Offshore wind turbine blade monitoring system based on signal processing and voiceprint recognition algorithm

Abstract

The present invention discloses an offshore wind turbine blade monitoring system based on signal processing and voiceprint recognition algorithm, and relates to the technical field of wind turbine blade detection. The system includes M industrial microphones, a signal processing unit, a voiceprint feature extraction unit, a blade fault recognition unit and a blade fault alarm unit, wherein the signal processing unit is used to adopt a joint denoising algorithm based on empirical mode decomposition and wavelet threshold to perform real-time noise reduction processing on M on-site sound signals from M industrial microphones corresponding to each other, and obtain M noise-reduced sound signals, the voiceprint feature extraction unit is used to perform voiceprint feature extraction processing on the M noise-reduced sound signals, and obtain voiceprint feature data, and the blade fault recognition unit is used to import the voiceprint feature data into a blade fault recognition model that has been pre-trained based on the voiceprint recognition algorithm, and output the blade fault recognition result, so as to improve the timeliness and accuracy of fault monitoring.

Description

Technical Field

[0001] The present invention belongs to the technical field of wind turbine blade detection, and in particular relates to an offshore wind turbine blade monitoring system based on signal processing and voiceprint recognition algorithms. Background Art

[0002] Offshore wind turbines are wind turbines specially designed for offshore wind farms. Compared with onshore wind turbines, offshore wind turbines need to cope with more complex technical challenges and environmental conditions. Offshore wind turbines usually have higher technical requirements and cover more and wider disciplines and professions. Since offshore wind turbines operate in harsh environments for a long time, the expensive blades, as the core components of the wind turbines, are subjected to huge stresses for a long time and work in harsh environments, and are extremely susceptible to damage such as fatigue, cracks and corrosion. At the same time, in the early stages of blade damage, maintenance is easier and the cost is lower. However, if the damage is not discovered in time, maintenance will be difficult, the cost will be high, and even the blades will be scrapped. Therefore, real-time monitoring of the health / fault status of the blades is of great significance.

[0003] Existing wind turbine blade fault monitoring technologies usually rely on regular manual inspections and physical sensors. There are problems such as time-consuming and labor-intensive inspection methods, poor real-time performance due to the need to shut down or long inspection cycles at low wind speeds, and inaccurate blade crack detection. For this reason, some researchers have begun to judge whether the wind turbine is damaged by the sound of the wind sweeping the blades when the wind turbine is running. When the blades have structural damage, the wind sweeping sound during operation will be mixed with abnormal sound signals. At present, some wind farm personnel use sound to judge whether the blades are damaged during inspections. However, at the bottom of the tower of the offshore wind turbine, there are sounds of waves hitting the tower, as well as whistles from distant ships, which will interfere with the identification of abnormal blade sound signals.

[0004] In summary, how to provide a new solution for online blade monitoring based on sound signals suitable for offshore wind turbines, so as to accurately detect and identify blade sweeping sound signals, help operation and maintenance personnel quickly discover wind turbine blade failures, thereby improving the timeliness and accuracy of fault monitoring and reducing the potential risks brought by equipment failures, is a topic that technical personnel in this field urgently need to study. Summary of the invention

[0005] The purpose of the present invention is to provide an offshore wind turbine blade monitoring system based on signal processing and voiceprint recognition algorithm, so as to solve the problem that the existing wind turbine blade fault monitoring technology is difficult to accurately extract the characteristics of the blade sweeping wind sound signal due to large interference from other sounds when used in offshore wind turbines, thereby affecting the accuracy of the blade fault identification results.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] The present invention provides an offshore wind turbine blade monitoring system based on signal processing and voiceprint recognition algorithm, comprising An industrial microphone, a signal processing unit, a voiceprint feature extraction unit, a blade fault identification unit and a blade fault alarm unit, among which: represents a positive integer greater than or equal to 3, Industrial microphones are installed at intervals on the bottom outer peripheral surface of the tower of the target offshore wind turbine;

[0008] The industrial microphone is communicatively connected to the signal processing unit, and is used to collect on-site sound signals in real time, and transmit the on-site sound signals to the signal processing unit in real time;

[0009] The signal processing unit is communicatively connected to the voiceprint feature extraction unit, and is used to adopt a joint denoising algorithm based on empirical mode decomposition and wavelet threshold to denoise the signals from the voiceprint feature extraction unit in a one-to-one correspondence. Industrial pickups The live sound signal is processed in real time to reduce noise, and the The live sound signals correspond one to one a noise-reduced sound signal, and the The noise-reduced sound signal is transmitted to the voiceprint feature extraction unit in real time;

[0010] The voiceprint feature extraction unit is communicatively connected to the blade fault identification unit, and is used to respectively Performing voiceprint feature extraction processing on the noise-reduced sound signal to obtain voiceprint feature data, and transmitting the voiceprint feature data to the blade fault identification unit;

[0011] The blade fault identification unit is communicatively connected to the blade fault alarm unit, and is used to import the voiceprint feature data into a blade fault identification model that has been pre-trained based on a voiceprint recognition algorithm, output a blade fault identification result for the wind turbine blade of the target offshore wind turbine, and transmit the blade fault identification result to the blade fault alarm unit;

[0012] The blade fault alarm unit is used to trigger a fault alarm action for the fan blade when it is found that the probability of judging that the fan blade is normal is less than the probability of judging that the fan blade is abnormal according to the blade fault identification result.

