Blade fault diagnosis method, device, system and storage medium

Abstract

The application discloses a kind of blade fault diagnosis method, device, system and storage medium, the method comprises: obtaining the blade rotation audio that audio acquisition equipment is collected in the operation process of wind turbine generator unit;Based on wind noise filtering algorithm, blade rotation audio is preprocessed, and the blade rotation audio after filtering wind noise is obtained;The blade rotation audio after filtering wind noise is segmented, and the audio segment corresponding to each blade is obtained;According to each audio segment, diagnose whether the blade corresponding to each audio segment exists fault.The method provided in the embodiment of the application can diagnose whether the corresponding blade occurs fault according to the audio segment of different blades respectively, and the accuracy of diagnostic result is improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of wind power generation, and particularly relates to a blade fault diagnosis method, device, system and storage medium. BACKGROUND

[0002] The blades of a wind turbine will emit sound when rotating. When the blades are damaged or have defects, the audio of the rotating blades will change to a certain extent. Therefore, when monitoring the state of the blades of the wind turbine, one implementation is to collect the audio of the rotating blades, and then determine whether the blades of the wind turbine have faults according to the audio signal.

[0003] However, the inventors have found that since the change in the audio signal is small when the blades have faults, and the audio signals of different blades have slight differences, the existing method of diagnosing whether the blades have faults through audio may cause misdiagnosis. SUMMARY

[0004] The embodiments of the present application provide a blade fault diagnosis method, device, system and storage medium, which can diagnose whether the corresponding blades have faults according to the audio segments of different blades respectively, and improve the accuracy of the diagnosis result.

[0005] In one aspect, the embodiments of the present application provide a blade fault diagnosis method, which comprises: obtaining blade rotating audio collected by an audio collection device in the operation process of a wind turbine; pre-processing the blade rotating audio based on a wind noise filtering algorithm to obtain blade rotating audio filtered of wind noise; segmenting the blade rotating audio filtered of wind noise to obtain audio segments corresponding to each blade; and diagnosing whether the blade corresponding to each audio segment has a fault according to each audio segment.

[0006] Optionally, segmenting the blade rotating audio filtered of wind noise to obtain audio segments corresponding to each blade comprises: processing the blade rotating audio filtered of wind noise by short-time Fourier transform to obtain first characteristic values, the first characteristic values being used to represent the frequency domain characteristics of the blade rotating audio filtered of wind noise; inputting the first characteristic values into a blade recognition model to obtain segmentation time points of the blade rotating audio filtered of wind noise, wherein the blade recognition model is a pre-trained model used to identify the switching time points of the rotating sound of different blades according to the first characteristic values; and segmenting the blade rotating audio filtered of wind noise according to the segmentation time points to obtain audio segments corresponding to each blade.

[0007] Optionally, diagnosing whether the corresponding blade has a fault according to each audio segment comprises: processing each audio segment respectively through Fourier transform to obtain second characteristic values of each audio segment, the second characteristic values being used to represent frequency domain characteristics of each audio segment; inputting the second characteristic values of each audio segment into a blade fault diagnosis model respectively to obtain first fault diagnosis results of each blade, the blade fault diagnosis model being a model pre-trained for identifying whether the corresponding blade has a fault according to the second characteristic values of the audio segment.

[0008] Optionally, diagnosing whether the corresponding blade has a fault according to each audio segment comprises: counting time lengths of the audio segments corresponding to each blade; judging whether there is a fault blade according to whether a difference between the time lengths of each two audio segments exceeds a preset threshold to obtain second fault diagnosis results; and combining the first fault diagnosis results and the second fault diagnosis results of each blade to judge whether there is a fault blade.

[0009] Optionally, before diagnosing whether the corresponding blade has a fault according to each audio segment, the method further comprises: obtaining an environmental parameter of the wind turbine, wherein the environmental parameter is used to represent a season and / or weather; and determining a model for identifying a fault type corresponding to the environmental parameter from a plurality of candidate fault diagnosis models respectively used to identify different fault types to determine the fault diagnosis model to be used.

[0010] Optionally, obtaining the blade rotating audio collected by the audio acquisition device in the operation process of the wind turbine comprises: obtaining blade rotating audios collected at a plurality of positions; and pre-processing the blade rotating audios based on a wind noise filtering algorithm to obtain blade rotating audios filtered of wind noise, comprising: processing each blade rotating audio respectively through the wind noise filtering algorithm to obtain a plurality of blade rotating audios filtered of wind noise; after obtaining the plurality of blade rotating audios filtered of wind noise, the method further comprises: calculating wind noise parameters of each blade rotating audio filtered of wind noise respectively through a wind noise identification model, wherein the wind noise parameters are used to represent wind noise sizes in the audios, and the wind noise identification model is a model pre-trained for evaluating wind noise parameters of audios; and segmenting the blade rotating audios filtered of wind noise to obtain audio segments corresponding to each blade, comprising: selecting a blade rotating audio filtered of wind noise with the smallest wind noise from the plurality of blade rotating audios filtered of wind noise according to the wind noise parameters to segment the blade rotating audio filtered of wind noise to obtain the audio segments corresponding to each blade.

[0011] Optionally, the wind noise parameters of each blade rotation audio after filtering wind noise are calculated by the wind noise identification model, including: each blade rotation audio after filtering wind noise is processed by Fourier transform respectively to obtain third characteristic values of each blade rotation audio after filtering wind noise, the third characteristic values being used to represent the frequency domain characteristics of the corresponding blade rotation audio after filtering wind noise; and the third characteristic values of each blade rotation audio after filtering wind noise are input into the wind noise identification model respectively to obtain the wind noise parameters of each blade rotation audio after filtering wind noise.

[0012] In another aspect, the embodiments of the present application provide a blade fault diagnosis device, which comprises: an acquisition module configured to acquire blade rotation audio collected by an audio acquisition device during operation of a wind turbine generator system; a preprocessing module configured to preprocess the blade rotation audio based on a wind noise filtering algorithm to obtain blade rotation audio after filtering wind noise; a segmentation module configured to segment the blade rotation audio after filtering wind noise to obtain audio segments corresponding to each blade; and a diagnosis module configured to diagnose whether each audio segment corresponding blade has a fault according to each audio segment.

[0013] Optionally, the blade rotation audio after filtering wind noise is segmented to obtain audio segments corresponding to each blade, including: the blade rotation audio after filtering wind noise is processed by short-time Fourier transform to obtain first characteristic values, the first characteristic values being used to represent the frequency domain characteristics of the blade rotation audio after filtering wind noise; the first characteristic values are input into a blade identification model to obtain segmentation time points of the blade rotation audio after filtering wind noise, wherein the blade identification model is a pre-trained model configured to identify switching time points of rotation sounds of different blades according to the first characteristic values; and the blade rotation audio after filtering wind noise is segmented according to the segmentation time points to obtain audio segments corresponding to each blade.

