Method for extracting signal segments from an acoustic signal, and fault warning method and device

Abstract

The application discloses an extraction method of a signal segment in an acoustic signal, a fault early warning method and device, and belongs to the technical field of networks. The time domain autocorrelation index and the frequency domain quality index of each signal segment are acquired based on the envelope signal of each signal segment, the frequency domain feature of the envelope signal and the time domain autocorrelation index, and the target signal segment in the acoustic signal is extracted based on the time domain autocorrelation index and the frequency domain quality index of each signal segment. The process comprehensively considers the time domain autocorrelation index and the frequency domain quality index, that is, the signal segment is cross quality detected through the time domain and the frequency domain, the false detection and the missed detection of the signal segment are greatly reduced, the accuracy of the extracted target signal segment is improved, the frequency domain quality index is the frequency domain index under the constraint of the time domain autocorrelation index, false positives caused by random spikes in the frequency domain feature are excluded, the accuracy of the extracted target signal segment is improved, and the accuracy of fault identification based on the target signal segment is higher.

Description

Technical Field

[0001] This disclosure relates to the field of network technology, and in particular to a method for extracting signal segments from acoustic signals, a fault early warning method, and an apparatus. Background Technology

[0002] To detect the condition of wind turbines (i.e., the blades in a wind turbine), acoustic signals are usually obtained by continuously collecting the sound signals generated during the rotation of the blades using acoustic sensors. However, due to wind noise interference in the environment where the wind turbine is located, the quality of the acoustic signals is severely degraded. If acoustic signals containing wind noise interference are directly input into the fault diagnosis model, a large number of false alarms or missed alarms will be generated.

[0003] In related technologies, acoustic signals are divided into multiple signal segments, and the RMS (Root Mean Square) of the signal amplitude at all sampling times in each signal segment is calculated. When the RMS is lower than a preset threshold, the signal segment is determined to be submerged by wind noise. However, wind noise itself may also have a large signal amplitude, and relying solely on RMS can easily lead to misjudgment. Therefore, how to accurately extract signal segments from acoustic signals that are not affected or are less affected by wind noise interference is an important research direction. Summary of the Invention

[0004] This disclosure provides a method for extracting signal segments from acoustic signals, a fault early warning method, and an apparatus, which improves the accuracy of the extracted signal segments.

[0005] According to one aspect of the present disclosure, a method for extracting signal segments from an acoustic signal is provided. The method includes: acquiring multiple signal segments of an acoustic signal from a wind turbine generator; acquiring a time-domain autocorrelation index for each signal segment based on the envelope signal of each signal segment, wherein the envelope signal indicates the curve of amplitude variation over time at each sampling moment in the signal segment, and the time-domain autocorrelation index indicates the periodicity of the envelope signal; acquiring a frequency-domain quality index for each signal segment based on the frequency-domain characteristics of the envelope signal of each signal segment and the time-domain autocorrelation index, wherein the frequency-domain quality index indicates the frequency-domain quality of the signal segment under the time-domain autocorrelation index; and extracting a target signal segment from the acoustic signal based on the time-domain autocorrelation index and the frequency-domain quality index of each signal segment, wherein at least one of the time-domain autocorrelation index and the frequency-domain quality index of the target signal segment is greater than or equal to a corresponding preset threshold.

[0006] According to another aspect of the embodiments of this disclosure, a fault early warning method is provided. The method includes: acquiring multiple signal segments of acoustic signals from a wind turbine generator; acquiring a time-domain autocorrelation index for each signal segment based on the envelope signal of each signal segment, wherein the envelope signal indicates the curve of amplitude variation over time at each sampling moment in the signal segment, and the time-domain autocorrelation index indicates the periodicity of the envelope signal; acquiring a frequency-domain quality index for each signal segment based on the frequency-domain characteristics of the envelope signal of each signal segment and the time-domain autocorrelation index, wherein the frequency-domain quality index indicates the frequency-domain quality of the signal segment under the time-domain autocorrelation index; extracting a target signal segment from the acoustic signals based on the time-domain autocorrelation index and the frequency-domain quality index of each signal segment, wherein at least one of the time-domain autocorrelation index and the frequency-domain quality index of the target signal segment is greater than or equal to a corresponding preset threshold; and performing fault identification based on the target signal segment, and issuing an early warning if the identification result indicates the presence of a fault.

[0007] In some embodiments, obtaining the time-domain autocorrelation index of each signal segment based on the envelope signal of each signal segment includes: obtaining multiple autocorrelation indices between the envelope signal of each signal segment and a plurality of corresponding first lag signals, wherein the plurality of first lag signals refer to the envelope signal that lags behind the envelope signal by different sampling times, and the plurality of autocorrelation indices indicate the similarity between the envelope signal and each of the corresponding first lag signals; and outputting the maximum value among the plurality of autocorrelation indices of each signal segment as the time-domain autocorrelation index of each signal segment.

[0008] In some embodiments, obtaining the frequency domain quality index of each signal segment based on the frequency domain characteristics of the envelope signal of each signal segment and the time domain autocorrelation index includes: obtaining the frequency domain quality index of the first signal segment based on the frequency domain characteristics of the envelope signal of the first signal segment and the time domain autocorrelation index, wherein the time domain autocorrelation index of the first signal segment is greater than or equal to a corresponding preset threshold.

[0009] In some embodiments, obtaining the frequency domain quality index of each signal segment based on the frequency domain features of the envelope signal of each signal segment and the time domain autocorrelation index includes: obtaining a frequency domain fusion quality index of each signal segment based on at least two frequency domain features of the envelope signal of each signal segment, wherein the at least two frequency domain features refer to different frequency domain features obtained based on each envelope signal through different extraction methods, and the frequency domain fusion quality index indicates the frequency domain quality of the signal segment; adjusting the frequency domain fusion quality index based on the time domain autocorrelation index of each signal segment to obtain the frequency domain quality index of each signal segment.

[0010] In some embodiments, the frequency domain features include at least two of spectral kurtosis, spectral entropy, and harmonic regularity factor. The spectral kurtosis indicates the sharpness of the energy distribution of the frequency domain signal corresponding to the envelope signal, the spectral entropy indicates the concentration of the energy distribution of the frequency domain signal corresponding to the envelope signal, and the harmonic regularity factor indicates the periodicity of the frequency domain signal corresponding to the envelope signal.

[0011] In some embodiments, the method further includes: performing a fast Fourier transform on the amplitude at each sampling time of the envelope signal to obtain a frequency domain signal, wherein the frequency domain signal indicates the amplitude intensity of different frequency signals in the envelope signal; acquiring multiple frequency domain autocorrelation indices between the frequency domain signal and multiple second hysteresis signals at different preset fundamental frequencies, wherein the preset fundamental frequency indicates the rotation frequency of the wind turbine, and the multiple second hysteresis signals refer to the frequency domain signals that lag behind the frequency domain signal by different integer multiples of the preset fundamental frequency, wherein the multiple frequency domain autocorrelation indices indicate the similarity between the frequency domain signal and the multiple second hysteresis signals respectively; acquiring the product of the multiple frequency domain autocorrelation indices corresponding to each preset fundamental frequency to obtain the autocorrelation product of each preset fundamental frequency, and outputting the maximum value among the multiple autocorrelation products as the harmonic regularity factor of the envelope signal.

[0012] In some embodiments, extracting the target signal segment from the acoustic signal includes: extracting a signal segment from the acoustic signal that satisfies both a first condition and a second condition as the target signal segment, wherein the first condition is that the time-domain autocorrelation index is greater than or equal to a first preset index, and the second condition is that the frequency-domain quality index is greater than or equal to a second preset index; or, extracting a signal segment from the acoustic signal that satisfies both the first and second conditions as the target signal segment; or, performing a weighted summation of the time-domain autocorrelation index and the frequency-domain quality index of each signal segment to obtain a comprehensive quality index for each signal segment, and extracting a signal segment from the acoustic signal whose comprehensive quality index is greater than or equal to a preset threshold as the target signal segment, wherein the comprehensive quality index indicates the comprehensive quality of the signal segment in the frequency and time domains.

[0013] In some embodiments, the method further includes: obtaining the first type of inter-class variance of the frequency domain quality index of each signal segment; and outputting the frequency domain quality index corresponding to the largest first type of inter-class variance among the first type of inter-class variances of each signal segment as a second preset index.

[0014] In some embodiments, the method further includes: outputting a preset value as the first preset index; or, sorting the time-domain autocorrelation indices of each signal segment in ascending order to obtain an ordered dataset, and outputting the quantile of a preset percentage of the ordered dataset as the first preset index; or, obtaining the second-type variance of the time-domain autocorrelation indices of each signal segment, and outputting the time-domain autocorrelation index corresponding to the largest second-type variance among the second-type variances of each signal segment as the first preset index.

[0015] In some embodiments, the method further includes: squaring the amplitude at all sampling times in each signal segment to obtain a first instantaneous relative power at each sampling time; removing the DC component of the first instantaneous relative power to obtain a second instantaneous relative power; and performing low-pass filtering based on the second instantaneous relative power at each sampling time to obtain the envelope signal of each signal segment.

[0016] In some embodiments, the method further includes: downsampling the envelope signal to obtain a target envelope signal; obtaining the time-domain autocorrelation index of each signal segment based on the envelope signal of each signal segment, including: obtaining the time-domain autocorrelation index of each signal segment based on the target envelope signal of each signal segment; obtaining the frequency-domain quality index of each signal segment based on the frequency-domain features of the envelope signal of each signal segment and the time-domain autocorrelation index, including: obtaining the frequency-domain quality index of each signal segment based on the frequency-domain features of the target envelope signal of each signal segment and the time-domain autocorrelation index.

[0017] In some embodiments, the method further includes performing at least one of the following preprocessing steps on the acoustic signal: removing the DC component of the acoustic signal; and performing bandpass filtering on the acoustic signal, wherein the passband range of the bandpass filter covers the frequency band of the sound signal generated by the blades of the wind turbine.

[0018] In some embodiments, extracting the target signal segment from the acoustic signal includes: extending each second signal segment in the acoustic signal forward and backward by a preset time period to obtain a third signal segment, wherein the second signal segment is a signal segment in which at least one of the time-domain autocorrelation index and the frequency-domain quality index is less than a corresponding preset threshold; and extracting the target signal segments from the acoustic signal other than the third signal segment.