[0013] Based on the above invention content, a new solution for online monitoring of blades based on sound signals for offshore wind turbines is provided, which includes M industrial microphones, a signal processing unit, a soundprint feature extraction unit, a blade fault identification unit and a blade fault alarm unit, wherein the signal processing unit is used to adopt a joint denoising algorithm based on empirical mode decomposition and wavelet threshold to perform real-time noise reduction processing on M on-site sound signals corresponding to each other from the M industrial microphones, so as to obtain M noise-reduced sound signals; the soundprint feature extraction unit is used to perform soundprint feature extraction processing on the M noise-reduced sound signals, so as to obtain soundprint feature data; the blade fault identification unit is used to import the soundprint feature data into a blade fault identification model that has been pre-trained based on the soundprint recognition algorithm, and output the blade fault identification result; in this way, the blade sweeping wind sound signal can be accurately detected and identified through the combination of the joint denoising algorithm and the soundprint recognition algorithm, so as to help operation and maintenance personnel quickly discover the fault of the wind turbine blade, thereby improving the timeliness and accuracy of fault monitoring, reducing the potential risks brought by equipment failure, and facilitating practical application and promotion.

[0014] In a possible design, the number of the industrial microphones is eight, and they are installed on the bottom outer peripheral surface of the tower at equal intervals in an annular direction around the center of the horizontal cross section of the tower.

[0015] In a possible design, a joint denoising algorithm based on empirical mode decomposition and wavelet threshold is used to denoise the noise from the two images one by one. Industrial pickups The live sound signal is processed in real time to reduce noise, and the The live sound signals correspond one to one The noise-reduced sound signal includes:

[0016] For the respective corresponding Industrial pickups Each sound signal in the live sound signal is subjected to empirical mode decomposition processing to obtain the corresponding IMF components, where represents a positive integer greater than or equal to 3;

[0017] For each of the sound signals, according to the corresponding Intrinsic mode function components, determine the corresponding component center frequencies, where The center frequency of the components is The intrinsic mode function components correspond one to one;

[0018] According to the component center frequency points, and finding whether there is at least one adjacent frequency point group, wherein the adjacent frequency point group contains frequency points corresponding to different sound signals. The component center frequency points, The maximum frequency difference between the center frequencies of the components is less than or equal to the preset frequency threshold. Indicates greater than or equal to 3 and less than or equal to A positive integer of ;

[0019] If so, for each adjacent frequency point group in the at least one adjacent frequency point group, according to the corresponding The center frequencies of the components correspond one to one The intrinsic mode function components and the The intrinsic mode function components correspond one to one The known installation positions of the industrial microphones are used to determine the corresponding sound source positions, and when it is determined that the sound source positions are far from the active area of ​​the wind turbine blades of the target offshore wind turbine, the sound source positions are respectively The intrinsic mode function components are subjected to wavelet threshold denoising to obtain the same The intrinsic mode function components correspond one to one The denoised intrinsic mode function components;

[0020] For each of the sound signals, a corresponding denoised sound signal is reconstructed based on all the corresponding denoised intrinsic mode function components and all the non-denoised intrinsic mode function components.

[0021] In a possible design, according to the component center frequency points, and find out whether there is at least one adjacent frequency point group, including:

[0022] In the frequency domain, the sound signals are examined in sequence from small to large frequency. a component center frequency point, if it is found that a component center frequency point of a sound signal in each of the sound signals is located at the current frequency point and the current frequency point is not within the established frequency domain window, then a new frequency domain window is created with the current frequency point as the starting frequency point and the frequency domain width is equal to the preset frequency threshold;

[0023] For each of the established frequency domain windows, determine whether the total number of frequency points of multiple component center frequency points corresponding to different sound signals in the corresponding window is greater than or equal to 3, and if so, include the multiple component center frequency points into an adjacent frequency point group, wherein the adjacent frequency point group includes the frequency points corresponding to different sound signals. The component center frequency points, The maximum frequency difference between the center frequency points of the components is less than or equal to the preset frequency threshold, Indicates greater than or equal to 3 and less than or equal to A positive integer.

[0024] In one possible design, for each adjacent frequency point group in the at least one adjacent frequency point group, according to the corresponding The center frequencies of the components correspond one to one The intrinsic mode function components and the The intrinsic mode function components correspond one to one The known installation positions of the industrial microphones are used to determine the corresponding sound source positions, including:

[0025] For a certain adjacent frequency point group in the at least one adjacent frequency point group, determine the The center frequencies of the components correspond one to one IMF components;

[0026] For the For each pair of intrinsic mode function components in the eigenmode function components, the corresponding signal propagation time difference value is calculated according to the corresponding two eigenmode function components;

[0027] According to the The intrinsic mode function components correspond one to one The known installation positions of the industrial microphones and the signal propagation time difference values ​​of the pairs of inherent mode function components are used to calculate the sound source position corresponding to the adjacent frequency point group using a time difference positioning algorithm.

[0028] In one possible design, for the For each pair of intrinsic mode function components in the eigenmode function components, the corresponding signal propagation time difference value is calculated according to the corresponding two eigenmode function components, including:

[0029] For the a pair of intrinsic modal function components among the eigenmodal function components, and searching, according to the corresponding two eigenmodal function components, whether there is at least one adjacent peak / trough moment group, wherein the adjacent peak / trough moment group includes two peak / trough moments corresponding to the two eigenmodal function components one by one, and the time difference between the two peak / trough moments is less than or equal to a preset time threshold;

[0030] If so, for each adjacent peak / trough moment group in the at least one adjacent peak / trough moment group, a corresponding time difference is calculated according to the corresponding two peak / trough moments;

[0031] The average value of the time difference of each adjacent peak / trough moment group is calculated to obtain the signal propagation time difference value corresponding to the certain pair of intrinsic mode function components.