[0014] Optionally, whether each audio segment corresponding blade has a fault is diagnosed according to each audio segment, including: each audio segment is processed by Fourier transform respectively to obtain second characteristic values of each audio segment, the second characteristic values being used to represent the frequency domain characteristics of each audio segment; and the second characteristic values of each audio segment are input into a blade fault diagnosis model respectively to obtain first fault diagnosis results of each blade, the blade fault diagnosis model being a pre-trained model configured to identify whether each audio segment corresponding blade has a fault according to the second characteristic values of the audio segment.

[0015] Optionally, whether each audio segment corresponding blade has a fault is diagnosed according to each audio segment, including: the time lengths of the audio segments corresponding to each blade are counted; whether there is a fault blade is determined according to whether the difference between the time lengths of each two audio segments exceeds a preset threshold to obtain second fault diagnosis results; and whether there is a fault blade is determined by combining the first fault diagnosis results and the second fault diagnosis results of each blade.

[0016] Optionally, before diagnosing whether the corresponding blade has a fault according to each audio segment, the method further comprises: obtaining an environmental parameter of the wind turbine, wherein the environmental parameter is used to represent a season and / or weather; determining a model used to identify a fault type corresponding to the environmental parameter from a plurality of candidate fault diagnosis models respectively used to identify different fault types, to determine the used fault diagnosis model.

[0017] Optionally, obtaining the blade rotating audio collected by the audio acquisition device during the operation of the wind turbine comprises: obtaining blade rotating audio collected at a plurality of positions; pre-processing the blade rotating audio based on a wind noise filtering algorithm to obtain blade rotating audio filtered of wind noise, comprising: performing wind noise filtering algorithm processing on each blade rotating audio to obtain a plurality of blade rotating audios filtered of wind noise; after obtaining the plurality of blade rotating audios filtered of wind noise, the method further comprises: calculating a wind noise parameter of each blade rotating audio filtered of wind noise by a wind noise identification model, wherein the wind noise parameter is used to represent the size of wind noise in the audio, and the wind noise identification model is a pre-trained model used to evaluate the wind noise parameter of the audio; segmenting the blade rotating audio filtered of wind noise to obtain an audio segment corresponding to each blade, comprising: selecting the blade rotating audio filtered of wind noise with the smallest wind noise from the plurality of blade rotating audios filtered of wind noise based on the wind noise parameter to segment to obtain an audio segment corresponding to each blade.

[0018] Optionally, calculating the wind noise parameter of each blade rotating audio filtered of wind noise by the wind noise identification model comprises: processing each blade rotating audio filtered of wind noise by a Fourier transform to obtain a third feature value of each blade rotating audio filtered of wind noise, the third feature value being used to represent the frequency domain feature of the corresponding blade rotating audio filtered of wind noise; inputting the third feature value of each blade rotating audio filtered of wind noise into the wind noise identification model to obtain the wind noise parameter of each blade rotating audio filtered of wind noise.

[0019] In still another aspect, the embodiments of the present application provide a blade fault diagnosis system, comprising: an audio acquisition device comprising an audio sensor, wherein at least one audio sensor is arranged at a main wind direction position or a leeward position of a tower of a wind turbine; a processor arranged inside the tower and connected to the audio sensor, used to receive blade rotating audio collected by the audio sensor, pre-process the blade rotating audio based on a wind noise filtering algorithm to obtain blade rotating audio filtered of wind noise; segment the blade rotating audio filtered of wind noise to obtain an audio segment corresponding to each blade; and diagnose whether the corresponding blade has a fault according to each audio segment.

[0020] Optionally, the system further comprises a server connected with the processor, configured to acquire a result of diagnosing whether the blade has a fault by the processor, and give a prompt in the case that the blade has a fault.

[0021] In another aspect, an embodiment of the present application provides a storage medium, in which computer program instructions are executed by a processor to implement the blade fault diagnosis method in the embodiment of the present application.

[0022] The blade fault diagnosis method, device, system and storage medium provided by the embodiment of the present application can filter out wind noise from the blade rotating audio through wind noise filtering algorithm processing, so as to remove the interference of wind noise in the blade rotating audio and obtain audio carrying more accurate and stronger blade rotating sound. Then, the blade rotating audio after filtering out wind noise is segmented to obtain audio segments corresponding to each blade, and whether each audio segment corresponds to a blade having a fault is diagnosed according to each audio segment, so that whether each blade has a fault can be diagnosed according to the audio segments of different blades, and the accuracy of the diagnosis result is improved. BRIEF DESCRIPTION OF DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiments of the present application will be briefly introduced. For those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0024] Figure 1 FIG. 1 is a schematic diagram of a blade fault diagnosis system provided by an embodiment of the present application;

[0025] Figure 2 FIG. 2 is a schematic diagram of a blade fault diagnosis system provided by another embodiment of the present application;

[0026] Figure 3 FIG. 3 is a flowchart of a blade fault diagnosis method provided by an embodiment of the present application;

[0027] Figure 4 FIG. 4 is a flowchart of training a model of a blade fault diagnosis method provided by an embodiment of the present application;

[0028] Figure 5 FIG. 5 is a flowchart of identifying wind noise by a wind noise identification model in a blade fault diagnosis method provided by an embodiment of the present application;

[0029] Figure 6 FIG. 6 is a flowchart of obtaining an audio segment by a blade identification model in a blade fault diagnosis method provided by an embodiment of the present application;

[0030] Figure 7is a flowchart of diagnosing a blade fault by a blade fault diagnosis model in a blade fault diagnosis method provided by an embodiment of the present application;

[0031] Figure 8 is a flowchart of a blade fault diagnosis method provided by another embodiment of the present application;

[0032] Figure 9 is a schematic diagram of a blade fault diagnosis system provided by another embodiment of the present application;

[0033] Figure 10 is a schematic diagram of a blade fault diagnosis system provided by another embodiment of the present application;

[0034] Figure 11 is a schematic diagram of a blade fault diagnosis device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0035] The features and exemplary embodiments of various aspects of the present application will be described below in detail, in order to make the purposes, technical solutions and advantages of the present application more clear and apparent, the present application will be further described in detail below in combination with the drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present application, but not to limit the present application. The present application can be implemented without some of these specific details by those skilled in the art. The following description of the embodiments is only to provide a better understanding of the present application by showing examples of the present application.

[0036] It should be noted that, in this paper, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device. Without more limitations, the elements defined by the statement "include" do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0037] In order to solve the problems in the prior art, the embodiments of the present application provide a blade fault diagnosis method, device, system and storage medium.

[0038] First, the blade fault diagnosis system provided by the embodiments of the present application will be introduced.

[0039] As Figure 1As shown, it is an optional structural schematic diagram of a blade fault diagnosis system provided by the embodiment of the present application, which comprises an audio acquisition device 110 and a processor 120, wherein the audio acquisition device 110 comprises at least one audio sensor 111.