[0019] In some embodiments, the method further includes: obtaining a quality score for each signal segment based on the envelope signal of each signal segment, the quality score indicating the quality of the signal segment; and extracting a target signal segment from the acoustic signal based on the time-domain autocorrelation index and the frequency-domain quality index of each signal segment, comprising: extracting a target signal segment from the acoustic signal based on the time-domain autocorrelation index, the frequency-domain quality index, and the quality score of each signal segment, wherein at least one of the time-domain autocorrelation index, the frequency-domain quality index, and the quality score of the target signal segment is greater than or equal to a corresponding preset threshold.

[0020] According to another aspect of the present disclosure, an apparatus for extracting signal segments from an acoustic signal is provided, the apparatus comprising: The first acquisition module is configured to acquire multiple signal segments of the acoustic signal of the wind turbine. The second acquisition module is configured to acquire the time-domain autocorrelation index of each signal segment based on the envelope signal of each signal segment. The envelope signal indicates the curve of amplitude change with time at each sampling moment in the signal segment, and the time-domain autocorrelation index indicates the periodicity of the envelope signal. The third acquisition module is configured to acquire the frequency domain quality index of each signal segment based on the frequency domain characteristics of the envelope signal of each signal segment and the time domain autocorrelation index. The frequency domain quality index indicates the frequency domain quality of the signal segment under the time domain autocorrelation index. The first extraction module is configured to extract target signal segments from the acoustic signal based on the time-domain autocorrelation index and frequency-domain quality index of each signal segment, wherein at least one of the time-domain autocorrelation index and the frequency-domain quality index of the target signal segment is greater than or equal to a corresponding preset threshold.

[0021] In some embodiments, the second acquisition module is configured to acquire multiple autocorrelation indices between the envelope signal of each signal segment and the corresponding plurality of first hysteresis signals, wherein the plurality of first hysteresis signals refer to the envelope signal that lags behind the envelope signal by different sampling times, and the plurality of autocorrelation indices indicate the degree of similarity between the envelope signal and the corresponding first hysteresis signals; and output the maximum value among the plurality of autocorrelation indices of each signal segment as the time-domain autocorrelation index of each signal segment.

[0022] In some embodiments, the third acquisition module is configured to acquire a frequency domain quality index of the first signal segment based on the frequency domain characteristics of the envelope signal of the first signal segment and the time domain autocorrelation index, wherein the time domain autocorrelation index of the first signal segment is greater than or equal to a corresponding preset threshold.

[0023] In some embodiments, the third acquisition module is configured to acquire a frequency domain fusion quality index for each signal segment based on at least two frequency domain features of the envelope signal of each signal segment. The at least two frequency domain features refer to different frequency domain features obtained by different extraction methods based on each envelope signal. The frequency domain fusion quality index indicates the frequency domain quality of the signal segment. The frequency domain fusion quality index is adjusted based on the time domain autocorrelation index of each signal segment to obtain the frequency domain quality index of each signal segment.

[0024] In some embodiments, the frequency domain features include at least two of spectral kurtosis, spectral entropy, and harmonic regularity factor. The spectral kurtosis indicates the sharpness of the energy distribution of the frequency domain signal corresponding to the envelope signal, the spectral entropy indicates the concentration of the energy distribution of the frequency domain signal corresponding to the envelope signal, and the harmonic regularity factor indicates the periodicity of the frequency domain signal corresponding to the envelope signal.

[0025] In some embodiments, the apparatus further includes a seventh acquisition module: the seventh acquisition module is configured to perform a fast Fourier transform on the amplitude at each sampling time in the envelope signal to obtain a frequency domain signal, the frequency domain signal indicating the amplitude intensity of different frequency signals in the envelope signal; at different preset fundamental frequencies, acquire multiple frequency domain autocorrelation indices between the frequency domain signal and multiple second hysteresis signals, the preset fundamental frequency indicating the rotation frequency of the wind turbine, the multiple second hysteresis signals referring to the frequency domain signals that lag behind the frequency domain signal by different integer multiples of the preset fundamental frequency, the multiple frequency domain autocorrelation indices indicating the similarity between the frequency domain signal and the multiple second hysteresis signals respectively; acquire the product of the multiple frequency domain autocorrelation indices corresponding to each preset fundamental frequency to obtain the autocorrelation product of each preset fundamental frequency, and output the maximum value among the multiple autocorrelation products as the harmonic regularity factor of the envelope signal.

[0026] In some embodiments, the first extraction module is configured to extract signal segments from the acoustic signal that satisfy both a first condition and a second condition as the target signal segment, wherein the first condition is that the time-domain autocorrelation index is greater than or equal to a first preset index, and the second condition is that the frequency-domain quality index is greater than or equal to a second preset index; or, to extract signal segments from the acoustic signal that satisfy both the first and second conditions as the target signal segment; or, to perform a weighted summation of the time-domain autocorrelation index and the frequency-domain quality index of each signal segment to obtain a comprehensive quality index of each signal segment, and to extract signal segments from the acoustic signal whose comprehensive quality index is greater than or equal to a preset threshold as the target signal segment, wherein the comprehensive quality index indicates the comprehensive quality of the signal segment in the frequency and time domains.

[0027] In some embodiments, the apparatus further includes an eighth acquisition module: the eighth acquisition module is configured to acquire the first type variance of the frequency domain quality index of each signal segment; and output the frequency domain quality index corresponding to the largest first type variance among the first type variances of each signal segment as a second preset index.

[0028] In some embodiments, the device further includes a ninth acquisition module: the ninth acquisition module is configured to output a preset value as the first preset index; or, to sort the time-domain autocorrelation indices of each signal segment in ascending order to obtain an ordered dataset, and output the quantile of a preset percentage of the ordered dataset as the first preset index; or, to acquire the second-type variance of the time-domain autocorrelation indices of each signal segment, and output the time-domain autocorrelation index corresponding to the largest second-type variance among the second-type variances of each signal segment as the first preset index.

[0029] In some embodiments, the apparatus further includes a tenth acquisition module: the tenth acquisition module is configured to square the amplitude of all sampling times in each signal segment to obtain a first instantaneous relative power quantity at each sampling time; remove the DC component of the first instantaneous relative power quantity to obtain a second instantaneous relative power quantity; and perform low-pass filtering based on the second instantaneous relative power quantity at each sampling time to obtain the envelope signal of each signal segment.

[0030] In some embodiments, the apparatus further includes a first downsampling module: the first downsampling module is configured to downsample the envelope signal to obtain a target envelope signal; the step of obtaining the time-domain autocorrelation index of each signal segment based on the envelope signal of each signal segment includes: obtaining the time-domain autocorrelation index of each signal segment based on the target envelope signal of each signal segment; the step of obtaining the frequency-domain quality index of each signal segment based on the frequency-domain features of the envelope signal of each signal segment and the time-domain autocorrelation index includes: obtaining the frequency-domain quality index of each signal segment based on the frequency-domain features of the target envelope signal of each signal segment and the time-domain autocorrelation index.

[0031] In some embodiments, the apparatus further includes a first preprocessing module: the first preprocessing module is configured to perform at least one of the following preprocessing steps on the acoustic signal: removing the DC component of the acoustic signal; and performing bandpass filtering on the acoustic signal, the passband of the bandpass filtering covering the frequency band of the sound signal generated by the blades of the wind turbine.

[0032] In some embodiments, the first extraction module is configured to extend each second signal segment in the acoustic signal forward and backward by a preset time period to obtain a third signal segment, wherein the second signal segment is a signal segment in which at least one of the time-domain autocorrelation index and the frequency-domain quality index is less than a corresponding preset threshold; and to extract target signal segments in the acoustic signal other than the third signal segment.

[0033] In some embodiments, the apparatus further includes a first value acquisition module; the first value acquisition module is configured to acquire a quality score for each signal segment based on the envelope signal of each signal segment, the quality score indicating the quality of the signal segment; the extraction of a target signal segment from the acoustic signal based on the time-domain autocorrelation index and the frequency-domain quality index of each signal segment includes: extracting a target signal segment from the acoustic signal based on the time-domain autocorrelation index, the frequency-domain quality index, and the quality score of each signal segment, wherein at least one of the time-domain autocorrelation index, the frequency-domain quality index, and the quality score of the target signal segment is greater than or equal to a corresponding preset threshold.

[0034] According to another aspect of the present disclosure, a fault early warning device is provided, the device comprising: The fourth acquisition module is configured to acquire multiple signal segments of the acoustic signal of the wind turbine. The fifth acquisition module is configured to acquire the time-domain autocorrelation index of each signal segment based on the envelope signal of each signal segment. The envelope signal indicates the curve of amplitude change with time at each sampling moment in the signal segment, and the time-domain autocorrelation index indicates the periodicity of the envelope signal. The sixth acquisition module is configured to acquire the frequency domain quality index of each signal segment based on the frequency domain characteristics of the envelope signal of each signal segment and the time domain autocorrelation index. The frequency domain quality index indicates the frequency domain quality of the signal segment under the time domain autocorrelation index. The second extraction module is configured to extract target signal segments from the acoustic signal based on the time-domain autocorrelation index and the frequency-domain quality index of each signal segment, wherein at least one of the time-domain autocorrelation index and the frequency-domain quality index of the target signal segment is greater than or equal to a corresponding preset threshold. The identification module is configured to identify faults based on the target signal segment, and to issue an early warning if the identification result indicates that a fault exists.

[0035] In some embodiments, the fifth acquisition module is configured to acquire multiple autocorrelation indices between the envelope signal of each signal segment and the corresponding multiple first hysteresis signals, wherein the multiple first hysteresis signals refer to the envelope signals that lag behind the envelope signals at different sampling times, and the multiple autocorrelation indices indicate the similarity between the envelope signals and the corresponding first hysteresis signals; and output the maximum value among the multiple autocorrelation indices of each signal segment as the time-domain autocorrelation index of each signal segment.

[0036] In some embodiments, the sixth acquisition module is configured to acquire a frequency domain quality index of the first signal segment based on the frequency domain characteristics of the envelope signal of the first signal segment and the time domain autocorrelation index, wherein the time domain autocorrelation index of the first signal segment is greater than or equal to a corresponding preset threshold.

