one or more examples, a fault score may be seen as numerical indication that the determination based on the fault parameter is accurate. For example, a fault score above a threshold could indicate that the fault parameter (for example indicating presence of a fault) is highly likely, such as provided with high confidence. For example, a fault score above a threshold could indicate the high confidence in the fault parameter indicating the absence of a fault. In one or more examples, the fault score can be used in the inverse of what is previously described, or in another way.

[0045] A recognition model may be seen as a mathematical model configured to determine one or more fault patterns in the wind turbine data, for example, an image recognition model. In one or more examples, a recognition model may be a model used for various components and / or for wind turbine types. In one or more examples, a recognition model may be a model customized specifically to fault detection based on the particular attributes of the wind turbine or the component. The recognition model may be comprised in the memory of the electronic device, which may be internal and / or external to the wind turbine.

[0046] In one or more example methods, the recognition model is configured to determine a fault pattern associated with the component based on the wind turbine data with overlayed references. In one or more examples, the recognition model may be used to more accurately predict a fault pattern associated with the relevant component as a “true fault”. In one or more examples, one or more patterns may be identifiable by the recognition model as a fault pattern, thus indicating a presence of a fault (e.g., a detection that a component may fail). In one or more examples, component-specific, reference-overlayed wind turbine data may be used to determine which, if any, of the fault patterns are “true” faults. One more examples herein, such as examples in Fig. 5, may further illustrate the use of the overlayed references with respect to the wind turbine data.

[0047] In one or more example methods, the recognition model is configured to classify the wind turbine data as indicative of a fault or of an absence of a fault. In one or more examples, the recognition model is configured to indicate when there is or is not a fault of the component based on the wind turbine data overlayed with references. For example, based on the recognition model processing using the overlayed references, the output may indicate the detection of a fault in the wind turbine data. In one or more examples, based on the recognition model processing, the output may indicate an absence of a fault, for example, a determination that the particular component is not likely to fail, based on the provided data and parameters.

[0048] In one or more example methods, the recognition model is a convolutional neural network model. In one or more examples, the recognition model comprises one or more of: a machine-learning model, a convolutional neural network model, and Support Vector Machine (SVM) model. A convolution neural network (CNN) may be seen as a feedforward neural network. In one or more examples, the CNN is internal and / or external to the wind turbine. For example, the recognition model processing may occur as a part of the CMS, including one or more processors present as part of the wind turbine system technology. In one or more examples, the recognition model may be stored, accessed, and used remotely, e.g., external to the wind turbine. In one or more examples, the training, use, and / or storage of the recognition model(s) may occur within the wind turbine processors, as part of the CMS equipment, and / or remote to the wind turbine, for example, via one or more distributed processors and / or in the cloud.

[0049] An SVM model may be seen as a supervised machine learning algorithm that classifies data by finding an optimal line or hyperplane that maximizes the distance between each class in an N-dimensional space. In one or more example methods, obtaining S102 the wind turbine data comprises obtaining (e.g. receiving and / or retrieving) S102A the wind turbine data from one or more sensors of a monitoring system of the wind turbine. A monitoring system may be seen as a condition monitoring system (CMS), as discussed herein. A monitoring system, like a CMS, may be used to detect faults before the faults cause secondary damage (for example, the failure of one part then causing damage to other components in the wind turbine). The CMS may be located close to the power train in the wind turbine. In one or more examples, the CMS collects vibration data (e.g., raw data and / or processed data) via transducers. In other words, in one or more examples, a CMS is designed to convert a physical measurement (for example, vibration, temperature, pressure and / or other measurements) into a digital form.

[0050] In one or more example methods, obtaining S104 the references associated with the component comprises determining S104A, based on the wind turbine data, the references (and optionally ordering the references). In one or more examples, the wind turbine data may be accessed from or provided by, for example, one or more of supplier data, historical wind turbine data, live data (e.g., data received or accessed at run-time), data from one or more databases for the same component and / or data from one or more databases for a different wind turbine type, model, or component. In one or more examples, the historical wind turbine data may be collected from the same or a different wind turbine, in the same or a different wind turbine farm. The historical wind turbine data may be stored and accessed according to profiles or in a different way, as is suitable to the particular wind turbine setup, and / or a particular wind turbine design. In one or more examples, the determining of the references may occur before runtime (e.g., in an initial phase) and may be stored, for example, in a database, in a device memory, in the cloud, or in another suitable manner. The references may then be accessed at runtime or as needed. In one or more examples, the references may be determined for a particular wind turbine and then used for an associated or in other ways relevant wind turbine, as described herein.