[0040] The at least one audio sensor 111 is arranged at a position of a main wind direction or a leeward direction of a tower of the wind turbine generator set. The audio sensor 111 is used to acquire a blade rotation audio signal of the wind turbine generator set. The blade of the wind turbine generator set will emit sound when rotating, and the sound of the blade rotation will change when the blade is faulty. Therefore, the audio signal acquired by the audio sensor 111 can be used to diagnose the blade fault. Optionally, the audio sensor 111 can be a pickup. Optionally, in addition to the audio sensor 111, the audio acquisition device 110 can comprise a signal processing module, wherein the signal processing module of the audio acquisition device 110 can be used to convert the blade rotation audio signal acquired by the audio sensor 111 into blade rotation audio, and optionally, the blade rotation audio can be further preprocessed, etc. The audio sensor 111 of the audio acquisition device 110 can be arranged on the outer wall of the tower of the wind turbine generator set, and other hardware parts of the audio acquisition device 110 can be arranged inside the tower of the wind turbine generator set. The hardware inside the tower can communicate with the audio sensor 111 through wired communication or wireless communication, for example, the hardware of the audio acquisition device 110 arranged inside the tower can be connected with the audio sensor 111 through an optical fiber communication cable to receive the audio signal acquired by the audio sensor 111.

[0041] As shown in FIG. 1, Figure 2 It is an optional installation schematic diagram of an audio sensor, Figure 1 The audio sensor 111 in FIG. 1 can comprise an audio sensor 141 and an audio sensor 142 of Figure 2 FIG. 1. Wherein the audio sensor 141 is arranged at a position of a main wind direction 151 of a tower 130 of the wind turbine generator set, Figure 2 which is shown as a cross section of the tower 130 of the wind turbine generator set, and the audio sensor 142 is arranged above a tower door 131 of the wind turbine generator set, i.e. in a direction with an angle of 90° with the main wind direction 151 or a leeward direction 152. Optionally, the audio sensor can be arranged at the outer wall of the bottom of the tower of the wind turbine generator set through magnetic attraction or other fixing methods.

[0042] The audio sensor is arranged at the position of the main wind direction or the leeward direction. The wind speed at the position of the main wind direction or the leeward direction is small, and the wind noise is small. Therefore, the wind noise interference in the collected blade rotating audio is small, and the fault diagnosis result is more accurate. Compared with being arranged on the nacelle of the wind turbine generator set or above the tower door at the bottom of the tower, the wind speed is large when being arranged near the nacelle. The wind noise in the collected audio is large, and it is difficult to record clear blade rotating audio. When being arranged above the tower door (at the direction of 90° with the main wind direction), the wind speed at the position of the tower door is the fastest in the airflow field of the wind. The wind noise in the collected audio is also large. The wind speed at the position of the main wind direction or the leeward direction is small in the airflow field. The wind noise interference in the collected audio is relatively small, and clear blade rotating audio can be collected.

[0043] When the audio collection device 110 includes two or more audio sensors 111, at least one audio sensor 111 is arranged at the position of the main wind direction or the leeward direction of the tower of the wind turbine generator set. The remaining audio sensors 111 can be arranged at other positions, for example, can be arranged at the position of 90° with the main wind direction of the tower, or can be uniformly distributed around the tower of the wind turbine generator set, and at least one audio sensor 111 is arranged at the position of the main wind direction or the leeward direction. Optionally, the two or more audio sensors 111 can be arranged at the same cross section of the tower, or can be arranged at different cross sections of the tower.

[0044] The processor 120 can be arranged inside the tower of the wind turbine generator set and connected with the audio sensor 111. The processor 120 is used to receive the blade rotating audio collected by the audio sensor 111, pre-process the blade rotating audio based on the wind noise filtering algorithm, obtain the blade rotating audio filtered by the wind noise, segment the blade rotating audio filtered by the wind noise, obtain the audio segment corresponding to each blade, and diagnose whether each audio segment corresponding to each blade exists fault according to each audio segment. The processor 120 can be executed by a program to realize the diagnosis of whether the blade exists fault according to the blade rotating audio.

[0045] Optionally, the system can further include a server connected with the processor 120. The server can be used to obtain the result of diagnosing whether the blade exists fault by the processor 120, and prompt in the case that the blade exists fault. The server can be arranged in the control room of the wind farm and communicate with the processor 120 arranged in the tower of the wind turbine generator set through the switch.

[0046] The embodiment of the present application also provides a blade fault diagnosis method, which can be executed by the processor in the blade fault diagnosis system provided by the embodiment of the present application. For the part of the steps executed by the processor in the blade fault diagnosis system provided by the embodiment of the present application which is not described in detail, reference can be made to the description of the blade fault diagnosis method provided by the embodiment of the present application.

[0047] The blade fault diagnosis method provided by the embodiment of the present application is introduced below.

[0048] Figure 3 The flowchart of the blade fault diagnosis method provided by the embodiment of the present application is shown. As shown in the figure, Figure 3 The method comprises the following steps 201-204:

[0049] Step 201: acquiring the blade rotation audio collected by the audio collection device in the operation process of the wind turbine generator.

[0050] The audio collection device can be used to collect the blade rotation audio. In the operation process of the wind turbine generator, the rotation of the blade will produce sound, and the sound signal collected by the audio collection device in real time can generate the blade rotation audio.

[0051] The audio collection device can communicate with the execution party of the blade fault diagnosis method provided by the embodiment of the present application through wired communication mode such as optical fiber communication cable or wireless communication mode, so as to send the blade rotation audio collected by the audio collection device to the execution party. For example, the execution party of the blade fault diagnosis method can be a server arranged in the central control room, or the execution party can also be the processor 120 of the blade fault diagnosis system of the embodiment shown in Figure 1 The processor 120 of the blade fault diagnosis system of the embodiment shown in

[0052] Optionally, the collection time length can be a preset time length, for example, 1 minute of blade rotation audio is collected each time. Step 202: pre-processing the blade rotation audio based on a wind noise filtering algorithm to obtain the blade rotation audio filtered from wind noise.

[0053] The wind noise filtering algorithm is used to filter the wind noise in the blade rotation audio. Since the audio collection device is in the wind field, the collected audio includes wind noise, that is, the noise of the wind. In order to obtain clearer blade rotation audio, the wind noise in the blade rotation audio can be eliminated by the wind noise filtering algorithm.

[0054] Optionally, the wind noise filtering algorithm can be a Kalman filtering algorithm, a median filtering algorithm, an arithmetic mean filtering algorithm, a sliding average filtering algorithm, etc., and the embodiments of the present application do not limit this.

[0055] Since the wind noise is a white noise, the frequency of the wind noise can be counted according to the historical record of the collected audio, and a frequency filter of the audio is designed to reduce the components in some frequency bands in the blade rotation audio.