[0037] In some embodiments, the sixth acquisition module is configured to acquire a frequency domain fusion quality index for each signal segment based on at least two frequency domain features of the envelope signal of each signal segment. The at least two frequency domain features refer to different frequency domain features obtained by different extraction methods based on each envelope signal. The frequency domain fusion quality index indicates the frequency domain quality of the signal segment. The frequency domain fusion quality index is adjusted based on the time domain autocorrelation index of each signal segment to obtain the frequency domain quality index of each signal segment.

[0038] In some embodiments, the frequency domain features include at least two of spectral kurtosis, spectral entropy, and harmonic regularity factor. The spectral kurtosis indicates the sharpness of the energy distribution of the frequency domain signal corresponding to the envelope signal, the spectral entropy indicates the concentration of the energy distribution of the frequency domain signal corresponding to the envelope signal, and the harmonic regularity factor indicates the periodicity of the frequency domain signal corresponding to the envelope signal.

[0039] In some embodiments, the device further includes an eleventh acquisition module: the eleventh acquisition module is configured to perform a fast Fourier transform on the amplitude at each sampling time in the envelope signal to obtain a frequency domain signal, the frequency domain signal indicating the amplitude intensity of different frequency signals in the envelope signal; at different preset fundamental frequencies, acquire multiple frequency domain autocorrelation indices between the frequency domain signal and multiple second hysteresis signals, the preset fundamental frequency indicating the rotation frequency of the wind turbine, the multiple second hysteresis signals referring to the frequency domain signals that lag behind the frequency domain signal by different integer multiples of the preset fundamental frequency, the multiple frequency domain autocorrelation indices indicating the similarity between the frequency domain signal and the multiple second hysteresis signals respectively; acquire the product between the multiple frequency domain autocorrelation indices corresponding to each preset fundamental frequency to obtain the autocorrelation product of each preset fundamental frequency, and output the maximum value among the multiple autocorrelation products as the harmonic regularity factor of the envelope signal.

[0040] In some embodiments, the second extraction module is configured to extract signal segments from the acoustic signal that satisfy both a first condition and a second condition as the target signal segment, wherein the first condition is that the time-domain autocorrelation index is greater than or equal to a first preset index, and the second condition is that the frequency-domain quality index is greater than or equal to a second preset index; or, to extract signal segments from the acoustic signal that satisfy both the first and second conditions as the target signal segment; or, to perform a weighted summation of the time-domain autocorrelation index and the frequency-domain quality index of each signal segment to obtain a comprehensive quality index of each signal segment, and to extract signal segments from the acoustic signal whose comprehensive quality index is greater than or equal to a preset threshold as the target signal segment, wherein the comprehensive quality index indicates the comprehensive quality of the signal segment in the frequency and time domains.

[0041] In some embodiments, the apparatus further includes a twelfth acquisition module: the twelfth acquisition module is configured to acquire the first type variance of the frequency domain quality index of each signal segment; and output the frequency domain quality index corresponding to the largest first type variance among the first type variances of each signal segment as a second preset index.

[0042] In some embodiments, the device further includes a thirteenth acquisition module: the thirteenth acquisition module is configured to output a preset value as the first preset index; or, to sort the time-domain autocorrelation indices of each signal segment in ascending order to obtain an ordered dataset, and output the quantile of a preset percentage of the ordered dataset as the first preset index; or, to acquire the second-type variance of the time-domain autocorrelation indices of each signal segment, and output the time-domain autocorrelation index corresponding to the largest second-type variance among the second-type variances of each signal segment as the first preset index.

[0043] In some embodiments, the apparatus further includes a fourteenth acquisition module: the fourteenth acquisition module is configured to square the amplitude of all sampling times in each signal segment to obtain a first instantaneous relative power quantity at each sampling time; remove the DC component of the first instantaneous relative power quantity to obtain a second instantaneous relative power quantity; and perform low-pass filtering based on the second instantaneous relative power quantity at each sampling time to obtain the envelope signal of each signal segment.

[0044] In some embodiments, the apparatus further includes a second downsampling module: the second downsampling module is configured to downsample the envelope signal to obtain a target envelope signal; the step of obtaining the time-domain autocorrelation index of each signal segment based on the envelope signal of each signal segment includes: obtaining the time-domain autocorrelation index of each signal segment based on the target envelope signal of each signal segment; the step of obtaining the frequency-domain quality index of each signal segment based on the frequency-domain features of the envelope signal of each signal segment and the time-domain autocorrelation index includes: obtaining the frequency-domain quality index of each signal segment based on the frequency-domain features of the target envelope signal of each signal segment and the time-domain autocorrelation index.

[0045] In some embodiments, the apparatus further includes a second preprocessing module: the second preprocessing module is configured to perform at least one of the following preprocessing steps on the acoustic signal: removing the DC component of the acoustic signal; and performing bandpass filtering on the acoustic signal, the passband range of the bandpass filtering covering the frequency band of the sound signal generated by the blades of the wind turbine.

[0046] In some embodiments, the second extraction module is configured to extend each second signal segment in the acoustic signal forward and backward by a preset time period to obtain a third signal segment, wherein the second signal segment is a signal segment in which at least one of the time-domain autocorrelation index and the frequency-domain quality index is less than a corresponding preset threshold; and to extract target signal segments in the acoustic signal other than the third signal segment.

[0047] In some embodiments, the apparatus further includes a second value acquisition module; the second value acquisition module is configured to acquire a quality score for each signal segment based on the envelope signal of each signal segment, the quality score indicating the quality of the signal segment; the extraction of a target signal segment from the acoustic signal based on the time-domain autocorrelation index and the frequency-domain quality index of each signal segment includes: extracting a target signal segment from the acoustic signal based on the time-domain autocorrelation index, the frequency-domain quality index, and the quality score of each signal segment, wherein at least one of the time-domain autocorrelation index, the frequency-domain quality index, and the quality score of the target signal segment is greater than or equal to a corresponding preset threshold.

[0048] According to another aspect of the present disclosure, an electronic device is provided, the electronic device comprising: one or more processors; a memory for storing processor-executable program code; wherein the processor is configured to execute the program code to implement the method for extracting signal segments from acoustic signals as described above or the fault warning method described in any of the preceding claims.

[0049] According to another aspect of the present disclosure, a computer-readable storage medium is provided that, when the instructions in the computer-readable storage medium are executed by a processor of an electronic device, enables the electronic device to perform a method for extracting signal segments from an acoustic signal as described in any of the preceding claims or a fault warning method as described in any of the preceding claims.

[0050] According to another aspect of the present disclosure, a computer program product is provided, including a computer program / instructions that, when executed by a processor, implement the method for extracting signal segments from acoustic signals as described above or the fault warning method described in any of the above.

[0051] In the technical solution provided by this disclosure, the time-domain autocorrelation index and frequency-domain quality index of each signal segment are obtained based on the envelope signal, frequency domain features, and time-domain autocorrelation index of each signal segment. Based on these indices, target signal segments in the acoustic signal are extracted. This process comprehensively considers both time-domain autocorrelation and frequency-domain quality indices, meaning that cross-quality detection of signal segments is performed in both the time and frequency domains. This significantly reduces false positives and false negatives, improving the accuracy of the extracted target signal segments. Furthermore, the frequency-domain quality index is a frequency-domain index constrained by the time-domain autocorrelation index, eliminating false positives caused by random spikes in the frequency domain features, further improving the accuracy of the extracted target signal segments. Therefore, when fault identification is performed based on the target signal segments, since these segments are less affected by wind noise and other noise, signal segments from non-wind turbine sources in the acoustic signal are excluded at the source, significantly improving the accuracy of fault identification, reducing false alarms and false negatives, and increasing the reliability of the early warning.

[0052] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0053] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure, and are not intended to unduly limit this disclosure.

[0054] Figure 1 This is a schematic diagram illustrating an implementation environment according to an exemplary embodiment.

[0055] Figure 2 This is a flowchart illustrating a method for extracting signal segments from an acoustic signal according to an exemplary embodiment.

[0056] Figure 3This is a flowchart illustrating another method for extracting signal segments from an acoustic signal according to an exemplary embodiment.

[0057] Figure 4 This is a schematic diagram of a first sample according to an exemplary embodiment.

[0058] Figure 5 This is a schematic diagram of a second sample according to an exemplary embodiment.

[0059] Figure 6 This is a schematic diagram of a signal segment of window 7 in a first sample, according to an exemplary embodiment.

[0060] Figure 7 This is a schematic diagram of the autocorrelation index curves of multiple signal segments of a first sample according to an exemplary embodiment.

[0061] Figure 8 This is a schematic diagram of the autocorrelation index curves of multiple signal segments of a second sample, according to an exemplary embodiment.

[0062] Figure 9 This is a window-by-window numerical diagram of the six dimensions of a first sample, according to an exemplary embodiment.

[0063] Figure 10 This is a window-by-window numerical diagram of the six dimensions of a second sample, according to an exemplary embodiment.

[0064] Figure 11 This is a frequency domain signal diagram of window 8 of the first sample, as illustrated in an exemplary embodiment.

[0065] Figure 12 This is a frequency domain signal diagram of window 2 of the second sample, as illustrated in an exemplary embodiment.

[0066] Figure 13 This is a schematic diagram of the determination label of a first sample according to an exemplary embodiment.

[0067] Figure 14 This is a schematic diagram of the determination label of a second sample according to an exemplary embodiment.

[0068] Figure 15 The above is a power spectral density diagram of a first sample and a second sample, as shown in an exemplary embodiment.

[0069] Figure 16 The first and second samples are shown according to an exemplary embodiment. and Comparison chart.

[0070] Figure 17The first and second samples shown according to an exemplary embodiment are based on and A comparison chart of the selected signal segments.

[0071] Figure 18 This is a flowchart illustrating a fault early warning method according to an exemplary embodiment.

[0072] Figure 19 This is a schematic diagram of a device for extracting signal segments from an acoustic signal according to an exemplary embodiment.

[0073] Figure 20 This is a schematic diagram of a fault warning device according to an exemplary embodiment.

[0074] Figure 21 This is a block diagram illustrating a terminal according to an exemplary embodiment.

[0075] Figure 22 This is a block diagram illustrating a server according to an exemplary embodiment. Detailed Implementation

[0076] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0077] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0078] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, data stored, data displayed, etc.) and signals involved in this disclosure are all authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0079] Figure 1 This is a schematic diagram illustrating an implementation environment according to an exemplary embodiment. Taking an electronic device provided as a terminal as an example, see [link to example]. Figure 1The implementation environment specifically includes: terminal 101 and server 102. Terminal 101 and server 102 can be connected directly or indirectly through wired or wireless communication, which is not limited herein.