[0032] In a possible design, it further includes a posture sensor installed on the hub of the target offshore wind turbine;

[0033] The posture sensor is communicatively connected to the voiceprint feature extraction unit, and is used to collect the posture data of the wheel hub in real time, and transmit the posture data to the voiceprint feature extraction unit in real time;

[0034] Respectively The de-noised sound signal is processed for voiceprint feature extraction to obtain voiceprint feature data, including:

[0035] Determining the real-time posture of the wheel hub according to the posture data;

[0036] Determining the real-time position of the wind turbine blade according to the real-time posture and the relative position relationship between the hub and the wind turbine blade of the target offshore wind turbine;

[0037] Determine, according to the real-time position, a time period during which the fan blade passes through the airspace below the hub;

[0038] Respectively for the said The noise-reduced sound signal is processed for voiceprint feature extraction to obtain voiceprint feature data.

[0039] In one possible design, the voiceprint recognition algorithm adopts an artificial intelligence algorithm based on a time-delay neural network.

[0040] In a possible design, the blade fault identification model is pre-trained in the following manner:

[0041] Acquire multiple copies of historical fault audio data from the database, wherein the historical fault audio data contains audio recorded when the fan blade fails. A historical noise-reduced sound signal;

[0042] A plurality of positive sample data corresponding to the plurality of historical fault audio data are obtained in the following manner: Perform voiceprint feature extraction processing on the historical noise-reduced sound signal to obtain historical voiceprint feature data, and use the historical voiceprint feature data as a model input item, and use the value "1" as a model output item, and then use the model input item and the model output item as a positive sample data;

[0043] The plurality of positive sample data are applied to calibrate and verify the artificial intelligence model based on the voiceprint recognition algorithm to obtain the blade fault recognition model.

[0044] In a possible design, a monitoring screen is further included which is respectively communicatively connected to the signal processing unit, the voiceprint feature extraction unit, the blade fault identification unit and / or the blade fault alarm unit, wherein the monitoring screen is used to output and display the wind turbine blade corresponding to the target offshore wind turbine. a noise-reduced sound signal, the voiceprint feature data, the blade fault identification result and / or fault alarm information.

[0045] Beneficial effects of the above scheme:

[0046] (1) The present invention provides a new solution for online monitoring of blades of offshore wind turbines based on sound signals, namely, comprising M industrial microphones, a signal processing unit, a soundprint feature extraction unit, a blade fault identification unit and a blade fault alarm unit, wherein the signal processing unit is used to adopt a joint denoising algorithm based on empirical mode decomposition and wavelet threshold to perform real-time denoising processing on M on-site sound signals corresponding to each other from the M industrial microphones, thereby obtaining M denoised sound signals; the soundprint feature extraction unit is used to perform soundprint feature extraction processing on the M denoised sound signals, thereby obtaining soundprint feature data; the blade fault identification unit is used to import the soundprint feature data into a blade fault identification model that has been pre-trained based on the soundprint recognition algorithm, and output a blade fault identification result. In this way, the blade sweeping sound signal can be accurately detected and identified by combining the joint denoising algorithm and the soundprint recognition algorithm, thereby helping operation and maintenance personnel to quickly discover the fault of the wind turbine blade, thereby improving the timeliness and accuracy of fault monitoring and reducing the potential risks caused by equipment failure;

[0047] (2) The location of the sound source can be determined and based on the distance relationship between the location of the sound source and the active area of ​​the wind turbine blades, it can be determined whether the corresponding sound source is a noise source (such as the sound source of the impact of the waves below the wind turbine and the tower, and the sound source of the whistle of a distant ship, etc.). If so, the wavelet threshold denoising method is used to denoise each intrinsic mode function component in the group to obtain the corresponding denoised intrinsic mode function component, thereby achieving the purpose of targeted denoising;

[0048] (3) The noise-reduced sound signal can also be selectively extracted and processed by combining it with the hub posture data, which can further improve the accuracy of blade fault identification results and the granularity of blade fault identification, facilitating practical application and promotion. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0050] Figure 1 A schematic diagram of the structure of an offshore wind turbine blade monitoring system based on signal processing and voiceprint recognition algorithm provided in an embodiment of the present invention.

[0051] Figure 2 This is an example diagram of the positional relationship between the industrial microphone and attitude sensor provided in an embodiment of the present invention and a target offshore wind turbine.

[0052] Figure 3 A schematic flow chart of a joint denoising algorithm based on empirical mode decomposition and wavelet threshold provided in an embodiment of the present invention.

[0053] Figure 4 An example diagram of the frequency domain window establishment result provided in an embodiment of the present invention.

[0054] Figure 5 A schematic diagram of the process of extracting voiceprint feature data provided by an embodiment of the present invention.

[0055] In the above drawings: 1-industrial microphone; 2-attitude sensor; 100-target offshore wind turbine; 101-tower; 102-wind turbine blades; 103-hub. DETAILED DESCRIPTION

[0056] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in combination with the drawings and the description of the embodiments or the prior art. Obviously, the following description of the structures of the drawings is only some embodiments of the present invention. For ordinary technicians in this field, other embodiments can be obtained based on these embodiments without creative work. It should be noted that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation of the present invention.

[0057] It should be understood that although the terms first and second, etc. may be used herein to describe various objects, these objects should not be limited by these terms. These terms are only used to distinguish one object from another object. For example, a first object can be referred to as a second object, and similarly, a second object can be referred to as a first object without departing from the scope of the exemplary embodiments of the present invention.

[0058] It should be understood that the term "and / or" that may appear in this article is merely a description of the association relationship of associated objects, indicating that there may be three relationships. For example, A and / or B can indicate three situations: A exists alone, B exists alone, or A and B exist at the same time. For another example, A, B and / or C can indicate the existence of any one of A, B and C or any combination of them. The term " / and" that may appear in this article describes another type of association object relationship, indicating that there may be two relationships. For example, A / and B can indicate two situations: A exists alone or A and B exist at the same time. In addition, the character " / " that may appear in this article generally indicates that the previous and next associated objects are in an "or" relationship.