[0056] Optionally, after the blade rotation audio is obtained, the blade rotation audio is converted to the frequency domain through Fourier transform, and then a designed frequency filter (wind noise filtering algorithm) is used in the frequency domain to reduce the components in the frequency band where the wind noise is located, thereby obtaining the blade rotation audio after filtering out the wind noise.

[0057] In step 203, the blade rotation audio after filtering out the wind noise is segmented to obtain the audio segment corresponding to each blade.

[0058] Since the audio sensor of the audio acquisition device is fixed, the sound intensity (unit: decibel) of each blade in one rotation cycle of the wind turbine generator set is from small to large and then to small, and therefore, this characteristic can be used to segment the blade rotation audio to obtain the audio segment corresponding to each blade, and in the audio segment corresponding to each blade, the sound intensity of the corresponding blade is the main component of the sound intensity in the audio segment. By dividing the blade rotation audio, the accuracy of subsequent diagnosis of blade faults can be improved.

[0059] Optionally, segmenting the blade rotation audio after filtering out the wind noise to obtain the audio segment corresponding to each blade in step 203 can include performing the following steps:

[0060] In step 2031, the blade rotation audio after filtering out the wind noise is processed through short-time Fourier transform to obtain a first feature value.

[0061] The first feature value is used to represent the frequency domain feature of the blade rotation audio after filtering out the wind noise.

[0062] The short-time Fourier transform uses a fixed window function, and then moves the window function to calculate the power spectrum at each time, i.e., to obtain the first feature value.

[0063] Optionally, the window length of the window function used by the short-time Fourier transform can be a preset value, for example, the blade rotation audio is Fourier transformed with a 10-second window length. Through the short-time Fourier transform, the audio frequency domain feature in each window can be better obtained, so that the result of identifying the segmentation time point according to the result of the short-time Fourier transform is more accurate.

[0064] In step 2032, the first feature value is input into the blade identification model to obtain the segmentation time point of the blade rotation audio filtered from wind noise.

[0065] After obtaining the short-time Fourier transform result (first feature value) of the blade rotation audio, the first feature value is input into the blade identification model.

[0066] The blade identification model is a pre-trained model, which is used to identify the switching time point of the rotation sound of different blades according to the first feature value.

[0067] Optionally, the blade identification model can be a machine learning-based algorithm model, for example, a support vector machine (SVM) based model, or a k-nearest neighbor algorithm, or a convolutional neural network (CNN) model, etc., which is not limited in the embodiments of the present application.

[0068] The support vector machine is a kind of supervised learning method, which is a generalized linear classifier that can perform binary classification on data, and its decision boundary is a maximum-margin hyperplane solved by learning samples. Supervised learning is a process of adjusting the parameters of the classifier using a set of known class samples to achieve the required performance. The hyperplane is a linear subspace of dimension one in n-dimensional Euclidean space, that is, it must be (n-1) dimensional.

[0069] In step 2033, the blade rotation audio filtered from wind noise is segmented according to the segmentation time point to obtain an audio segment corresponding to each blade.

[0070] The output result of the blade identification model is the segmentation time point, and the audio segment corresponding to each blade can be cut from the blade rotation audio according to the segmentation time point. Optionally, the audio segment of each blade in one rotation period can be cut, that is, the number of cut audio segments is the same as the number of blades of the wind turbine generator.

[0071] For example, for the blade rotation audio, the segmentation time points are 1 second, 2.1 seconds, 3.08 seconds, and 4.13 seconds, which are used to represent the plurality of audio segments obtained by segmenting the blade rotation audio, including: an audio segment 1 in a period of 1-2.1 seconds, an audio segment 2 in a period of 2.1-3.08 seconds, and an audio segment 3 in a period of 3.08-4.13 seconds. It should be noted that each audio segment cannot distinguish which blade in the actual corresponds to, and different audio segments corresponding to different blades can be distinguished by different identifiers, for example, the audio segment 1 corresponds to the blade 1, the audio segment 2 corresponds to the blade 2, and the audio segment 3 corresponds to the blade 3.

[0072] In step 204, according to each audio segment, whether the blade corresponding to each audio segment has a fault is diagnosed.

[0073] Since each audio segment is the rotating sound of the corresponding blade, according to the characteristics of the audio segment in the time domain and / or the frequency domain, whether the blade corresponding to each audio segment has a fault can be diagnosed. Since the rotating sound of the blade will change when the blade has a fault, such as breaking, icing, etc., the sound intensity and frequency change, and thus whether the blade has a fault can be identified by analyzing the audio segment.

[0074] Optionally, step 204 of diagnosing whether the corresponding blade has a fault according to each audio segment can include the following steps:

[0075] In step 2041, each audio segment is processed by Fourier transform respectively to obtain a second characteristic value of each audio segment, and the second characteristic value is used to represent the frequency domain characteristics of each audio segment.

[0076] After Fourier transform of the audio segment, the amplitude of each frequency of the audio segment, that is, the second characteristic value, can be obtained, which can represent the characteristics of the audio segment in the frequency domain, wherein the amplitude of each frequency represents the signal power of the audio segment at the corresponding frequency.

[0077] In step 2042, the second characteristic value of each audio segment is input into a blade fault diagnosis model to obtain a first fault diagnosis result of each blade.

[0078] The blade fault diagnosis model is a model pre-trained to identify whether the corresponding blade has a fault according to the second characteristic value of the audio segment. Optionally, the blade fault diagnosis model can also be a machine learning-based algorithm model, for example, a model of support vector machine (SVM) or a k-nearest neighbor algorithm.

[0079] Optionally, in addition to the above fault diagnosis according to the frequency domain features of each audio segment, fault diagnosis can also be performed according to the correlation of the audio segment time lengths of multiple blades. In the case where the blades are not faulty, the audio segment time lengths of each blade should be similar. After a faulty blade exists, the corresponding blade rotation sound may become larger, that is, the sound intensity becomes larger, so that when the blade rotation audio is segmented according to the sound intensity, the time length of the corresponding blade increases, the time lengths of other blades decrease, and the difference in time length becomes larger.

[0080] Based on the above principle, step 204 can further include the following steps:

[0081] Step 2043, counting the time length of the audio segment corresponding to each blade;

[0082] Step 2044, determining whether there is a faulty blade according to whether the difference in time length of each two audio segments exceeds a preset threshold, to obtain a second fault diagnosis result;

[0083] If a blade is faulty, the rotation sound intensity of the blade will change, and usually the rotation sound of the faulty blade becomes larger, which will cause the audio segment time length of the faulty blade to be longer when the audio segment of each blade is intercepted, and there is a more obvious difference in the audio segment time length of the faulty blade and that of other normal blades. According to the above principle, a time length threshold (preset threshold) can be preset. If the time length difference of any two audio segments exceeds the preset threshold, the wind turbine generator set has a faulty blade.

[0084] After step 2042 and step 2044 are performed, whether there is a faulty blade is determined in combination with the first fault diagnosis result and the second fault diagnosis result of each blade.