[0080] Terminal 101 is at least one of the following devices: smartphone, smartwatch, desktop computer, laptop, and mobile computer. Terminal 101 runs an application that supports methods for extracting signal segments from acoustic signals or for fault warning. Users can log in to this application through Terminal 101 to access the services provided by the application. For example, users can use the application on Terminal 101 to extract signal segments from acoustic signals and perform fault warnings.

[0081] Terminal 101 generally refers to one of a plurality of terminals; this embodiment uses terminal 101 as an example. Those skilled in the art will understand that the number of terminals can be more or less. For example, there may be several terminals, or dozens or hundreds of terminals, or even more. This disclosure does not limit the number of terminals or the type of device.

[0082] Server 102 can be at least one of a single server, multiple servers, a cloud computing platform, and a virtualization center. Server 102 provides background services for extracting signal segments from acoustic signals or for fault early warning. Server 102 can connect to terminal 101 via a wireless or wired network. Server 102 can provide database services to the terminal, and of course, it can also provide other types of services, such as data preprocessing. In some embodiments, the number of servers may be more or fewer, and this disclosure does not limit this. Of course, server 102 may also include other functional servers to provide more comprehensive and diversified services.

[0083] Among related technologies, wind power, as a clean and renewable energy source, occupies an increasingly important position in the global energy structure. The blades of wind turbines (i.e., wind generators or wind turbines) are key components of the unit, and their structural integrity directly affects power generation efficiency and operational safety. Statistics show that blade failures account for 34% of total wind turbine downtime, and their maintenance costs account for 30% of the total lifecycle cost. Therefore, real-time online monitoring of the operating status of wind turbine blades to promptly detect early damage and performance degradation has significant engineering value and economic implications.

[0084] Acoustic signal-based wind turbine blade condition monitoring technology has received widespread attention in recent years due to its advantages such as non-contact operation, low cost, and easy installation. This technology continuously collects acoustic signals generated during blade rotation by installing acoustic sensors (such as microphones or microphone arrays) on the wind turbine tower or nacelle, extracting characteristic parameters representing the blade's health status to achieve fault detection and diagnosis. In practical applications, the acoustic signal energy of the passing blade is mainly concentrated in the high-frequency band of 8–18 kHz, and its envelope exhibits quasi-periodic fluctuations related to the blade rotation frequency (typically 0.1–2 Hz).

[0085] However, wind noise interference in wind farm environments is a major challenge affecting the effectiveness of acoustic monitoring. Strong wind noise disrupts the time-domain periodicity of the blade acoustic signal envelope and disturbs the harmonic structure of the envelope spectrum, leading to a severe deterioration in signal quality. If low-quality signals are directly input into subsequent fault diagnosis models without screening, a large number of false alarms and false negatives will be generated, severely reducing the reliability of the monitoring system.

[0086] To assess the quality of acoustic signals acquired by acoustic sensors during blade rotation and extract signal segments that are unaffected or minimally affected by wind noise, the relevant techniques mainly fall into the following categories: Category 1: Threshold method based on signal amplitude or signal energy This type of method directly calculates the RMS or peak value of each signal segment of the acoustic signal. When the RMS is lower than a preset threshold, the signal segment is determined to be submerged in noise and is discarded. This method is simple to implement and has a fast calculation speed, but it has the following problems: wind noise itself may also have a large signal amplitude, and it is impossible to distinguish between "useful blade sound signals with high sound pressure levels" and "wind noise interference signals with high sound pressure levels" based solely on energy level, which can easily lead to misjudgment, especially under high wind noise conditions.

[0087] The second category: methods based on SNR (Signal-to-Noise Ratio) estimation. Such methods require first estimating the "clean signal" and "noise" components in the acoustic signal, and then calculating their power ratio to obtain the signal-to-noise ratio (SNR). Typical methods for estimating the "clean signal" and "noise" components in an acoustic signal include spectral subtraction to estimate the noise floor and noise estimation based on minimum statistics. However, in the acoustic monitoring scenario of wind turbine blades, the blade signal itself has time-varying and non-stationary characteristics, and there is a lack of clear statistical separability between the "clean signal" and "noise." The accuracy of SNR estimation is difficult to guarantee. That is, in the non-stationary wind turbine blade acoustic signal scenario, the statistical assumptions of the "clean signal" and "noise" deviate significantly from the actual operating conditions, making SNR estimation unreliable.

[0088] The third category: methods based on single frequency domain features These methods rely solely on a single frequency domain feature of the acoustic signal's envelope (such as spectral centroid, spectral flatness, and spectral entropy) to determine the quality of the acoustic signal. However, the dimensionality of a single frequency domain feature is limited, failing to comprehensively characterize the multidimensional attributes of acoustic signal quality. For example, spectral entropy alone cannot distinguish between "blade acoustic signals with clear harmonic structures" and "noise signals containing isolated single-frequency interference." In other words, using only a single statistical quantity (such as spectral entropy) cannot effectively differentiate between "signals with clear harmonic structures" and "noise containing isolated single-frequency interference." This lack of dimensionality results in limited accuracy.

[0089] The fourth category: SQI (Signal Quality Index) method SQI (Signal Quality Indicator) originated from biomedical time-series signals such as ECG (Electrocardiogram) and PPG (Photoplethysmography). It uses a 0-1 quantized score obtained by normalizing a single frequency domain feature to score a signal segment, intuitively judging the cleanliness / noise contamination level of the signal segment. However, these SQI methods do not integrate time-domain periodic information, lack a regular quantitative assessment of the envelope signal of the acoustic signal, and rely on manually set fixed thresholds. When the detection environment changes or the audio signal-to-noise ratio fluctuates, the fixed thresholds are difficult to adapt to, requiring repeated manual parameter tuning, and have limited adaptability to different signal-to-noise ratio environments.

[0090] In summary, the methods used in the field of acoustic signal quality assessment for wind turbine blades share the following common problems: either they have a single feature dimension and insufficient discrimination ability, or they rely on manual thresholds and lack adaptability, or they cannot effectively distinguish between signals with similar amplitude characteristics and interference. To address these problems, this disclosure proposes a method for extracting signal segments from acoustic signals.

[0091] Figure 2 This is a flowchart illustrating a method for extracting signal segments from an acoustic signal according to an exemplary embodiment. See also... Figure 2 In this embodiment, the method is executed by an electronic device, and the method includes the following steps: In step 201, multiple signal segments of the acoustic signal of the wind turbine are acquired.

[0092] In this step, the acoustic signal is acquired through an acoustic sensor. The acoustic sensor is used to continuously collect the acoustic signals generated during the blade rotation to obtain the acoustic signal, which includes the amplitude at each sampling time (or sampling point). A signal segment refers to a segment with a fixed time length obtained by continuously framing the acoustic signal without overlap. This avoids the repeated evaluation of signal segments with the same operating condition and the same rotation cycle in subsequent iterations. For example, for an acoustic signal with a time length of 2 minutes, if the signal segment time length is 10 seconds, 12 signal segments can be obtained, with the corresponding time lengths of 0-10s, 10s-20s, 20s-30s...110s-120s, respectively. In order to fully capture the periodic information of each signal segment, the duration of the signal segment must be no less than the rotation period of the fan. For example, if the fan's rotation frequency is 0.1-2Hz, which means the fan's rotation period is 0.5-10s, then the duration of the signal segment can be 10s (seconds) or greater than 10s, such as 11s.

[0093] In step 202, based on the envelope signal of each signal segment, the time-domain autocorrelation index of each signal segment is obtained. The envelope signal indicates the curve of amplitude change with time at each sampling moment in the signal segment, and the time-domain autocorrelation index indicates the periodicity of the envelope signal.

[0094] In this step, the envelope signal refers to the contour line formed by connecting the peaks of the amplitude at each sampling moment in the signal segment. In other words, the envelope signal is the curve showing how the amplitude changes over time at each sampling moment in the signal segment. Correspondingly, the waveform of the envelope signal changes slowly, primarily reflecting the rotational frequency of the wind turbine blades (0.1~2Hz). The time-domain autocorrelation index is used to reflect the strength of the periodicity of the envelope signal; that is, the larger the time-domain autocorrelation index, the stronger the periodicity of the envelope signal, and the smaller the time-domain autocorrelation index, the weaker the periodicity of the envelope signal. Typically, wind noise is random and aperiodic. When there is little or no wind noise, the signal of blade rotation in the signal segment is relatively clear, the periodicity of the envelope signal is good, and the time-domain autocorrelation index is large. However, when the wind noise is large or even strong enough to drown out the blade sound, wind noise dominates, the periodicity of the envelope signal is disrupted, and the time-domain autocorrelation index is small.

[0095] In step 203, based on the frequency domain characteristics and time domain autocorrelation index of the envelope signal of each signal segment, the frequency domain quality index of each signal segment is obtained. The frequency domain quality index indicates the frequency domain quality of the signal segment under the time domain autocorrelation index.

[0096] In this step, frequency domain characteristics refer to the amplitude distribution pattern on the frequency axis after the envelope signal is converted into frequency-amplitude (or energy) by Fourier transform, such as spectral centroid, spectral kurtosis, spectral entropy, etc. Frequency domain quality indicators are obtained based on the frequency domain characteristics and time-domain autocorrelation index of the signal segment's envelope signal. The frequency domain quality index is obtained by adjusting the value of the frequency domain characteristics using the time-domain autocorrelation index, and it accurately reflects the frequency domain quality of the signal segment.

[0097] In step 204, target signal segments are extracted from the acoustic signal based on the time-domain autocorrelation index and frequency-domain quality index of each signal segment. At least one of the time-domain autocorrelation index and frequency-domain quality index of the target signal segment is greater than or equal to the corresponding preset threshold.

[0098] In this step, the target signal segment refers to a segment of the acoustic signal in which at least one of the time-domain autocorrelation index and the frequency-domain quality index is greater than or equal to the corresponding preset threshold. For example, the time-domain autocorrelation index of the target signal segment is greater than the corresponding threshold and the frequency-domain quality index is greater than the corresponding threshold.