[0059] Example

[0060] like Figures 1 to 5 As shown, the offshore wind turbine blade monitoring system provided in this embodiment and based on signal processing and voiceprint recognition algorithm includes but is not limited to An industrial microphone 1, a signal processing unit, a voiceprint feature extraction unit, a blade fault identification unit and a blade fault alarm unit, etc., among which, represents a positive integer greater than or equal to 3, The industrial microphones 1 are installed at intervals on the bottom outer peripheral surface of the tower 101 of the target offshore wind turbine 100. Since the industrial microphones 1 are installed below the offshore wind turbine, they will not affect the wind turbine, so the early installation and later maintenance are relatively convenient. Figure 2 As shown, the number of the industrial microphones 1 is eight, and they are installed on the bottom outer peripheral surface of the tower 101 at equal intervals in an annular direction around the center of the horizontal cross section of the tower 101 .

[0061] The industrial microphone 1 is communicatively connected to the signal processing unit, and is used to collect the on-site sound signal in real time, and transmit the on-site sound signal to the signal processing unit in real time. The industrial microphone 1 collects the sound of the wind turbine blade in a non-contact manner, and the corresponding product can adopt a windproof sponge cover physical design to effectively reduce wind noise interference. Specifically, the industrial microphone 1 is preferably implemented by an existing product that meets the following standards: (1) meets the requirements of GB / T3785.1-2010 Class 1 sound level meter; (2) has strong applicability and is compatible with all IEPE (Integrated Electronics Piezo-Electric, piezoelectric integrated circuit) acquisition cards; (3) is windproof, dustproof, rainproof and birdproof; (4) has an operating temperature of -40 to 70°, suitable for outdoor and industrial scenes; (5) uses a titanium film microphone, which is corrosion-resistant. The on-site sound signal can be, but is not limited to, the sound of wind from the blades, the sound of waves hitting the tower, and the sound of the whistle of a distant ship. In addition, since the on-site sound signal is transmitted by data transmission, the on-site sound signal and other subsequent signals are all digital signals.

[0062] The signal processing unit is communicatively connected to the voiceprint feature extraction unit, and is used to adopt a joint denoising algorithm based on empirical mode decomposition and wavelet threshold to denoise the signals from the voiceprint feature extraction unit in a one-to-one correspondence. Industrial Pickup 1 The live sound signal is processed in real time to reduce noise, and the The live sound signals correspond one to one a noise-reduced sound signal, and the The de-noised sound signal is transmitted to the voiceprint feature extraction unit in real time. Since the on-site sound signal is mixed with non-blade wind sounds such as the impact of waves and towers and the whistle of distant ships, denoising is required. Empirical Mode Decomposition (EMD) denoising method and wavelet threshold denoising method are currently commonly used denoising methods. Among them, the Empirical Mode Decomposition denoising method is to decompose the noisy signal into empirical mode, calculate the intrinsic mode function (IMF) of each order, and then reconstruct some IMF components to form the effect of high-pass and / or low-pass filters; the wavelet threshold denoising method is a denoising method based on wavelet transform. Its principle is to use a threshold function to process the high-frequency components containing the main noise, and it is widely used to process nonlinear and non-stationary signals. However, the above two methods are very likely to cause the loss of effective signal information, especially when the noise content is too high, it will greatly weaken the weak characteristic components of the sudden change in the signal and increase the measurement error. Therefore, after combining the advantages of the two denoising methods, this embodiment proposes to use a joint denoising algorithm based on empirical mode decomposition and wavelet threshold to denoise the The live sound signal is processed in real time for noise reduction. Figure 3 As shown, a joint denoising algorithm based on empirical mode decomposition and wavelet threshold is used to denoise the noise from the above Industrial Pickup 1 The live sound signal is processed in real time to reduce noise, and the The live sound signals correspond one to one The noise-reduced sound signal includes but is not limited to the following steps S11 to S15.

[0063] S11. For the respective Industrial Pickup 1 Each sound signal in the live sound signal is subjected to empirical mode decomposition processing to obtain the corresponding IMF components, where Represents a positive integer greater than or equal to 3.

[0064] In step S11, since the empirical mode decomposition denoising method assumes that any complex sequence is formed by the superposition of multiple single frequency signals, it can be decomposed into a combination of several intrinsic mode functions IMF. Assume that the on-site sound signal is ( represents the time variable), then The EMD decomposition formula is:

[0065]

[0066] In the formula, Indicates less than or equal to A positive integer, Indicates Intrinsic mode function components (i.e. single frequency signals), In addition, the specific process of the empirical mode decomposition processing is a prior art and will not be described in detail here.

[0067] S12. For each of the sound signals, according to the corresponding Intrinsic mode function components, determine the corresponding component center frequencies, where The center frequency of the components is The intrinsic mode function components correspond one to one.

[0068] In step S12, since the intrinsic mode function component is a single frequency signal, the spectrum of the intrinsic mode function component can be obtained by conventional time domain to frequency domain conversion (such as Fourier transform), and then the center point of the spectrum is selected as the corresponding component center frequency point.

[0069] S13. According to the respective sound signals component center frequency points, and finding whether there is at least one adjacent frequency point group, wherein the adjacent frequency point group contains frequency points corresponding to different sound signals. The component center frequency points, The maximum frequency difference between the center frequencies of the components is less than or equal to the preset frequency threshold. Indicates greater than or equal to 3 and less than or equal to A positive integer.

[0070] In step S13, the technical idea of ​​searching the adjacent frequency point group is that the sound signals coming from the same sound source and arriving at different positions (i.e., the positions of different industrial microphones 1) have the same or similar frequencies, so it can be inferred that different natural mode function components with the same or similar frequencies may come from the same sound source, that is, one adjacent frequency point group can correspond to one sound source by default. Specifically, according to the component center frequency points, and searching whether there is at least one adjacent frequency point group, including but not limited to the following steps S131-S132.