[0085] The first fault diagnosis result is a result obtained by independently performing fault diagnosis on each audio segment through the blade fault identification model. For the audio segment corresponding to each blade, a corresponding first fault diagnosis result can be obtained, for example, state values 0 and 1 can be used to respectively represent that the corresponding blade is not faulty and is faulty.

[0086] The second fault diagnosis result is a result of determining whether there is a faulty blade according to whether the time length of the audio segment is consistent.

[0087] Combining the first fault diagnosis result and the second fault diagnosis result for determination can more accurately determine whether there is a faulty blade.

[0088] In different seasons or weather, blades may be faulty due to environmental reasons, for example, lightning is prone to occur in summer, wind speed is too large in autumn, and blades may be covered with ice in winter, etc.

[0089] Therefore, different blade fault diagnosis models can be trained according to different fault types, and then a suitable blade fault diagnosis model can be selected according to different environments to perform fault diagnosis on the collected blade rotating audio. Specifically, in an optional embodiment, before diagnosing whether the corresponding blade has a fault according to each audio segment, an environmental parameter of the wind turbine generator is obtained, wherein the environmental parameter is used to represent the season and / or the weather, and then among a plurality of candidate fault diagnosis models respectively used to identify different fault types, a model used to identify the fault type corresponding to the environmental parameter is selected to determine the used fault diagnosis model. That is, each environmental parameter corresponds to a fault type, so that the blade fault diagnosis model corresponding to the fault type is selected to perform fault diagnosis.

[0090] The blade fault diagnosis method provided by the embodiments of the present application can obtain blade rotating audio after filtering wind noise by performing a wind noise filtering algorithm on the blade rotating audio, so as to remove the interference of wind noise in the blade rotating audio and obtain audio carrying more accurate and stronger blade rotating sound; then, the blade rotating audio after filtering wind noise is segmented to obtain audio segments corresponding to each blade, and whether the corresponding blade of each audio segment has a fault is diagnosed according to each audio segment, which can diagnose whether the corresponding blade has a fault according to the audio segments of different blades respectively, thereby improving the accuracy of the diagnosis result.

[0091] In an optional embodiment, the audio acquisition device can obtain blade rotating audio collected at a plurality of positions, and then a blade rotating audio with the smallest wind noise and the best audio quality can be selected from the blade rotating audio at the plurality of positions to perform the following steps 203 and 204.

[0092] Correspondingly, the blade rotating audio collected by the audio acquisition device during the operation of the wind turbine generator in step 201 can be blade rotating audio collected at a plurality of positions;

[0093] The blade rotating audio after filtering wind noise obtained by preprocessing the blade rotating audio based on the wind noise filtering algorithm in step 202 can be processed by the wind noise filtering algorithm for each blade rotating audio to obtain a plurality of blade rotating audios after filtering wind noise.

[0094] After obtaining the plurality of blade rotation audios filtered wind noise, the wind noise parameters of each blade rotation audio filtered wind noise can be calculated respectively by a wind noise identification model. The wind noise identification model is a pre-trained model for evaluating the wind noise parameters of the audio. Optionally, the blade fault diagnosis model can also be a machine learning-based algorithm model, such as an SVM model, or a k-nearest neighbor algorithm, a CNN algorithm, etc. The embodiments of the present application do not limit this. Then, according to the wind noise parameters, the blade rotation audio filtered wind noise with the minimum wind noise is selected from the plurality of blade rotation audios filtered wind noise for segmentation, to obtain the audio segment corresponding to each blade. In this way, the blade rotation audio with the minimum wind noise can be selected, which can increase the probability of using the blade rotation audio and improve the real-time performance of the blade fault diagnosis method.

[0095] Optionally, when calculating the wind noise parameters of each blade rotation audio filtered wind noise by the wind noise identification model, an optional implementation is that first, each blade rotation audio filtered wind noise is processed respectively by Fourier transform to obtain third feature values of each blade rotation audio filtered wind noise, wherein the third feature values are used to represent the frequency domain characteristics of the corresponding blade rotation audio filtered wind noise. Then, the third feature values of each blade rotation audio filtered wind noise are input into the wind noise identification model to obtain the wind noise parameters of each blade rotation audio filtered wind noise.

[0096] The blade fault diagnosis model, the wind noise identification model, and the blade identification model described in the embodiments of the present application can be trained machine learning-based models, for example, an SVM model, etc. An optional model training method is shown in Figure 4 The training method shown in Figure 4 The flow of the training method is described below. It should be noted that the other models described in the embodiments of the present application can also be trained based on the training method shown in Figure 4 .

[0097] First, a plurality of training audio files are obtained. The training audio file is an audio file used for training the model. Then, each training audio file is labeled. For example, in order to train the blade identification model, each training audio file is a blade rotation audio collected for a preset time length (such as 10 seconds). After obtaining the blade rotation audio, a waveform diagram of the sound intensity of the blade rotation audio can be displayed. Since the blade rotates from far to near and then to far, the sound intensity of the blade rotation also changes from small to large and then to small. Therefore, the blade rotation audio can be segmented according to the sound intensity. A wind turbine usually has three blades, so three audio segments corresponding to one rotation period are segmented in the blade rotation audio.

[0098] For example, for the blade rotation audio 1, the tags can be 1 second, 2.1 seconds, 3.08 seconds, 4.13 seconds, which are used to represent a plurality of audio segments obtained by segmenting the blade rotation audio 1, including: an audio segment 1 in a period of 1-2.1 seconds, an audio segment 2 in a period of 2.1-3.08 seconds, and an audio segment 3 in a period of 3.08-4.13 seconds. It should be noted that each audio segment cannot distinguish which blade in the actual corresponds to, and different audio segments corresponding to different blades can be distinguished by different identifications, for example, the audio segment 1 corresponds to the blade 1, and the audio segment 2 corresponds to the blade 2.

[0099] After being tagged, a part of the blade rotation audios with segmented time tags is taken as a training set, and another part is taken as a test set. The blade recognition model is trained through the training set. Each time a blade rotation audio in the training set is obtained, the segmentation time point is recognized through the blade recognition model, and the result obtained by the model is compared with the tag. According to the preset parameter modification algorithm, the parameters of the model are modified, and the updated model is used to recognize the next training audio file in the training set again.

[0100] When the training of the model using all the files in the training set is completed, the model is used to recognize the test set, and the accuracy of the recognition is evaluated. If it is qualified, a plurality of new blade rotation audios can be used to generate a training set and a test set, and the model is further modified. After multiple modifications, the final model is obtained. If it is not qualified, the tags on the files can be adjusted, and the model is retrained using the current training set until the accuracy is qualified.