[0099] The technical solution provided in this disclosure obtains the time-domain autocorrelation index and frequency-domain quality index of each signal segment based on the envelope signal, frequency domain features, and time-domain autocorrelation index of each signal segment. Based on the time-domain autocorrelation index and frequency-domain quality index of each signal segment, the target signal segment in the acoustic signal is extracted. This process comprehensively considers the time-domain autocorrelation index and the frequency-domain quality index, that is, it performs cross-quality detection of the signal segments in the time and frequency domains, which greatly reduces the false detection and false negative detection of the signal segments and improves the accuracy of the extracted target signal segments. Furthermore, the frequency-domain quality index is a frequency-domain index constrained by the time-domain autocorrelation index, which eliminates false positives caused by random spikes in the frequency domain features, further improving the accuracy of the extracted target signal segments.

[0100] Figure 3 This is a flowchart illustrating another method for extracting signal segments from an acoustic signal according to an exemplary embodiment, see [link to flowchart]. Figure 3 In this embodiment, the method is executed by an electronic device, and the method includes the following steps: In step 301, the acoustic signal of the wind turbine is acquired, and at least one of the following preprocessing steps is performed on the acoustic signal: removing the DC component of the acoustic signal; and performing bandpass filtering on the acoustic signal, wherein the passband range of the bandpass filter covers the frequency band of the sound signal generated by the blades of the wind turbine.

[0101] In this step, the principle of the acoustic signal is the same as in the above embodiment, and will not be repeated here. The DC component is a constant bias in the acoustic signal that does not change with time. It mainly comes from the zero-point drift and other biases in the acquisition circuit of the sound sensor. It does not belong to the real acoustic vibration signal. It will cause the baseline of the envelope signal of the acoustic signal to shift, interfering with the subsequent extraction of the envelope signal. By removing the DC component, this part of the error can be eliminated from the source. It can also eliminate the strong peak of the frequency domain signal of the acoustic signal at 0Hz, preventing it from spreading to adjacent low-frequency bands through tailing, thereby masking the real low-frequency signal. Bandpass filtering only retains the acoustic signal in the frequency band specified by the passband range, filtering out all signals below the lower limit of the specified frequency band and above the upper limit of the specified frequency band. This achieves the filtering out of interference signals other than the rotation of the fan blades, improving the signal-to-noise ratio of the acoustic signal.

[0102] In some embodiments, removing the DC component of the acoustic signal includes subtracting the average amplitude of all sampling times in the acoustic signal from the amplitude of each sampling time in the acoustic signal.

[0103] For example, the acoustic signal after removing the DC component is calculated using the following formula. : Formula 1:

[0104] in, is the amplitude at sampling time t in the acoustic signal; N is the total number of sampling times in the acoustic signal; Let n be the amplitude at the nth sampling time, where n takes the value [1, N]. This is the average amplitude from the first sampling time to the Nth sampling time. For example, the sampling rate of the acoustic signal... Signal duration , This corresponds to 5.76 million sampling points (that is, 5.76 million sampling times).

[0105] As the wind turbine blades rotate, sound waves are continuously radiated outward through the interaction between the turbulent boundary layer on the blade surface and the air. The frequency range of the sound signal generated by the blades in this process is mainly concentrated in the range of 8kHz–18kHz. Therefore, the passband range of the bandpass filter can be 8kHz–18kHz. Through bandpass filtering, low-frequency environmental noise (such as mechanical vibration and wind rumble) and extremely high-frequency (>20kHz) environmental interference are effectively suppressed.

[0106] Bandpass filtering is performed using a 128th-order FIR (Finite Impulse Response) bandpass filter. The passband range of the bandpass filter is... to Hamming windows can be used to design FIR bandpass filters.

[0107] For example, the bandpass filter (i.e., the bandpass filter coefficients) can be calculated using the following formulas two through five. : Formula 2:

[0108] Formula 3:

[0109] Formula 4:

[0110] Formula 5:

[0111] in, This is the lower limit of the passband range of the bandpass filter, that is... ; This is the upper limit of the passband range of the bandpass filter, that is... ; The sampling rate of the acoustic signal is, that is... ; and They are respectively and The normalized frequency, in terms of the Nyquist frequency ( (That is, half of the sampling rate) is used as the benchmark, and and Normalize to the range of 0 to 1; The value of 64 refers to the center of symmetry of the sequence; For Hamming window functions; Let be the unit impulse response of an ideal bandpass filter; n is the position index of the discrete sequence, with values ​​ranging from [1, 128]. It is a discrete sequence, that is, a series of coefficients arranged in order. n is the position index of the discrete sequence, used to indicate which coefficient it is.

[0112] Based on the above bandpass filter Zero-phase filtering (forward-backward filtering) is used to filter the signal, eliminating the phase distortion introduced by the bandpass filter, and obtaining the bandpass-filtered signal. Zero-phase filtering includes: The forward pass is processed through h[n] (performed once by convolution) to obtain the forward result y1 (i.e., forward filtering); the time sequence of y1 is reversed as a whole, and then the same result is passed through h[n] again. Perform a convolution to obtain the inverse result y2 (i.e., inverse filtering); flip y2 back to its original time order to obtain the final bandpass output. For example, forward filtering includes the following steps: moving the window to the right by only one sampling point at a time, where the window length is... The total length (with 129 coefficients, the window length is 129 sampling points) is such that, at each position, the overlapping sampling points are... Multiply each product point by point, then sum all the products to get the output for the current position, and move the window accordingly. After considering all positions, the positive result y1 is obtained. The number of sampling points in the positive result y1 is... The number of sampling points is the same. The principle of inverse filtering is the same as that of forward filtering, and will not be repeated here.

[0113] For example, the following Formula 6 is used to calculate the bandpass filtered signal. : Formula Six:

[0114] in, The acoustic signal after removing the DC component; These are the denominator coefficients of the bandpass filter, used to indicate that an FIR filter is currently being used. There is no feedback recursion; it relies solely on the first parameter. Complete the convolutional filtering.

[0115] The zero-phase filtering of the aforementioned 128th-order FIR bandpass filter includes two filters: forward filtering and reverse filtering. This allows for more thorough filtering of noise below 8kHz and above 18kHz. Furthermore, the forward filtering (which introduces a delay) and the reverse filtering (which introduces an equal amount of time advance) result in equal and opposite phase shifts, ultimately canceling each other out. It is perfectly aligned with the time axis of the original acoustic signal, ensuring linear phase characteristics and avoiding signal waveform distortion.

[0116] This step standardizes the acoustic signal, providing unified input data for subsequent dual-channel analysis (i.e., obtaining time-domain autocorrelation and frequency-domain quality indices separately).

[0117] In step 302, the envelope signals of each signal segment of the preprocessed acoustic signal are obtained.

[0118] In this step, the signal segment is obtained by framing the preprocessed acoustic signal. For example, a sliding window can be used to frame the preprocessed acoustic signal, and the length of the sliding window... Sliding step size (That is, non-overlapping continuous framing), for a 2-minute acoustic signal, a total of A signal segment, The duration is 2 minutes. The fan's rotation frequency is 0.1-2Hz (i.e., the fan's rotation cycle is 0.5-10s). To identify the blade harmonic structure at the lowest rotation speed and capture the complete time-domain rotation cycle based on the signal segment, the sliding window (i.e., the signal segment) length is set to 10 seconds. At this time, Δf = 1 / L = 0.1Hz, which is exactly equal to the lowest frequency of 0.1Hz within the 0.1-2Hz range, and the duration is equal to the fan's rotation cycle. If the sliding window is too short (e.g., 2-3 seconds), Δf = 1 / L = 0.33-0.5Hz, which is higher than 0.1Hz, resulting in insufficient frequency resolution and inability to identify the frequency domain characteristics in the signal segment. If the sliding window is too long (e.g., more than 30 seconds), the quality of the sound signal within the sliding window fluctuates greatly, stability is lost, quality discrimination is distorted, and the effective data utilization rate decreases. By selecting the length of the sliding window, the obtained signal segment meets the harmonic identification requirements of the full speed range of 0.1~2Hz, without the problem of insufficient resolution or excessive redundancy. The duration of the signal segment is equal to the rotation cycle of the lowest speed of the wind turbine. Each window contains at least one complete blade sweep process, and the periodic related features are calculated accurately.

[0119] In some embodiments, step 302 includes: squaring the amplitude at all sampling times in each signal segment to obtain a first instantaneous relative power value at each sampling time; removing the DC component of the first instantaneous relative power value to obtain a second instantaneous relative power value; and performing low-pass filtering based on the second instantaneous relative power value at each sampling time to obtain the envelope signal of each signal segment. In this step, by squaring, the instantaneous amplitude domain of the signal segment is converted into instantaneous power, moving the slowly varying periodic information (i.e., the envelope signal) originally hidden in the high-frequency carrier amplitude to the low-frequency band; by removing the DC component of the first instantaneous relative power value, the fixed bias generated by the squaring operation is eliminated, preventing the envelope signal from being boosted; by low-pass filtering, the frequency doubling carrier and high-frequency noise are filtered out, retaining the envelope signal reflecting the slowly varying energy. Through this step, the low-frequency blade rotation information (i.e., the envelope signal) submerged in high-frequency clutter signals is extracted.

[0120] In some embodiments, the cutoff frequency of the low-pass filter is 10Hz. That is, the passband range is 0-10Hz. This cutoff frequency just covers the 0.1~2Hz fundamental frequency (wind turbine blade rotation frequency) and multiple harmonics, completely preserving the waveform shape of the envelope signal while maximizing the filtering of high-frequency interference signals. If the cutoff frequency is too low, higher harmonics will be lost, resulting in envelope signal distortion; if the cutoff frequency is too high, high-frequency noise will remain, while also taking into account computational efficiency. This step can be performed using a 64th-order FIR low-pass filter for zero-phase filtering. Generally, the higher the filter order, the more accurate the filtering, but the greater the amount of convolution computation; the lower the order, the faster the computation speed, but the coarser the filtering. The filter order of this embodiment is 64, which is moderate, balancing a certain level of filtering accuracy and computational speed. Zero-phase filtering ensures that the envelope signal and the original acoustic signal are perfectly time-aligned, with no phase delay, ensuring accurate time coordinates in subsequent time-domain periodic calculations, etc. The principle of this low-pass filter is the same as that of the band-pass filter described above (the passband range and filter order are different), so it will not be repeated here.