[0051] In one or more example methods, obtaining S104 the references associated with the component comprises obtaining (e.g. receiving and / or retrieving) S104B references associated with a type of component to which the component pertains. In one or more examples, the references may be associated with a particular component or component type (for example, a gearbox, or more specifically, a gearbox by a particular manufacturer, or a gearbox by a particular manufacturer of a certain lot number). In one or more examples, the references may be interchangeable or specific to the particular component, component type, or other identifier. In one or more examples, the references may be obtained for a ‘best fit’ based on available component data and / or wind turbine data. For example, references may exist for a particular wind turbine component by a particular manufacturer. However, for any reason, the reference may not be available nor determinable for the specific wind turbine component by a particular manufacturer. In one or more examples, a reference number that has been determined for the particular manufacturer may be obtained and used for the wind turbine data component of the different manufacturer. In other words, in one or more examples, references may be associated with the same or different components, based on a variety of factors, and the references may be used to accurately provide fault data, as described herein.

[0052] In one or more example methods, obtaining S104 the references associated with the component comprises receiving S104C the references associated with the component from one or more databases. In one or more examples, the references associated with the component may be accessed from one or more databases. In one or more examples, the one or more databases may be local or distributed, for example, in the cloud or over a series of remote databases. In one or more examples, the database may be populated according to manufacturer-supplied data. In one or more examples, the database may be a component-manufacturer database.

[0053] In one or more example methods, the references associated with the component obtained are ordered based on a speed property of the component. In one or more examples, ordering of the references associated with the component may be based on the generator speed. For example, spectrum data is converted into first order or second order so that the frequency is normalized. In one or more examples, this method is for machines with varying speeds to compensate for (e.g., normalize) the speed variations. Thus, in one or more examples, a variety of speeds may be used without negatively impacting the references and therefore the fault detection (e.g., the results).

[0054] In one or more example methods, the recognition model is trained based on historical wind turbine data. Historical wind turbine data may be seen as wind turbine data from past operations of one or more wind turbines. The wind turbine for which the historical wind turbine data is being provided may be the wind turbine for which fault data is being provided. The historical wind turbine data may also be for one or more other wind turbines. In one or more examples, the one or more other wind turbines may be of the same type or model of wind turbine, or they may be for another type and / or model of wind turbine.

[0055] In one or more example methods, the method 100 comprises training S107 the recognition model by fine tuning the model using the wind turbine data. In one or more examples, the recognition model may be trained according to standard processes for the training of CNNs, such as feed-forward neural networks. In one or more examples, the training may comprise a prioritization of fine-tuning the last two layers of the CNN for an improved performance. In other words, a standardized CNN may be used for training, with the final two layers of the model being customized for the particular purpose of fault detection. In one or more examples, one or more parameters may be tuned (e.g., by fine tuning) to find a model well-suited to the CNN requirements.

[0056] In one or more example methods, the method 100 comprises controlling S110 an operation of the wind turbine based on the fault data. In one or more examples, based on the fault data determination, one or more wind turbines may be mechanically or electronically controlled. For example, in response to a determination of a “true fault”, as described herein, the wind turbine may be paused in anticipation of an immediate repair, to prevent further damage, and / or to allow for manual (e.g., operator) intervention, or for another reason. For example, other ways of operating the wind turbine in response to determination of fault data comprises operating the wind turbine in a derated mode so that loads, e.g. drive train loads, are limited.

[0057] Fig. 3 shows a block diagram of an exemplary electronic device 300 according to the disclosure. The electronic device 300 comprises memory circuitry 301, processor circuitry 302, and an interface 303. The electronic device 300 is configured to perform any of the methods disclosed in Fig. 2. In other words, the electronic device 300 is configured for enabling control of operation of a wind turbine, such as providing fault data associated with a component of a wind turbine. For example, the electronic device 300 may be a controller of a wind turbine, such as an internal controller of the wind turbine. For example, the electronic device may be a device remote from the wind turbine (such device 30 or server 40 of Fig. 1). For example, the electronic device may be an external controller of the wind turbine. The electronic device 300 is configured to obtain (e.g., via processor circuitry 302 and / or interface 303) wind turbine data having a spectral element . The wind turbine data is associated with the component .