[0101] An optional flowchart for identifying wind noise in a blade rotation audio through a wind noise recognition model is shown in Figure 5 First, a blade rotation audio is read, and data preprocessing is performed, such as preprocessing through a wind noise filtering algorithm. Then, the blade rotation audio is subjected to Fourier transform to obtain characteristic values. Then, the wind noise parameter is calculated by the wind noise recognition model according to the characteristic values of the blade rotation audio. The larger the wind noise, the larger the wind noise parameter, and the smaller the wind noise, the smaller the wind noise parameter. In the case where the wind noise parameter exceeds the threshold value, the wind noise is too large, and the corresponding blade rotation audio can be abandoned for fault diagnosis. In the case where the wind noise parameter does not exceed the threshold value, the blade rotation audio with the smallest wind noise parameter can be selected from a plurality of blade rotation audios that do not exceed the threshold value, as the audio for diagnosing blade faults.

[0102] An optional flowchart for segmenting a blade rotation audio through a blade recognition model can be as shown in Figure 6As shown, first, a blade rotation audio is read, and data preprocessing is performed, such as noise removal and the like. Then, the blade rotation audio is subjected to short-time Fourier transform according to a preset window length to obtain first characteristic values. The first characteristic values can be a sequence of amplitudes corresponding to multiple frequencies. After obtaining the first characteristic values, the first characteristic values are input into a blade identification model, and the blade identification model outputs time points of blade segmentation. Then, the blade rotation audio can be segmented according to the segmentation time points to extract audio segments corresponding to each blade in a rotation period.

[0103] An optional flowchart for diagnosing blade faults in combination with a blade fault diagnosis model is shown in FIG. 6. Figure 7 After the three-blade audio segments are segmented, the blade fault model can be used to diagnose the fault of each audio segment to determine the state of the blade. Specifically, Fourier transform can be performed on each audio segment to obtain second characteristic values of the corresponding audio segment, and then the second characteristic values of each audio segment are input into the corresponding blade fault diagnosis model to obtain the fault diagnosis result of the corresponding audio segment. For example, 0 can represent normal and 1 can represent abnormal, and for each audio segment, 0 or 1 is output to represent the state of the corresponding blade. In addition, after each segmented blade audio segment is obtained, the length of the audio segment can be used to determine whether there is an abnormal blade to obtain a second fault diagnosis result. For example, if the lengths of audio segment 1 and audio segment 2 differ by 0.1 seconds, which is less than the threshold of 0.2 seconds, and the lengths of audio segment 1 and audio segment 3 differ by 0.4 seconds, and the lengths of audio segment 2 and audio segment 3 differ by 0.3 seconds, which is greater than the threshold of 0.2 seconds, then it is determined that the states of audio segment 1 and 2 are consistent, and the state of audio segment 3 is inconsistent. Further, the blade state decision maker can determine the state of the blade and output the state of the blade in combination with the first fault diagnosis result of each blade and the consistency detection result (second fault diagnosis result) according to the audio length.

[0104] As an optional specific embodiment, the flowchart of the blade fault diagnosis model provided by the embodiments of the present application is shown in FIG. 6. Figure 8As shown. After the blade fault diagnosis system is powered on, the delay waits for the start of other devices of the wind turbine generator. After the delay preset time, communication is performed with the server arranged in the control room of the wind farm. If the connection fails, further delay is waited. If the connection is successful, the initialization file can be loaded for initialization configuration. The initialization configuration can include configuration of parameters such as the number of audio sensors, the time length of audio data acquisition, and the time length of audio data storage. Then, the audio sensor can send the real-time acquired audio data to the blade fault diagnosis system. Then, it is judged whether there is wind noise through the wind noise identification model. If there is noise, filtering is performed through the wind noise filtering algorithm, and then it is identified whether there is noise through the wind noise identification model. If there is still noise, a state value of the fault diagnosis result is returned: the blade is not identified. If no noise is identified through the wind noise identification model, a blade rotating audio with the smallest wind noise can be selected from the multiple blade rotating audios according to the wind noise parameters obtained through the wind noise identification model.

[0105] Then, the selected blade rotating audio is segmented through the blade identification algorithm to obtain an audio segment corresponding to each blade. Then, it can be diagnosed whether there is a fault through the fault diagnosis algorithm as shown in Figure 7 If there is a fault, a state value of the fault diagnosis result is returned: there is a blade fault, and the file of the blade rotating audio of the fault can be saved for record saving. If there is no fault, a state value of the blade normal is returned. Next, a period of time can be delayed, and then real-time stream data acquired by the audio sensor is acquired again, and then fault diagnosis is performed to realize the effect of real-time monitoring of the blade.

[0106] The application also provides another embodiment of a blade fault diagnosis system, as shown in Figure 9 In a wind farm (wind farm one), a plurality of wind turbines are arranged. Each wind turbine is provided with an audio sensor for collecting a sound signal of a blade rotating of the wind turbine and sending the sound signal to an edge processor. The edge processor can be arranged in a tower of the wind turbine. The edge processor can process the sound signal collected by the audio sensor into an audio to obtain a blade rotating audio, and the blade rotating audio is sent to a server in a control room of the wind farm through a cable. Figures 3 to 8 Any one of the optional embodiments provides a blade fault diagnosis method to obtain a diagnosis result. Further, the blade fault diagnosis result can be sent to a core optical fiber switch of a control room of a wind farm area one ring network through a switch at the bottom of the wind turbine tower via a cable, and then sent to a server of the control room. The server can display a display interface of a WEB front end, display the blade fault diagnosis condition through an interface program, and issue an alarm prompt when the blade has a fault. Another schematic diagram of the blade fault diagnosis system provided by the embodiment of the application can be as shown in Figure 10As shown, the server can be configured with multiple candidate blade fault diagnosis models, based on the environmental parameters selected by the user on the front end, select the corresponding blade fault diagnosis model, and inform the edge processor. After receiving the blade fault diagnosis model selected by the server, the edge processor configures the edge processor program to perform fault diagnosis according to the blade fault diagnosis model selected by the server.

[0107] Optionally, an audio processing unit can be included in the edge processor of each fan, which can be used to collect real-time audio stream data, perform fault diagnosis, save fault data (which can include fault diagnosis results and fault audio), and communicate with the server. Among them, when performing fault diagnosis, the data can be cached, and if there is a fault, the fault data is saved. After obtaining the fault diagnosis result, the data stored in the cache area can be uploaded to the server host computer. In addition, the edge processor can also accept the control of the server host computer, for example, according to the server initialization configuration, configure the number of audio sensors, data storage time, maximum data cache space, etc. Optionally, it can also receive the selection of the blade fault diagnosis model by the server according to the environmental parameters.

[0108] The server algorithm program of the central control room can be provided with functions such as unit connection state verification, unit blade state communication, control configuration file distribution, and fault data reception; the WEB front-end interface of the server of the central control room can be provided with functions such as unit connection state display, unit blade state display, and unit fault data playback.