[0121] For example, the first instantaneous relative power quantity is calculated using the following formula seven. : Formula 7:

[0122] in, The signal after bandpass filtering The i-th signal segment, where i is a positive integer.

[0123] For example, the second instantaneous relative power quantity is calculated using the following formula eight. : Formula 8:

[0124] in, This represents the relative power at the first instant. It is the average amplitude of all sampling times in the signal segment.

[0125] For example, the envelope signal is calculated using the following formula nine. : Formula Nine:

[0126] in, 1 represents the coefficients of the low-pass filter, order 64; 1 represents the denominator coefficients of the low-pass filter. This is the relative power quantity at the second instant.

[0127] In the above steps, the envelope signal is extracted by squaring and low-pass filtering. Other methods can also be used to extract the envelope signal, such as using the Hilbert Transform or the RMS envelope method. This disclosure does not specifically limit the specific methods used.

[0128] In step 303, the envelope signal is downsampled to obtain the target envelope signal.

[0129] In some embodiments, downsampling refers to reducing the sampling rate of the envelope signal. With an original acoustic signal sampling rate of 48kHz (48,000 samples per second), for an envelope signal with a bandwidth of only 10Hz, the signal fluctuates at most 10 times per second. This means that for each signal fluctuation, 48,000 completely repetitive and informationless sampling points are repeatedly collected, resulting in severe oversampling. To improve computational efficiency, the envelope signal is downsampled to a target sampling rate. It can be At this time, the downsampling factor By downsampling, the computational load of subsequent autocorrelation calculations is significantly reduced (from 480,000 points to 500 points). Furthermore, the effective information related to the blade state is concentrated in the fundamental frequency and harmonic range of 0.1–10 Hz. The Nyquist frequency of 25 Hz (defined as half the sampling frequency) completely covers this range, providing ample margin. Therefore, effective time-domain and frequency-domain characteristics of the target envelope signal are not lost. During downsampling, the sampling interval is... Finally, the target envelope signal is obtained. For example, each interval From the envelope signal Extract a sampling point from the sample to obtain the target envelope signal. Alternatively, reduce the sampling factor. For example, every 959 sampling points, from the envelope signal Extract a sampling point from the sample to obtain the target envelope signal. .

[0130] Target sampling rate after downsampling for The corresponding Nyquist frequency is 25Hz. According to the Nyquist theorem, sampling at 50Hz will not cause distortion only when all frequency components in the envelope signal are below 25Hz. If there are frequency components above 25Hz, these components will fold and alias into the lower frequency band after decimation, mixing with the envelope signal and becoming inseparable. Therefore, a low-pass filter (anti-aliasing filter) must be used to filter out all frequency components above 25Hz before decimation. Accordingly, in some embodiments, before downsampling the envelope signal, the method further includes: performing a low-pass filter on the envelope signal, the bandwidth of which is determined according to the target sampling rate of downsampling. The bandwidth of the low-pass filter is 0-25Hz, completely suppressing all frequency components >25Hz and eliminating the risk of aliasing.

[0131] In step 304, based on the target envelope signal of each signal segment, the time-domain autocorrelation index of each signal segment is obtained. The target envelope signal indicates the curve of amplitude change with time at each sampling moment in the signal segment, and the time-domain autocorrelation index indicates the periodicity of the envelope signal.

[0132] In some embodiments, step 304 includes: obtaining multiple autocorrelation indices between the target envelope signal of each signal segment and the corresponding multiple first lag signals, wherein the multiple first lag signals refer to target envelope signals that lag behind the target envelope signal at different sampling times, and the multiple autocorrelation indices indicate the similarity between the target envelope signal and the corresponding first lag signals; and outputting the maximum value among the multiple autocorrelation indices of each signal segment as the time-domain autocorrelation index of each signal segment.

[0133] In this step, the target envelope signal corresponding to the blade rotation has quasi-periodicity, with the fundamental frequency of the blade rotation as the basis. For example, its cycle =0.5–10 seconds, each revolution of the blade will produce similar impacts and force changes at the same position, therefore the target envelope signal approximately satisfies Since a wind turbine consists of multiple blades, if the wind turbine has three blades, the period T of the target envelope signal... BPF = / 3, the lag time of a certain first lag signal of the target envelope signal relative to the target envelope signal is equal to or When the autocorrelation index is a multiple of the target envelope signal, it is relatively large, while at other times it is relatively small. When wind noise is high, regardless of the lag time of the first lag signal relative to the target envelope signal (0.5–10 seconds), the waveforms of the target envelope signal and the corresponding first lag signal do not match. At this time, the autocorrelation index is close to 0 across the entire range, and the maximum value among the multiple autocorrelation indices of the signal segment is extremely low. This indicates that the signal segment is too chaotic to extract a stable period; that is, wind noise interference disrupts the periodicity of the target envelope signal, significantly reducing the autocorrelation index. Therefore, the maximum value of the autocorrelation index within the lag range can be used as a direct measure of the periodicity of the target envelope signal. Specifically, the lag time of the first lag signal relative to that signal segment is within the range of 0.5–10 seconds, meaning the lag range of the first lag signal is 0.5–10 seconds. Thus, only within one period... The calculation can be done internally.

[0134] The peak value M of the autocorrelation index in the lag range ac (That is, the time-domain autocorrelation index) directly measures the strength of the time-domain periodicity of the target envelope signal. It only depends on the prior range of the wind turbine blade rotation period [0.5, 10] seconds. The autocorrelation index has low computational complexity, with a computational complexity of O(NlogN), where N is the number of sampling points in each signal segment. It can be accelerated by FFT (Fast Fourier Transform), making it suitable for real-time initial screening of signal segments.

[0135] For example, the autocorrelation index of the target envelope signal is calculated using the following formula. : Formula 10:

[0136] Where m is the lag index, representing a lag of m sampling times (or sampling points), which means the target envelope signal is shifted to the right by m sampling times. M is a positive integer, and the value of M can be the total number of sampling points included in the target envelope signal. n represents the nth sampling time in the target envelope signal. This represents the amplitude at the nth sampling time in the target envelope signal; This represents the sum of squares of the amplitudes at all sampling times in the target envelope signal; That is, when m=0 (zero lag), the autocorrelation index is 1.

[0137] For example, the time-domain autocorrelation index of the target envelope signal is calculated using Formula 11 below. : Formula 11:

[0138] in, Let be the time-domain autocorrelation index of the i-th target envelope signal, where i is a positive integer. ∈[0,1], A value close to 1 indicates that the target envelope signal has strong periodicity (high-quality signal). A value close to 0 indicates that the target envelope signal has no obvious quasi-periodicity (severe wind noise pollution). Let m be the lag time of the first lag signal corresponding to the i-th target envelope signal. Then, convert the lag index m of the first lag signal into a lag time. (Second); Let be multiple autocorrelation indices for the i-th target envelope signal, where each autocorrelation indice represents the degree of similarity between the i-th target envelope signal and a first lag signal with a different lag time.

[0139] In step 305, based on the frequency domain characteristics and time domain autocorrelation index of the target envelope signal of each signal segment, the frequency domain quality index of each signal segment is obtained. The frequency domain quality index indicates the frequency domain quality of the signal segment under the time domain autocorrelation index.

[0140] In this step, the frequency domain quality index of each of the signal segments can be obtained, as well as the frequency domain quality index of the signal segments whose time domain autocorrelation index is greater than or equal to the corresponding preset threshold.

[0141] In some embodiments, step 305 includes: obtaining a frequency domain quality index of the first signal segment based on the frequency domain characteristics and time domain autocorrelation index of the target envelope signal of the first signal segment, wherein the time domain autocorrelation index of the first signal segment is greater than or equal to a corresponding preset threshold. That is, the signal segments are initially screened using the time domain autocorrelation index, and the first signal segments with time domain autocorrelation indices greater than or equal to the corresponding preset thresholds are selected to obtain the frequency domain quality index of the first signal segment, instead of obtaining the frequency domain quality index for all signal segments. This reduces the computational load of obtaining the frequency domain quality index and improves the real-time performance of the signal segment selection.

[0142] In some embodiments, step 305 includes: obtaining a frequency domain fusion quality index for each signal segment based on at least two frequency domain features of the target envelope signal of each signal segment, wherein the at least two frequency domain features refer to different frequency domain features obtained by different extraction methods based on each target envelope signal, and the frequency domain fusion quality index indicates the frequency domain quality of the signal segment; adjusting the frequency domain fusion quality index based on the time domain autocorrelation index of each signal segment to obtain the frequency domain quality index of each signal segment. In this step, for the target envelope signal of each signal segment, at least two frequency domain features are fused to obtain a frequency domain fusion quality index. Since different frequency domain features reflect the frequency domain characteristics of the corresponding frequency domain signal of the target envelope signal in different aspects, different frequency domain features can complement each other from different angles, so that the frequency domain fusion quality index can more accurately reflect the frequency domain quality of the signal band. Furthermore, the frequency domain fusion quality index is adjusted based on the time domain autocorrelation index to obtain the frequency domain quality index of each signal segment. The frequency domain quality index is further constrained by the time domain autocorrelation index, which further eliminates the inherent blind spot of the frequency domain fusion quality, so that the frequency domain quality index can more accurately reflect the frequency domain quality of the signal band.

[0143] In some embodiments, the frequency domain characteristics include at least two of spectral kurtosis, spectral entropy, and harmonic regularity factor. Spectral kurtosis indicates the sharpness of the energy distribution of the frequency domain signal corresponding to the envelope signal, spectral entropy indicates the concentration of the energy distribution of the frequency domain signal corresponding to the envelope signal, and harmonic regularity factor indicates the periodicity of the frequency domain signal corresponding to the envelope signal.

[0144] In order to obtain at least two frequency domain features of the target envelope signal for each signal segment, the target envelope signal is first subjected to a fast Fourier transform to obtain the frequency domain signal.