[0058] The electronic device 300 is configured to obtain (e.g., via processor circuitry 302 and / or interface 303) references associated with the component.

[0059] The electronic device 300 is configured to overlay (e.g., via processor circuitry 302) the references onto the wind turbine data.

[0060] The electronic device 300 is configured to determine (e.g., via processor circuitry 302) the fault data by applying a recognition model to the wind turbine data with overlayed references.

[0061] The processor circuitry 302 is optionally configured to perform any of the operations disclosed in Fig. 2 (such as any one or more of: S102, S102A, S104, S104A, S104B, S104C, S106, S107, S108, S110). The operations of the electronic device 300 may be embodied in the form of executable logic routines (e.g., lines of code, software programs, etc.) that are stored on a non-transitory computer readable medium (e.g., the memory circuitry 301) and are executed by the processor circuitry 302.

[0062] Furthermore, the operations of the electronic device 300 may be considered a method that the electronic device 300 is configured to carry out. Also, while the described functions and operations may be implemented in software, such functionality may as well be carried out via dedicated hardware or firmware, or some combination of hardware, firmware and / or software.

[0063] The memory circuitry 301 may be one or more of a buffer, a flash memory, a hard drive, a removable media, a volatile memory, a non-volatile memory, a random-access memory (RAM), or other suitable device. In a typical arrangement, the memory circuitry 301 may include a non-volatile memory for long term data storage and a volatile memory that functions as system memory for the processor circuitry 302. The memory circuitry 301 may exchange data with the processor circuitry 302 over a data bus. Control lines and an address bus between the memory circuitry 301 and the processor circuitry 302 also may be present (not shown in Fig. 3). The memory circuitry 301 is considered a non-transitory computer readable medium. The memory circuitry 301 may be configured to store wind turbine data, references, trained recognition model parameters in a part of the memory.

[0064] Fig. 4 shows an example graphical representation of vibration patterns with references overlayed, according to the disclosure. In one or more examples, graphical representation 400 illustrates an example plot of spectral data 404 (e.g., spectral data as described herein) with references overlayed as described herein. An example reference is indicated at 402 (such as markers (optionally order tracked) plotted on the CMS data). The references, as discussed herein, may be determined (e.g., calculated) based on kinetic markers and the specific type of component for which the reference is generated. In one or more examples, references may be obtained or derived from a set of manufacture- provided data (e.g., from a reference manual or database).

[0065] Fig. 5 shows an example illustration of a comparison between a model’s prediction on a vibration pattern that resembles a failure vibration pattern with and without references, according to the disclosure. As further depicted and described herein, the use of references (e.g., “markers”) on plots of spectral data (for example, spectral data obtained from wind turbine data, as described herein ) when training and / or running the recognition model allows the recognition model to successfully categorize a normal pattern that resembles a fault pattern as “no fault”. For example, a recognition model trained or run on plots without references (e.g., without markers) mistakenly categorizes the data (e.g., fault data) as a fault. In other words, with the references, the recognition model can detect when the seeming fault data is in fact, a “false” fault, thereby reducing false positives. For example, with the references, the recognition model can detect more accurately a “true” fault.

[0066] Graphical representations 52A, 52B, and 52C (collectively “52”) illustrate the recognition model’s prediction on a vibration pattern that resembles the failure vibration pattern (e.g., fault data). In graphical representations 52A, 52B, 52C, the model accurately categorizes such patterns as “no-failure” with a high probability thanks to references such as reference 53A. This is a correct identification, despite the resemblance of the pattern to a fault pattern.

[0067] Graphical representations 54A, 54B, and 54C (collectively “54”) illustrate the model’s (e.g., a recognition model, e.g., a CNN, as described herein), prediction on a vibration pattern that resembles the failure vibration pattern (e.g., fault data). In this case, the model incorrectly labels such patterns as “failure” with a high probability. As described herein, this incorrect identification of faults (e.g., misidentifying “false faults” as “true faults”) may result in significant loss of time and resources. In one or more examples, the use of markers significantly improves the accuracy (e.g., reduces the misidentifications) of fault detection in the CMS for wind turbines.

[0068] It should further be noted that any reference signs do not limit the scope of the claims, that the exemplary embodiments may be implemented at least in part by means of both hardware and software, and that several "means", "units" or "devices" may be represented by the same item of hardware.