[0109] The blade fault diagnosis method, device and system provided by the embodiment of the application can monitor the state of the fan blade in real time by processing audio data, which has the advantages of not being affected by light, being able to monitor in real time at night, not being affected by heavy fog, and being able to effectively monitor the blade state for a long time, so as to identify the blade abnormal sound problem caused by the blade itself due to reasons such as leading edge corrosion, lightning strike, fracture, and crack in the first time, and then timely maintenance can be performed to avoid blade failure caused by accumulated blade faults.

[0110] The embodiment of the application also provides a blade fault diagnosis device which can be used to execute the blade fault diagnosis method provided by the embodiment of the application. The parts not described in detail in the blade fault diagnosis device provided by the embodiment of the application can refer to the description of the blade fault diagnosis method provided by the embodiment of the application, which will not be described here.

[0111] As shown in the figure, Figure 11 The blade fault diagnosis device provided by the embodiment of the application includes an acquisition module 11, a preprocessing module 12, a segmentation module 13, and a diagnosis module 14.

[0112] The acquisition module is configured to acquire blade rotation audio collected by the audio acquisition device during operation of the wind turbine generator set; the preprocessing module is configured to preprocess the blade rotation audio based on a wind noise filtering algorithm to obtain blade rotation audio from which wind noise is filtered out; the segmentation module is configured to segment the blade rotation audio from which wind noise is filtered out to obtain audio segments corresponding to each blade; and the diagnosis module is configured to diagnose whether each blade corresponding to each audio segment has a fault based on the audio segment.

[0113] Optionally, segmenting the blade rotation audio from which wind noise is filtered out to obtain audio segments corresponding to each blade comprises: processing the blade rotation audio from which wind noise is filtered out through short-time Fourier transform to obtain first characteristic values, the first characteristic values being used to represent frequency domain characteristics of the blade rotation audio from which wind noise is filtered out; inputting the first characteristic values into a blade recognition model to obtain segmentation time points of the blade rotation audio from which wind noise is filtered out, wherein the blade recognition model is a pre-trained model used to identify switching time points of rotation sounds of different blades based on the first characteristic values; and segmenting the blade rotation audio from which wind noise is filtered out based on the segmentation time points to obtain audio segments corresponding to each blade.

[0114] Optionally, diagnosing whether each blade corresponding to each audio segment has a fault based on the audio segment comprises: processing each audio segment through Fourier transform respectively to obtain second characteristic values of each audio segment, the second characteristic values being used to represent frequency domain characteristics of each audio segment; inputting the second characteristic values of each audio segment into a blade fault diagnosis model respectively to obtain first fault diagnosis results of each blade, the blade fault diagnosis model being a pre-trained model used to identify whether each blade corresponding to each audio segment has a fault based on the second characteristic values of the audio segment.

[0115] Optionally, diagnosing whether each blade corresponding to each audio segment has a fault based on the audio segment comprises: counting lengths of the audio segments corresponding to each blade; determining whether there is a blade with a fault based on whether a difference between the lengths of each two audio segments exceeds a preset threshold to obtain second fault diagnosis results; and determining whether there is a blade with a fault by combining the first fault diagnosis results of each blade and the second fault diagnosis results.

[0116] Optionally, before diagnosing whether each blade corresponding to each audio segment has a fault based on the audio segment, the method further comprises: acquiring an environmental parameter of the wind turbine generator set, wherein the environmental parameter is used to represent a season and / or weather; and determining, from a plurality of candidate fault diagnosis models respectively used to identify different fault types, a model used to identify a fault type corresponding to the environmental parameter to determine a fault diagnosis model to be used.

[0117] Optionally, the blade rotating audio collected by the audio collection device during the operation of the wind turbine generator set is obtained, including: obtaining blade rotating audios collected at multiple positions; pre-processing the blade rotating audios based on a wind noise filtering algorithm to obtain blade rotating audios filtered of wind noise, including: performing wind noise filtering algorithm processing on each blade rotating audio to obtain multiple blade rotating audios filtered of wind noise; after obtaining the multiple blade rotating audios filtered of wind noise, the method further includes: calculating a wind noise parameter of each blade rotating audio filtered of wind noise by a wind noise identification model, wherein the wind noise parameter is used to represent the wind noise in the audio, and the wind noise identification model is a pre-trained model for evaluating the wind noise parameter of the audio; segmenting the blade rotating audios filtered of wind noise to obtain audio segments corresponding to each blade, including: according to the wind noise parameter, selecting the blade rotating audio filtered of wind noise with the smallest wind noise from the multiple blade rotating audios filtered of wind noise for segmentation to obtain audio segments corresponding to each blade.

[0118] Optionally, the wind noise parameter of each blade rotating audio filtered of wind noise is calculated by a wind noise identification model, including: processing each blade rotating audio filtered of wind noise by a Fourier transform to obtain a third feature value of each blade rotating audio filtered of wind noise, the third feature value being used to represent the frequency domain characteristics of the corresponding blade rotating audio filtered of wind noise; inputting the third feature value of each blade rotating audio filtered of wind noise into the wind noise identification model to obtain the wind noise parameter of each blade rotating audio filtered of wind noise.

[0119] The blade fault diagnosis device provided by the embodiments of the present application can process the blade rotating audio by a wind noise filtering algorithm to obtain blade rotating audio filtered of wind noise, thereby removing the interference of wind noise in the blade rotating audio and obtaining audio carrying more accurate and stronger blade rotating sound; then, the blade rotating audio filtered of wind noise is segmented to obtain audio segments corresponding to each blade, and whether each audio segment corresponds to a blade with a fault is diagnosed according to each audio segment, which can diagnose whether each blade has a fault according to the audio segments of different blades, thereby improving the accuracy of the diagnosis result.

[0120] It should be noted that the present application is not limited to the specific configurations and processes described above and shown in the drawings. For the sake of brevity, detailed descriptions of well-known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present application is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between steps, after understanding the spirit of the present application.

[0121] The functions noted in the description of the structural block diagrams above can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, and the like. When implemented in software, the elements of the present application are program or code segments that are used to perform the required tasks. The program or code segments can be stored in a machine-readable medium, or transmitted through a data signal carried in a carrier wave over a transmission medium or communication link. A "machine-readable medium" includes any medium that can store or transport information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, and the like. The code segments can be downloaded via computer networks such as the Internet, intranets, and the like.

[0122] It is also important to note that the examples mentioned in the present application describe some methods or systems based on a series of steps or devices. However, the present application is not limited to the order of the steps mentioned above, that is, the steps can be performed in the order mentioned in the examples, or in an order different from the examples, or several steps can be performed simultaneously.

[0123] The computer program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other processing devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other processing devices to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks. These computer program instructions can also be stored in a computer readable medium that can direct a computer, other programmable data processing apparatus, or other processing devices to operate in a particular manner, such that the instructions stored in the computer readable medium produce an article of manufacture including instructions which implement the function / act specified in the flowchart and / or block diagram block or blocks. The computer program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other processing devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other processing devices to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks. These computer program instructions can also be stored in a computer readable medium that can direct a computer, other programmable data processing apparatus, or other processing devices to operate in a particular manner, such that the instructions stored in the computer readable medium produce an article of manufacture including instructions which implement the function / act specified in the flowchart and / or block diagram block or blocks.