[0145] For example, the following formula 12 is applied to calculate the frequency domain signal. : Formula 12:

[0146] in, For the kth frequency, If the duration of the signal segment L = 10s, the frequency resolution For example, when k=0, f0=0Hz, which corresponds to the amplitude of the DC component; when k=1, f1=0.1Hz, which corresponds to the amplitude of the 0.1Hz frequency. It is a frequency domain signal, including amplitudes (or vibrations) at different frequencies. The sampling rate f is the total number of sampling points participating in the Fast Fourier Transform. If the duration of the signal segment is L = 10s, the sampling rate f is... s,ds =50Hz, the target envelope signal has a total of 500 sampling points. It can be 500, or all sampling points in the target envelope signal can be padded with zeros to integer powers of 2 (such as 512 points), that is... It can also be 512; The amplitude of the nth sampling point in the target envelope signal participating in the Fast Fourier Transform is given, where n takes the value [0, ..., ...]. ]; j is the imaginary unit; let Ignoring the DC component, the main analysis is... The energy (or amplitude) distribution in the range of 0–5 Hz.

[0147] For example, the spectral kurtosis is calculated using formulas thirteen to fifteen below. : Formula Thirteen:

[0148] Formula Fourteen:

[0149] Formula 15:

[0150] in, It is the sum of the amplitudes of all frequencies in the frequency domain signal, where j represents the j-th frequency; The amplitude of any frequency in the frequency domain signal; For the k-th frequency component f k exist The proportion of the total; For frequency; The mean frequency; Let V be the frequency variance. The spectral kurtosis is defined as the normalized fourth-order central moment – ​​spectral kurtosis. For an ideal single frequency (corresponding to the fundamental frequency of a clean blade), the spectral kurtosis is very high; for a flat noise spectrum (the frequency domain signal of white noise is approximately flat), the spectral kurtosis tends to a lower value (for example, the baseline spectral kurtosis of Gaussian noise is often referenced in engineering, which is 3). Therefore, a high spectral kurtosis value indicates good quality of the frequency domain signal.

[0151] For example, the spectral entropy is calculated using the following formulas sixteen to seventeen. : Formula Sixteen:

[0152] Formula 17: ,

[0153] in, It is the square of the amplitude of each frequency in the frequency domain signal, which is also the instantaneous power of each frequency; This is the sum of the instantaneous power at each frequency in the frequency domain signal; Instantaneous power at each frequency exist The proportion of the total; For example, a very small constant This is to prevent logarithmic divergence. The lower the spectral entropy, the more concentrated the energy is on a few frequencies, and the stronger the regularity of the frequency domain signal; the higher the spectral entropy, the more dispersed the energy is on many frequencies, indicating dispersed frequency domain energy, greater uncertainty, and corresponding to the disordered characteristics of wind noise. Low spectral entropy indicates good quality frequency domain signal.

[0154] In some embodiments, the method further includes: performing a fast Fourier transform on the amplitude at each sampling time in the envelope signal to obtain a frequency domain signal, wherein the frequency domain signal indicates the amplitude intensity of signals at different frequencies in the envelope signal; acquiring multiple frequency domain autocorrelation indices between the frequency domain signal and multiple second hysteresis signals at different preset fundamental frequencies, wherein the preset fundamental frequency indicates the rotation frequency of the wind turbine, and the multiple second hysteresis signals refer to frequency domain signals that lag behind the frequency domain signal by different integer multiples of the preset fundamental frequency, and the multiple frequency domain autocorrelation indices indicate the similarity between the frequency domain signal and the multiple second hysteresis signals; acquiring the product between the multiple frequency domain autocorrelation indices corresponding to each preset fundamental frequency to obtain the autocorrelation product of each preset fundamental frequency, and outputting the maximum value among the multiple autocorrelation products as the harmonic regularity factor of the envelope signal.

[0155] In this step, the frequency domain signal is obtained using Formula Twelve above; each preset fundamental frequency refers to the fundamental frequency candidate range. One of the values ​​within; under the same preset base frequency, obtain the frequency domain autocorrelation index between the frequency domain signal and the second lag signal that is lagging by different integer multiples of the preset base frequency, calculate the product of multiple frequency domain autocorrelation indices under the same preset base frequency, that is, calculate the autocorrelation product at the lag of the first preset number (e.g., the first 4) harmonics under the same preset base frequency. If a certain preset base frequency is exactly equal to the wind turbine blade rotation base frequency, then each frequency domain autocorrelation index of the preset base frequency is relatively large. At this time, the autocorrelation product of the preset base frequency is the maximum value among all autocorrelation products, that is, the autocorrelation product at this time is relatively large as the harmonic regularity factor. This method does not require peak picking based on the frequency domain signal in advance to determine the fundamental frequency of the wind turbine blade rotation, that is, it does not require explicit fundamental frequency estimation. During peak picking, only peaks in the frequency domain signal with amplitudes exceeding a threshold are identified as candidate fundamental frequencies. If the threshold is too high, it is easy to miss detections. If the threshold is too low, it is easy to misjudge small peaks generated by random noise as valid peaks. It is also easy to identify interference peaks with amplitudes higher than the true fundamental frequency as fundamental frequencies. The method of this disclosure avoids the problems of thresholds and false peaks in explicit fundamental frequency estimation.

[0156] For example, the harmonic regularity factor H is calculated using the following formulas 18 to 20: Formula 18:

[0157] Formula 19:

[0158] Formula 20:

[0159] in, This is due to frequency lag; It is a frequency domain signal; It is a second hysteresis signal of the frequency domain signal; It is a frequency domain autocorrelation index. ,and exist A peak will appear at that location; The preset fundamental frequency is within the fundamental frequency range of the wind turbine blade rotation. Take evenly inside There are 50 candidate values, which means we get 50 preset base frequencies. k represents different integer multiples of the preset fundamental frequency, and the value of k is a positive integer. The maximum value of k is [1, 4], which is the maximum value before calculation. The autocorrelation product at the hysteresis of the second harmonic; To a preset base frequency The autocorrelation product. If the envelope spectrum (i.e., the frequency domain signal) has a clear and equally spaced fundamental-harmonic sequence, then in For the correct fundamental frequency At that time, all They are all relatively large. Larger; if there is no harmonic structure (such as isolated single frequency or noise), then at least for The lag, Very small, leading to Extremely low. This method does not rely on explicit fundamental frequency estimation, but automatically discovers the optimal fundamental frequency match by maximizing the autocorrelation product.

[0160] Among them, in the above formula It can be other values, such as The calculation is faster at this point, but the suppression of wind noise is slightly weaker, for example... or This results in stronger suppression of wind noise, but also increases computational overhead and the number of preset fundamental frequencies. It can be more or less. With more (e.g., 100), the frequency resolution can be increased. Fewer (e.g., 25) can reduce the amount of computation, but this disclosure does not specifically limit this aspect.

[0161] In the above method, at each preset fundamental frequency, the product of the frequency domain autocorrelation index R(Δf) at the lag of the first four harmonics is calculated. For the 50 preset base frequencies The maximum value is taken to obtain the harmonic regularity factor. This method has three advantages: First, it does not require explicit fundamental frequency estimation, thus avoiding the threshold and spurious peak problems in peak picking; second, by using the product of multiple frequency domain autocorrelation indices as the harmonic regularity factor, it naturally suppresses isolated single-frequency interference (the autocorrelation of a single frequency approaches zero at a lag of k≥2); third, the preset dense sampling of the fundamental frequency (50) ensures coverage of any rotating frequency.

[0162] For example, the frequency domain fusion quality index is calculated using the following formula 21. : Formula 21:

[0163] in, For spectral kurtosis; Spectral entropy; This is the harmonic regularity factor. The larger the numerator Kurt, the smaller the denominator. The smaller +0.1 is, the larger the factor H is. The higher the value, the more likely it is to be divided by zero, where the constant 0.1 is used to prevent division by zero.

[0164] For example, the frequency domain quality index is calculated using the following formula twenty-two. : Formula 22:

[0165] in, Let be the frequency domain quality index of the i-th signal segment; Let be the frequency domain fusion quality index for the i-th signal segment; The time-domain autocorrelation index for the i-th signal segment (or the superscript used to represent the i-th segment can be removed, and it can be expressed as...) (Representation). Coefficient Guaranteed: When the multiplicative factor decreases to Retain half of the frequency domain fusion quality indicators to avoid excessive penalties; when When the multiplicative factor is close to To maintain high frequency domain fusion quality indicators. The range of values ​​is In practical applications, the typical value of a normal signal is in between.

[0166] Formula 22 above can also be: or

[0167] in, and These are the weighting coefficients.

[0168] In step 306, target signal segments are extracted from the acoustic signal based on the time-domain autocorrelation index and frequency-domain quality index of each signal segment. At least one of the time-domain autocorrelation index and frequency-domain quality index of the target signal segment is greater than or equal to the corresponding preset threshold.

[0169] Based on the time-domain autocorrelation index and frequency-domain quality index of each signal segment, the target signal segments extracted from the acoustic signal are different. Accordingly, in some embodiments, step 306 includes: extracting signal segments from the acoustic signal that satisfy both a first condition and a second condition as target signal segments, wherein the first condition is that the time-domain autocorrelation index is greater than or equal to a first preset index, and the second condition is that the frequency-domain quality index is greater than or equal to a second preset index; or, extracting signal segments from the acoustic signal that satisfy both the first and second conditions as target signal segments; or, performing a weighted summation of the time-domain autocorrelation index and the frequency-domain quality index of each signal segment to obtain a comprehensive quality index of each signal segment, and extracting signal segments from the acoustic signal whose comprehensive quality index is greater than or equal to a preset threshold as target signal segments, wherein the comprehensive quality index indicates the comprehensive quality of the signal segment in the frequency and time domains. In other words, the target signal segment can be the intersection of the signal segment that satisfies the first condition and the signal segment that satisfies the second condition. This method achieves a conservative screening of the target signal segment and avoids misjudgment by ignoring the time-domain periodicity based solely on the frequency domain quality index. The target signal segment can also be the union of the signal segment that satisfies the first condition and the signal segment that satisfies the second condition. The target signal segment can also be a signal segment whose comprehensive quality index is greater than or equal to a preset threshold.

[0170] The second preset index is dynamically determined based on the frequency domain quality index of each signal segment. Accordingly, in some embodiments, the method further includes: obtaining the inter-class variance of the frequency domain quality index of each signal segment; and outputting the frequency domain quality index corresponding to the largest inter-class variance among the inter-class variances of each signal segment as the second preset index. In this step, the inter-class variance is a statistical indicator that measures the degree of dispersion of the mean of different categories of data relative to the overall mean. The larger the value, the higher the separation between categories and the better the classification effect. By maximizing the inter-class variance (i.e., the Otsu algorithm), the frequency domain quality index of each signal segment is regarded as a one-dimensional data distribution, which includes high-quality data and low-quality data. This step automatically determines the dividing point with the largest variance between high-quality data and low-quality data, which is the second preset index. The frequency domain quality index of signals collected in different wind farms and under different meteorological conditions is different. This method can automatically adjust the second preset index to adapt to environmental changes. This method is completely parameterless, avoiding the tedious process of repeated manual parameter tuning, and is suitable for large-scale industrial deployment.