[0069] The various exemplary methods, devices, nodes, and systems described herein are described in the general context of method steps or processes, which may be implemented in one aspect by a computer program product, embodied in a computer- readable medium, including computer-executable instructions, such as program code, executed by computers in networked environments. A computer-readable medium may include removable and non-removable storage devices including, but not limited to, Read Only Memory (ROM), Random Access Memory (RAM), compact discs (CDs), digital versatile discs (DVD), etc. Generally, program circuitries may include routines, programs, objects, components, data structures, etc. that perform specified tasks or implement specific abstract data types. Computer-executable instructions, associated data structures, and program circuitries represent examples of program code for executing steps of the methods disclosed herein. The particular sequence of such executable instructions or associated data structures represents examples of corresponding acts for implementing the functions described in such steps or processes.

[0070] It is to be noted that the term "indicative of" may be seen as “associated with”, “related to”, “descriptive of’, “characterizing”, and / or “defining”. The terms “indicative of’, “associated with” “related to”, “descriptive of’, “characterizing”, and “defining” can be used interchangeably. The term “indicative of” can be seen as indicating a relation. For example, weight data indicative of weight may comprise one or more weight parameters.

[0071] It is to be noted that the word "based on" may be seen as “as a function of’ and / or “derived from”. The terms “based on” and “as a function of” can be used interchangeably. For example, a parameter determined “based on” a data set can be seen as a parameter determined “as a function of” the data set. In other words, the parameter may be an output of one or more functions with the data set as an input.

[0072] A function may be characterizing a relation between an input and an output, such as mathematical relation, a database relation, a hardware relation, logical relation, and / or other suitable relations.

[0073] Although features have been shown and described, it will be understood that they are not intended to limit the claimed disclosure, and it will be made obvious to those skilled in the art that various changes and modifications may be made without departing from the scope of the claimed disclosure. The specification and drawings are, accordingly, to be regarded in an illustrative rather than restrictive sense. The claimed disclosure is intended to cover all alternatives, modifications, and equivalents.

Claims

CLAIMS1 . A method, performed by an electronic device, for providing fault data associated with a component of a wind turbine, the method comprising: obtaining wind turbine data having a spectral element, wherein the wind turbine data is associated with the component; obtaining predetermined references associated with the component; overlaying the references onto the wind turbine data; and determining the fault data by applying a recognition model to the wind turbine data with overlayed references.

2. The method according to claim 1 , wherein the wind turbine data comprises vibration data associated with the component and / or acoustic data associated with the component.

3. The method according to any of the previous claims, wherein obtaining the wind turbine data comprises obtaining the wind turbine data from one or more sensors of a monitoring system of the wind turbine.

4. The method according to any of the previous claims, wherein the spectral element comprises a frequency element indicative of a frequency of vibration and / or sound in a spectrum of frequencies.

5. The method according to any of the previous claims, wherein obtaining the references associated with the component comprises determining, based on the wind turbine data, the references.

6. The method according to any of the previous claims, wherein obtaining the references associated with the component comprises obtaining references associated with a type of component to which the component pertains.

7. The method according to any of the previous claims, wherein obtaining the references associated with the component comprises receiving the references associated with the component from one or more databases.

8. The method according to any of the previous claims, wherein the references associated with the component obtained are ordered -based on a speed property of the component.

9. The method according to any of the previous claims, wherein the fault data is indicative of a probability of presence of a fault or indicative of a probability of absence of a fault.

10. The method according to claim 9, wherein the fault data comprise a fault parameter indicating the presence or absence of a fault, and a fault score indicating the probability of presence of the fault or the probability of absence of a fault.

11. The method according to any of the previous claims, wherein the recognition model is configured to- determine a fault pattern associated with the component based on the wind turbine data with overlayed references; and- classify the wind turbine data as indicative of a fault or of an absence of a fault.

12. The method according to any of the previous claims, wherein the recognition model is a convolutional neural network model.

13. The method according to any of the previous claims, wherein the recognition model is trained based on historical wind turbine data.

14. The method according to any of the previous claims, wherein the references comprise one or more of: markers, and graphical references.

15. An electronic device comprising a memory circuitry, a processor circuitry, and an interface, wherein the electronic device is configured to perform any of the methods according to any of claims 1-14.

16. A wind turbine comprising the electronic device according to claim 15.

17. A computer readable storage medium storing one or more programs, the one or more programs comprising instructions, which when executed by an electronic device cause the electronic device to perform any of the methods of claims 1-14.

Images