[0071] S131. In the frequency domain, examine the respective sound signals in order from small to large frequency. If it is found that a component center frequency point of a sound signal in each of the sound signals is located at the current frequency point and the current frequency point is not in the established frequency domain window, a new frequency domain window is created with the current frequency point as the starting frequency point and the frequency domain width equal to the preset frequency threshold.

[0072] In step S131, the idea of ​​establishing the frequency domain window is: Among the component center frequency points, in the frequency domain, along the direction from small to large frequency, the first component center frequency point is used as the trigger starting point (i.e. Figure 4 The frequency domain window is delineated backwards from the red point in the figure, and the window width is set (it should be noted that the center frequency of the component in the window can no longer trigger the establishment of a new window, that is, it is necessary to avoid window overlap). Based on the example of 8 sound signals (they correspond to the 8 industrial pickups 1), the establishment results of the frequency domain window are as follows: Figure 4 In addition, the preset frequency threshold can be preset based on historical experience.

[0073] S132. For each of the established frequency domain windows, determine whether the total number of frequency points of multiple component center frequencies corresponding to different sound signals in the corresponding window is greater than or equal to 3. If so, include the multiple component center frequencies into an adjacent frequency point group, wherein the adjacent frequency point group includes the frequency points corresponding to different sound signals. The component center frequency points, The maximum frequency difference between the center frequency points of the components is less than or equal to the preset frequency threshold, Indicates greater than or equal to 3 and less than or equal to A positive integer.

[0074] In step S132, based on the example of eight sound signals (which correspond one to one to eight industrial pickups 1), Figure 4 As shown, an adjacent frequency point group including component center frequency point 1A, component center frequency point 3A, component center frequency point 6A and component center frequency point 8A can be obtained.

[0075] S14. If so, for each adjacent frequency point group in the at least one adjacent frequency point group, according to the corresponding The center frequencies of the components correspond one to one The intrinsic mode function components and the The intrinsic mode function components correspond one to one The known installation position of the industrial microphone 1 is used to determine the corresponding sound source position, and when it is determined that the sound source position is far from the active area of ​​the wind turbine blade 102 of the target offshore wind turbine 100, the sound source position is respectively detected by the target offshore wind turbine 100. The intrinsic mode function components are subjected to wavelet threshold denoising to obtain the same The intrinsic mode function components correspond one to one The denoised intrinsic mode function components.

[0076] In step S14, since one of the adjacent frequency point groups can correspond to a sound source by default, it is possible to determine whether the corresponding sound source is a noise source (such as the sound source of the impact of the waves below the wind turbine and the tower and the sound source of the whistle of a distant ship, etc.) based on the corresponding sound source position and the distance between the sound source position and the active area of ​​the wind turbine blade 102. If so, the wavelet threshold denoising method is used to denoise each inherent modal function component in the group to obtain the corresponding denoised inherent modal function component, thereby achieving the purpose of targeted denoising. Specifically, for each adjacent frequency point group in the at least one adjacent frequency point group, according to the corresponding The center frequencies of the components correspond one to one The intrinsic mode function components and the The intrinsic mode function components correspond one to one The known installation position of an industrial microphone 1 is determined to determine the corresponding sound source position, including but not limited to the following steps S141 to S143.

[0077] S141. For a certain adjacent frequency point group in the at least one adjacent frequency point group, determine the corresponding The center frequencies of the components correspond one to one Intrinsic mode function components.

[0078] S142. For each pair of intrinsic mode function components in the eigenmode function components, the corresponding signal propagation time difference value is calculated according to the corresponding two intrinsic mode function components.

[0079] In the step S142, specifically, for the For each pair of intrinsic mode function components in the eigenmode function components, the corresponding signal propagation time difference value is calculated according to the corresponding two eigenmode function components, including but not limited to the following steps S1421 to S1423.

[0080] S1421. For a pair of intrinsic modal function components in the eigenmodal function components, according to the corresponding two eigenmodal function components, searching whether there is at least one adjacent peak / trough moment group, wherein the adjacent peak / trough moment group includes two peak / trough moments corresponding one-to-one to the two eigenmodal function components, and the time difference between the two peak / trough moments is less than or equal to a preset time threshold.

[0081] In the step S1421, the method for searching the adjacent peak / trough time group can be conventionally derived by referring to the method for searching the adjacent frequency point group, which will not be described in detail herein.

[0082] S1422. If so, for each adjacent peak / trough moment group in the at least one adjacent peak / trough moment group, a corresponding time difference is calculated based on the corresponding two peak / trough moments.

[0083] S1423. Calculate the average value of the time difference of each adjacent peak / trough moment group to obtain the signal propagation time difference value corresponding to the pair of intrinsic mode function components.

[0084] S143. According to the The intrinsic mode function components correspond one to one The known installation positions of the industrial microphones 1 and the signal propagation time difference values ​​of the pairs of inherent mode function components are used to calculate the sound source position corresponding to the adjacent frequency point group using the time difference positioning algorithm.

[0085] In step S143, due to the The intrinsic mode function components correspond to different sound signals, and the different sound signals come from different industrial microphones 1, so it can be determined that the sound signals are related to the industrial microphones 1. The intrinsic mode function components correspond one to one Industrial pickup 1. In addition, due to represents a positive integer greater than or equal to 3, so the sound source position corresponding to the adjacent frequency point group can be calculated based on the existing time difference positioning algorithm.