[0124] The above merely describes a specific implementation of the present application. Those skilled in the art can clearly understand the specific working processes of the system, modules and units described above for the convenience and brevity of description, and can refer to the corresponding processes in the foregoing method embodiments, which will not be described herein again. It should be understood that the protection scope of the present application is not limited to this, and any person skilled in the art can easily think of various equivalent modifications or replacements within the technical range disclosed by the present application, and these modifications or replacements should be covered within the protection scope of the present application.

Claims

1. A blade failure diagnosis method characterized by comprising: The method comprises: obtaining blade rotation audio collected by an audio collection device during operation of a wind turbine generator set; preprocessing the blade rotation audio based on a wind noise filtering algorithm to obtain blade rotation audio filtered of wind noise; segmenting the blade rotation audio filtered of wind noise to obtain audio segments corresponding to each blade; diagnosing whether each audio segment corresponds to a blade with a fault; the diagnosis of whether each audio segment corresponds to a blade with a fault comprises: processing each audio segment by Fourier transform to obtain second characteristic values of each audio segment, the second characteristic values being used to represent frequency domain characteristics of each audio segment; inputting the second characteristic values of each audio segment into a blade fault diagnosis model to obtain first fault diagnosis results of each blade, the blade fault diagnosis model being a pre-trained model for identifying whether a corresponding blade has a fault based on the second characteristic values of the audio segments; counting the lengths of the audio segments corresponding to each blade; judging whether there is a blade with a fault based on whether the difference between the lengths of each two audio segments exceeds a preset threshold to obtain second fault diagnosis results; combining the first fault diagnosis results of each blade and the second fault diagnosis results to judge whether there is a blade with a fault.

2. The blade failure diagnostic method according to claim 1, characterized by, segmenting the blade rotation audio filtered of wind noise to obtain audio segments corresponding to each blade comprises: processing the blade rotation audio filtered of wind noise by short-time Fourier transform to obtain first characteristic values, the first characteristic values being used to represent frequency domain characteristics of the blade rotation audio filtered of wind noise; inputting the first characteristic values into a blade recognition model to obtain segmentation time points of the blade rotation audio filtered of wind noise, wherein the blade recognition model is a pre-trained model for identifying switching time points of rotation sounds of different blades based on the first characteristic values; segmenting the blade rotation audio filtered of wind noise based on the segmentation time points to obtain audio segments corresponding to each blade.

3. The blade failure diagnostic method according to claim 1, characterized by, Before diagnosing whether each audio segment corresponds to a blade with a fault, the method further comprises: obtaining environmental parameters of the wind turbine generator set, wherein the environmental parameters are used to represent seasons and / or weather; in a plurality of candidate fault diagnosis models respectively used to identify different fault types, determining a model used to identify a fault type corresponding to the environmental parameters to determine the fault diagnosis model to be used.

4. The blade failure diagnostic method according to any one of claims 1 to 3, characterized by, The method comprises: obtaining blade rotation audio collected by an audio collection device during operation of a wind turbine generator set; preprocessing the blade rotation audio based on a wind noise filtering algorithm to obtain blade rotation audio filtered of wind noise; the preprocessing of the blade rotation audio based on a wind noise filtering algorithm to obtain blade rotation audio filtered of wind noise comprises: performing wind noise filtering algorithm processing on each blade rotation audio to obtain a plurality of blade rotation audios filtered of wind noise; After obtaining the plurality of blade rotating audios filtered wind noise, the method further comprises: calculating a wind noise parameter of each blade rotating audio filtered wind noise by a wind noise identification model, wherein the wind noise parameter is used to represent the wind noise size in the audio, and the wind noise identification model is a pre-trained model used to evaluate the wind noise parameter of the audio; The method further comprises: selecting the blade rotating audio filtered wind noise with the smallest wind noise from the plurality of blade rotating audios filtered wind noise according to the wind noise parameter, and segmenting the blade rotating audio filtered wind noise to obtain an audio segment corresponding to each blade.

5. The blade failure diagnostic method according to claim 4, characterized by, The method further comprises: The method further comprises: processing each blade rotating audio filtered wind noise by Fourier transform to obtain a third feature value of each blade rotating audio filtered wind noise, wherein the third feature value is used to represent the frequency domain feature of the corresponding blade rotating audio filtered wind noise; The method further comprises: inputting the third feature value of each blade rotating audio filtered wind noise into the wind noise identification model to obtain the wind noise parameter of each blade rotating audio filtered wind noise.

6. A blade failure diagnosis device characterized by comprising: The device comprises: An acquisition module configured to acquire blade rotating audio collected by an audio acquisition device during operation of a wind turbine generator set; A preprocessing module configured to preprocess the blade rotating audio based on a wind noise filtering algorithm to obtain blade rotating audio filtered wind noise; A segmentation module configured to segment the blade rotating audio filtered wind noise to obtain an audio segment corresponding to each blade; A diagnosis module configured to diagnose whether each audio segment corresponding to each blade has a fault based on each audio segment; The diagnosis module is specifically configured to: The diagnosis module is specifically configured to: The diagnosis module is specifically configured to: The diagnosis module is specifically configured to: The diagnosis module is specifically configured to: The device comprises:

7. A blade failure diagnosis system characterized by comprising: An audio acquisition device comprising an audio sensor, wherein at least one audio sensor is arranged at a main wind direction position or a leeward position of a tower of a wind turbine generator set; ​ A processor is arranged in the interior of the tower tube and connected with the audio sensor, configured to receive the blade rotating audio collected by the audio sensor, pre-process the blade rotating audio based on a wind noise filtering algorithm to obtain blade rotating audio filtered of wind noise, segment the blade rotating audio filtered of wind noise to obtain audio segments corresponding to each blade, and diagnose whether each audio segment corresponding blade has a fault according to each audio segment. The processor is specifically configured to: process each audio segment by Fourier transform to obtain second characteristic values of each audio segment, the second characteristic values being used to represent frequency domain characteristics of each audio segment; input the second characteristic values of each audio segment into a blade fault diagnosis model to obtain first fault diagnosis results of each blade, the blade fault diagnosis model being a model pre-trained to identify whether a corresponding blade has a fault according to the second characteristic values of the audio segments; count the time lengths of the audio segments corresponding to each blade; determine whether there is a blade with a fault according to whether the difference between the time lengths of each two audio segments exceeds a preset threshold to obtain second fault diagnosis results; determine whether there is a blade with a fault in combination with the first fault diagnosis results of each blade and the second fault diagnosis results.

8. A storage medium, characterized by The storage medium has computer program instructions stored thereon, and the computer program instructions are executed by the processor to implement the blade fault diagnosis method in any one of claims 1-5.

Images