[0086] In step S14, the specific method of judging whether the sound source position is far away from the active area of ​​the fan blade 102 may be, but is not limited to, conventionally determined based on the comparison result of the closest distance from the sound source position to the active area of ​​the fan blade 102 and a preset distance threshold. For example, when the closest distance is greater than a preset threshold, it is judged that the sound source position is far away from the active area of ​​the fan blade 102. In addition, the specific process of the wavelet threshold denoising process is a prior art and will not be described in detail here.

[0087] S15. For each of the sound signals, reconstruct the corresponding denoised sound signal according to all the corresponding denoised intrinsic mode function components and all the non-denoised intrinsic mode function components.

[0088] The voiceprint feature extraction unit is communicatively connected to the blade fault identification unit, and is used to respectively The noise-reduced sound signal is subjected to voiceprint feature extraction processing to obtain voiceprint feature data, and the voiceprint feature data is transmitted to the blade fault identification unit. The specific process of the above-mentioned voiceprint feature extraction processing is prior art and will not be described in detail here.

[0089] The blade fault identification unit is communicatively connected to the blade fault alarm unit, and is used to import the voiceprint feature data into a blade fault identification model that has been pre-trained based on a voiceprint recognition algorithm, output a blade fault identification result for the wind turbine blade 102 of the target offshore wind turbine 100, and transmit the blade fault identification result to the blade fault alarm unit. The voiceprint recognition algorithm is a deep learning algorithm (specifically machine learning based on deep neural network models and methods) that identifies the characteristics of the sound source by analyzing the voiceprint features; it is developed based on algorithm models such as statistical machine learning and artificial neural networks, combined with the development of contemporary big data and large computing power. The most important technical feature of deep learning is the ability to automatically extract features. The extracted features are also called deep features or deep feature representations. Compared with artificially designed features, deep features have stronger and more robust representation capabilities. Specifically, it can be, but not limited to, artificial intelligence algorithms based on classification and regression tree (abbreviated as CART, which is a learning method for conditional probability distribution of output random variable Y under given input random variable X), perceptron, multi-layer perceptron, recurrent neural network, long short-term memory network, autoencoder, variational autoencoder or time-delay neural network, etc., wherein, since the time-delay neural network is a neural network structure for processing sequence data, it is mainly used in speech recognition, text-independent speaker recognition and other fields, so it can be used preferentially. Specifically, the blade fault recognition model is pre-trained according to the following steps S21 to S23.

[0090] S21. Acquire multiple copies of historical fault audio data from the database, wherein the historical fault audio data contains audio recorded when the fan blade fails. A historical noise-reduced sound signal.

[0091] In the step S21, the The historical noise-reduced sound signal may be from the An industrial pickup 1, or from other offshore wind turbines Other industrial pickups can be obtained by denoising through the corresponding signal processing units and stored in the database.

[0092] S22. Obtain multiple positive sample data corresponding to the multiple copies of historical fault audio data in the following manner: The voiceprint feature extraction process is performed on the historical noise-reduced sound signal to obtain historical voiceprint feature data, and the historical voiceprint feature data is used as the model input item, and the value "1" is used as the model output item, and then the model input item and the model output item are used as a positive sample data.

[0093] S23. Apply the multiple positive sample data to calibrate and verify the artificial intelligence model based on the voiceprint recognition algorithm to obtain the blade fault recognition model.

[0094] The blade fault alarm unit is used to trigger a fault alarm action for the fan blade 102 when it is found that the probability of judging that the fan blade 102 is normal is less than the probability of judging that the fan blade 102 is abnormal according to the blade fault identification result.

[0095] Based on the detailed description of the offshore wind turbine blade monitoring system, it can be seen that the present embodiment provides a new solution for online blade monitoring based on sound signals for offshore wind turbines, namely, it includes M industrial microphones, a signal processing unit, a soundprint feature extraction unit, a blade fault identification unit and a blade fault alarm unit, wherein the signal processing unit is used to adopt a joint denoising algorithm based on empirical mode decomposition and wavelet threshold to perform real-time noise reduction processing on M on-site sound signals from M industrial microphones corresponding to each other, and obtain M noise-reduced sound signals, the soundprint feature extraction unit is used to perform soundprint feature extraction processing on the M noise-reduced sound signals respectively, and obtain soundprint feature data, and the blade fault identification unit is used to import the soundprint feature data into a blade fault identification model that has been pre-trained based on the soundprint recognition algorithm, and output the blade fault identification result, so that the blade sweeping sound signal can be accurately detected and identified through the combination of the joint denoising algorithm and the soundprint recognition algorithm, so as to help the operation and maintenance personnel quickly discover the fault of the wind turbine blade, thereby improving the timeliness and accuracy of fault monitoring, reducing the potential risks brought by equipment failure, and facilitating practical application and promotion.

[0096] Preferably, it also includes a posture sensor 2 installed on the hub 103 of the target offshore wind turbine 100; the posture sensor 2 is communicatively connected to the voiceprint feature extraction unit, and is used to collect the posture data of the hub 103 in real time, and transmit the posture data to the voiceprint feature extraction unit in real time. The posture sensor 2 can be implemented by using existing related devices. In this case, specifically, Figure 5 As shown, respectively The noise-reduced sound signal is subjected to voiceprint feature extraction processing to obtain voiceprint feature data, including but not limited to the following steps S31 to S34.

[0097] S31. Determine the real-time posture of the wheel hub 103 according to the posture data.

[0098] S32 . Determine the real-time position of the wind blade 102 according to the real-time posture and the relative position relationship between the hub 103 and the wind blade 102 of the target offshore wind turbine 100 .

[0099] In step S32 , since the relative position relationship is stable, the real-time position of the wind turbine blade 102 can be determined based on conventional geometric knowledge.

[0100] S33. Determine the time period during which the wind turbine blade 102 passes through the airspace below the hub 103 according to the real-time position.

[0101] S34. respectively for the said The noise-reduced sound signal is processed for voiceprint feature extraction to obtain voiceprint feature data.

[0102] In step S34, since the airspace below the wheel hub 103 is The installation position of the industrial microphone 1 is closest, which will make the wind blade sweeping sound energy in the collected sound signal the largest, that is, it has a stronger anti-interference ability, so the said industrial microphone 1 in the said time period is respectively By performing voiceprint feature extraction processing on the noise-reduced sound signal, more accurate voiceprint feature data can be obtained, which is further conducive to improving the accuracy of blade fault identification results. In addition, since offshore wind turbines generally have three blades, when a blade passes through the airspace below the hub 103, the other two blades are in other airspaces, so the voiceprint feature data can be bound to the blade, and then the blade fault identification result for the blade can be obtained, that is, the blade fault identification granularity can also be improved.

[0103] Preferably, it also includes a monitoring screen which is respectively connected to the signal processing unit, the voiceprint feature extraction unit, the blade fault identification unit and / or the blade fault alarm unit, wherein the monitoring screen is used to output and display the corresponding wind turbine blade 102 of the target offshore wind turbine 100. a noise-reduced sound signal, the voiceprint feature data, the blade fault identification result and / or fault alarm information, etc.

[0104] In summary, the offshore wind turbine blade monitoring system provided by this embodiment has the following technical effects:

[0105] (1) This embodiment provides a new solution for online monitoring of blades of offshore wind turbines based on sound signals, namely, it includes M industrial microphones, a signal processing unit, a soundprint feature extraction unit, a blade fault identification unit and a blade fault alarm unit, wherein the signal processing unit is used to adopt a joint denoising algorithm based on empirical mode decomposition and wavelet threshold to perform real-time denoising processing on M on-site sound signals corresponding to each other from the M industrial microphones, so as to obtain M denoised sound signals, the soundprint feature extraction unit is used to perform soundprint feature extraction processing on the M denoised sound signals, so as to obtain soundprint feature data, and the blade fault identification unit is used to import the soundprint feature data into a blade fault identification model that has been pre-trained based on the soundprint recognition algorithm, and output a blade fault identification result. In this way, the blade sweeping sound signal can be accurately detected and identified by combining the joint denoising algorithm and the soundprint recognition algorithm, so as to help operation and maintenance personnel quickly discover the fault of the wind turbine blade, thereby improving the timeliness and accuracy of fault monitoring and reducing the potential risks caused by equipment failure.

[0106] (2) The location of the sound source can be determined and based on the distance relationship between the location of the sound source and the active area of ​​the wind turbine blades, it can be determined whether the corresponding sound source is a noise source (such as the sound source of the impact of the waves below the wind turbine and the tower, and the sound source of the whistle of a distant ship, etc.). If so, the wavelet threshold denoising method is used to denoise each intrinsic mode function component in the group to obtain the corresponding denoised intrinsic mode function component, thereby achieving the purpose of targeted denoising;

[0107] (3) The noise-reduced sound signal can also be selectively extracted and processed by combining it with the hub posture data, which can further improve the accuracy of blade fault identification results and the granularity of blade fault identification, facilitating practical application and promotion.

[0108] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An offshore wind turbine blade monitoring system based on signal processing and voiceprint recognition algorithm, characterized in that: Included An industrial microphone, a signal processing unit, a voiceprint feature extraction unit, a blade fault identification unit and a blade fault alarm unit, among which: represents a positive integer greater than or equal to 3, Industrial microphones are installed at intervals on the bottom outer peripheral surface of the tower of the target offshore wind turbine; An industrial microphone is connected to a signal processing unit for real-time acquisition of on-site sound signals and transmission of the on-site sound signals to the signal processing unit; The signal processing unit is connected to the voiceprint feature extraction unit for adopting a joint denoising algorithm based on empirical mode decomposition and wavelet threshold to perform one-to-one correspondence on the voiceprint feature extraction unit. Industrial pickups The live sound signal is processed in real time to obtain The live sound signals correspond one to one A noise-reduced sound signal, including: for each of the noise-reduced sound signals from Industrial pickups Each sound signal in the live sound signal is subjected to empirical mode decomposition processing to obtain the corresponding Intrinsic mode function components, where represents a positive integer greater than or equal to 3; for each sound signal, according to the corresponding Intrinsic mode function components, determine the corresponding component center frequencies, where The center frequency of the component The intrinsic mode function components correspond one to one; according to the component center frequency points, and finding whether there is at least one adjacent frequency point group, wherein the adjacent frequency point group contains frequency points corresponding to different sound signals. The center frequency of the components, The maximum frequency difference between the center frequencies of the components is less than or equal to the preset frequency threshold. Indicates greater than or equal to 3 and less than or equal to A positive integer; if it exists, then for each adjacent frequency point group in at least one adjacent frequency point group, according to the corresponding The center frequencies of the components correspond one to one The intrinsic mode function components and The intrinsic mode function components correspond one to one The corresponding sound source position is determined by the known installation position of the industrial microphone, and when the sound source position is determined to be far away from the active area of ​​the wind turbine blades of the target offshore wind turbine, the sound source position is determined by the corresponding industrial microphone. The intrinsic mode function components are subjected to wavelet threshold denoising and the obtained The intrinsic mode function components correspond one to one For each sound signal, the corresponding denoised sound signal is reconstructed according to all corresponding denoised intrinsic modal function components and all non-denoised intrinsic modal function components, and The noise-reduced sound signal is transmitted to the voiceprint feature extraction unit in real time; The voiceprint feature extraction unit is connected to the blade fault identification unit for respectively Performing voiceprint feature extraction processing on the noise-reduced sound signal to obtain voiceprint feature data, and transmitting the voiceprint feature data to the blade fault identification unit; a blade fault identification unit, communicatively connected to the blade fault alarm unit, for importing the voiceprint feature data into a blade fault identification model that has been pre-trained based on the voiceprint recognition algorithm, outputting a blade fault identification result for a wind turbine blade of a target offshore wind turbine, and transmitting the blade fault identification result to the blade fault alarm unit; The blade fault alarm unit is used to trigger a fault alarm action for the fan blade when it is found that the probability of judging that the fan blade is normal is less than the probability of judging that the fan blade is abnormal according to the blade fault identification result.

2. The offshore wind turbine blade monitoring system according to claim 1, characterized in that: The number of the industrial pickups is eight, and they are installed on the bottom outer peripheral surface of the tower at equal intervals in the annular direction around the center of the horizontal cross section of the tower.

3. The offshore wind turbine blade monitoring system according to claim 1, characterized in that: According to each sound signal component center frequency points, and find out whether there is at least one adjacent frequency point group, including: In the frequency domain, we examine the frequency of each sound signal in order from small to large. If it is found that a component center frequency point of a sound signal in each sound signal is located at the current frequency point and the current frequency point is not within the established frequency domain window, a new frequency domain window is created with the current frequency point as the starting frequency point and the frequency domain width equal to the preset frequency threshold; For each established frequency domain window, determine whether the total number of frequency points of multiple component center frequency points corresponding to different sound signals in the corresponding window is greater than or equal to 3. If so, the multiple component center frequency points are included in an adjacent frequency point group, wherein the adjacent frequency point group contains the frequency points corresponding to different sound signals. The center frequency of the components, The maximum frequency difference between the center frequencies of the components is less than or equal to the preset frequency threshold. Indicates greater than or equal to 3 and less than or equal to A positive integer.

4. The offshore wind turbine blade monitoring system according to claim 1, characterized in that: For each adjacent frequency point group in at least one adjacent frequency point group, according to the corresponding The center frequencies of the components correspond one to one The intrinsic mode function components and The intrinsic mode function components correspond one to one The known installation positions of industrial microphones are used to determine the corresponding sound source positions, including: For a certain adjacent frequency point group in at least one adjacent frequency point group, determine The center frequencies of the components correspond one to one IMF components; For For each pair of intrinsic mode function components in the eigenmode function components, the corresponding signal propagation time difference value is calculated according to the corresponding two eigenmode function components; According to The intrinsic mode function components correspond one to one The known installation positions of the industrial microphones and the signal propagation time difference values ​​of each pair of intrinsic mode function components are used to calculate the sound source position corresponding to a certain adjacent frequency point group using the time difference positioning algorithm.

5. The offshore wind turbine blade monitoring system according to claim 4, characterized in that: For For each pair of intrinsic mode function components in the eigenmode function components, the corresponding signal propagation time difference value is calculated according to the corresponding two eigenmode function components, including: For a pair of intrinsic modal function components among the eigenmodal function components, and searching whether there is at least one adjacent peak / trough moment group according to the corresponding two eigenmodal function components, wherein the adjacent peak / trough moment group includes two peak / trough moments corresponding to the two eigenmodal function components one by one, and the time difference between the two peak / trough moments is less than or equal to a preset time threshold; If so, for each adjacent peak / trough moment group in at least one adjacent peak / trough moment group, a corresponding time difference is calculated according to the corresponding two peak / trough moments; The average value of the time difference between each adjacent peak / trough moment group is calculated to obtain the signal propagation time difference value corresponding to a pair of intrinsic mode function components.

6. The offshore wind turbine blade monitoring system according to claim 1, characterized in that: Also included is an attitude sensor mounted on a hub of a target offshore wind turbine; A posture sensor, which is communicatively connected to the voiceprint feature extraction unit, is used to collect the posture data of the wheel hub in real time and transmit the posture data to the voiceprint feature extraction unit in real time; Respectively The de-noised sound signal is processed for voiceprint feature extraction to obtain voiceprint feature data, including: Determine the real-time posture of the wheel hub based on the posture data; Determine the real-time position of the wind turbine blades according to the real-time attitude and the relative position relationship between the hub and the wind turbine blades of the target offshore wind turbine; According to the real-time position, determine the time period when the fan blades pass through the airspace below the hub; For the time periods The noise-reduced sound signal is processed for voiceprint feature extraction to obtain voiceprint feature data.

7. The offshore wind turbine blade monitoring system according to claim 1, characterized in that: The voiceprint recognition algorithm adopts an artificial intelligence algorithm based on a time-delay neural network.

8. The offshore wind turbine blade monitoring system according to claim 1, characterized in that: The blade fault recognition model is pre-trained as follows: Multiple copies of historical fault audio data are obtained from the database, wherein the historical fault audio data contains the audio recorded when the fan blade fails. A historical noise-reduced sound signal; A plurality of positive sample data corresponding to a plurality of historical fault audio data are obtained in the following manner: Perform voiceprint feature extraction on the historical noise-reduced sound signals to obtain historical voiceprint feature data, and use the historical voiceprint feature data as the model input item, and the value "1" as the model output item, and then use the model input item and the model output item as a positive sample data; Using multiple positive sample data, the artificial intelligence model based on the voiceprint recognition algorithm was calibrated and verified to obtain the blade fault recognition model.

9. The offshore wind turbine blade monitoring system according to claim 1, characterized in that: The device also includes a monitoring screen that is respectively connected to the signal processing unit, the voiceprint feature extraction unit, the blade fault identification unit and / or the blade fault alarm unit, wherein the monitoring screen is used to output and display the wind turbine blade corresponding to the target offshore wind turbine. The noise-reduced sound signal, voiceprint feature data, blade fault identification result and / or fault alarm information are collected.

Images