[0171] For example, the second preset index is calculated by applying formulas 23 to 25 below. : Formula 23:

[0172] Formula 24:

[0173] Formula 25:

[0174] in, Let be the frequency domain quality index of the i-th signal segment; This represents the minimum frequency domain quality index among the frequency domain quality indices of each signal segment; This represents the maximum value of the frequency domain quality index among the frequency domain quality indices for each signal segment; The normalized score for each signal segment, with a value range of [0,1]; each candidate threshold , This represents the total number of signal segments; For all values ​​above t The average value; For all values ​​below t The average value; For all values ​​above t The proportion of all signal segments; For all values ​​below t The proportion of all signal segments; The variance between types I; Let t be the set of all inter-class variances, and let t represent the frequency domain quality index corresponding to the largest inter-class variance obtained by iterating through all inter-class variances. .

[0175] The determination of the first preset index can take various forms, such as a fixed preset value, or determination based on the time-domain autocorrelation index of each signal segment, etc. Accordingly, in some embodiments, the method further includes: outputting the preset value as the first preset index; or, sorting the time-domain autocorrelation indices of each signal segment in ascending order to obtain an ordered dataset, and outputting the quantile of the preset percentage of the ordered dataset as the first preset index; or, obtaining the inter-type variance of the time-domain autocorrelation indices of each signal segment, and outputting the time-domain autocorrelation index corresponding to the largest inter-type variance among the inter-type variances of each signal segment as the first preset index. For example, the preset value can be determined based on experience, and the quantile of the preset percentage can also be set based on experience; for example, the first preset index For example, the preset percentage quantile can be the 30th percentile.

[0176] For example, the first preset index is calculated using formulas 26 to 28 below. : Formula 26:

[0177] Formula 27:

[0178] Formula 28:

[0179] in, Let be the time-domain autocorrelation index of the i-th signal segment; It is the minimum value among the time-domain autocorrelation indices of each signal segment; This represents the maximum value among the time-domain autocorrelation indices for each signal segment; The normalized score for each signal segment, with a value range of [0,1]; each candidate threshold , This represents the total number of signal segments; For all values ​​above t The average value; For all values ​​below t The average value; For all values ​​above t The proportion of all signal segments; For all values ​​below t The proportion of all signal segments; The variance between types 2; Let t be the set of all inter-type variances, and let represent the frequency domain quality index corresponding to the largest inter-type variance obtained by iterating through all inter-type variances. .

[0180] Taking the example that the time-domain autocorrelation index of the target signal segment is greater than or equal to a first preset index and the frequency-domain quality index is greater than or equal to a second preset index, the steps for determining the judgment label of each signal segment are explained. The judgment label indicates whether the signal segment is usable, including the following steps: Based on the time-domain autocorrelation index of each signal segment Compared with the first preset index The relative size determines the first decision label for each signal segment; if the signal segment's... The first determination label of the signal segment is available; if the signal segment's... The first determination label for this signal segment is unavailable; Based on the frequency domain quality index of each signal segment With the second preset index The relative size is used to determine the second decision label for each signal segment. If the signal segment's... The second determination tag for the signal segment is available if the signal segment's... The first determination label for this signal segment is unavailable; The signal segment is considered usable if both the first and second determination tags are usable; otherwise, the signal segment is considered unusable.

[0181] For example, the first determination label of the signal segment is determined by applying the following formula twenty-nine: Formula 29:

[0182] in, The first preset indicator has a value of 0.4; Let be the time-domain autocorrelation index of the i-th signal segment; It is the first determination label for the i-th signal segment.

[0183] For example, the second determination label of the signal segment is determined by applying the following formula: Formula 30:

[0184] in, This is the second preset indicator; Let be the frequency domain quality index of the i-th signal segment; This is the second determination label for the i-th signal segment.

[0185] For example, the following formula thirty-one is applied to determine the determination label of the signal segment: Formula 31:

[0186] in, This is the first determination label for the i-th signal segment; This is the second determination label for the i-th signal segment; Let i be the determination label for the i-th signal segment, when and Both are available (i.e., AND logic). Available if available, otherwise The signal segments are deemed unusable. The time-domain autocorrelation index is used to quickly filter out signal segments with basic periodicity, while the frequency-domain quality index is used to accurately identify high-quality signal segments from a multi-dimensional feature perspective. This step is equivalent to taking the intersection of the first and second usable signal segments, ensuring that the usable (finally retained) signal segments simultaneously satisfy both "time-domain periodicity" and "frequency-domain harmonics," effectively reducing the risk of misjudgment based on a single index.

[0187] When extracting target signal segments from acoustic signals, in order to ensure the quality of the extracted target signal segments, signal segments labeled as unusable can be extended forward and backward by a preset time period before extraction, in order to avoid extracting degraded transition bands near the boundaries of signal segments labeled as unusable. In some embodiments, step 306 includes: extending each second signal segment in the acoustic signal forward and backward by a preset time period to obtain a third signal segment, wherein the second signal segment is a signal segment in which at least one of the time-domain autocorrelation index and the frequency-domain quality index is less than a corresponding preset threshold; and extracting target signal segments from the acoustic signal other than the third signal segment.

[0188] In this step, the second signal segment, which is the one labeled as unusable, has an arbitrary preset time period, such as 1 second. The buffering strategy of extending the preset time period before and after takes into account the possible quality gradient transition zone near the signal segment boundary, ensuring the complete removal of the deterioration transition zone and guaranteeing the quality of the extracted target signal segment.

[0189] After extending the second signal segment forward and backward, the resulting third signal segment may have temporal overlap. The third signal segment can be merged before extraction. Accordingly, in some embodiments, the method further includes merging the third signal segments with temporal overlap. This step merges the overlapping regions, avoiding redundant cutting operations on the overlapping regions when extracting the target signal segment if adjacent signal segments are all determined to be unusable.

[0190] When extracting target signal segments, a binary mask indicating whether a signal segment is usable can be generated first based on the determination label, and then extraction can be performed based on the binary mask. Accordingly, in some embodiments, extracting target signal segments from the acoustic signal other than the third signal segment includes: generating a binary mask for each sampling time step based on the determination label of the signal segment, wherein the first value in the binary mask indicates usability (or retention) and the second value indicates unusability (removal); scanning along the time axis of the acoustic signal to extract all target signal segments corresponding to the intervals of the first value.

[0191] For example, the time interval of the third signal segment is determined by applying the following formula thirty-two. : Formula 32:

[0192] in, for The time interval of the signal segment That is, the tag is determined to be unusable; [0,T] is the time interval of the acoustic signal; For 0 and The maximum value in [0,T], where 0 is the left endpoint of [0,T]. for The left endpoint; For T and The minimum value in [0,T], where T is the right endpoint of [0,T]. for The right endpoint.

[0193] In addition to the aforementioned time-domain autocorrelation index and frequency-domain quality index, the quality of each signal segment can be classified to obtain a quality score for each signal segment. Accordingly, in some embodiments, the method further includes: obtaining a quality score for each signal segment based on the envelope signal of each signal segment, wherein the quality score indicates the quality of the signal segment; and extracting a target signal segment from the acoustic signal based on the time-domain autocorrelation index and frequency-domain quality index of each signal segment, including: extracting a target signal segment from the acoustic signal based on the time-domain autocorrelation index, frequency-domain quality index, and quality score of each signal segment, wherein at least one of the time-domain autocorrelation index, frequency-domain quality index, and quality score of the target signal segment is greater than or equal to a corresponding preset threshold.

[0194] In this step, the quality score can include 0 and 1, or other values, or even more values. This embodiment does not specifically limit this. Different quality scores represent different qualities of the signal segments. For example, 1 represents a signal segment with better quality (i.e., less affected by wind noise), and 0 represents a signal segment with poorer quality (i.e., more affected by wind noise). The quality score achieves an initial screening of the signal segment's quality. For example, an LSTM (Long Short-Term Memory) time-series classifier can be used to obtain the quality score of the envelope signal of each signal segment. An LSTM time-series classifier is a deep learning model specifically designed for processing ordered time-series data. Its input is the envelope signal of each signal segment, and its output is the quality score of the envelope signal of each signal segment. For example, taking a quality score including 0 and 1 as an example, the threshold for the quality score can be 1, as long as it can separate 0 and 1; for example, it can also be 0.5.

[0195] Accordingly, after increasing the quality score, the target signal segment can be a signal segment whose time-domain autocorrelation index, frequency-domain quality index, and quality score are greater than or equal to the corresponding preset thresholds. Alternatively, the target signal segment can be a signal segment whose time-domain autocorrelation index, frequency-domain quality index, and quality score are any two of the corresponding preset thresholds, and so on.

[0196] The time-domain autocorrelation index and frequency-domain quality index are obtained based on the corresponding physical laws, while the quality score is obtained by the LSTM time-series classifier based on the learned patterns of the envelope signal. Since the LSTM time-series classifier learns the full-dimensional temporal differences of envelope signals affected by different wind noise from massive samples, including many waveform details that are not quantified by the time-domain autocorrelation index and frequency-domain quality index (such as the density of glitches in the envelope, the smoothness of fluctuations, etc.), it can identify the implicit interference missed by the time-domain autocorrelation index and frequency-domain quality index. Therefore, this step is equivalent to combining physical rules and the patterns of the envelope signal learned by deep learning for joint screening, which further improves the accuracy of the obtained target signal segments.

[0197] In this embodiment, the acquisition processes of the time-domain autocorrelation index and the frequency-domain quality index are both based on the target envelope signal obtained by preprocessing the acoustic signal, extracting the envelope, and downsampling, avoiding redundant calculations. The rapid screening based on the time-domain autocorrelation index and the fine-grained frequency-domain discrimination based on the frequency-domain quality index complement each other, balancing efficiency and accuracy. From the preprocessing of the acoustic signal to the extraction of the target signal segment, no manual parameter setting is required throughout the process, improving the automation level of the method. The frequency-domain quality index integrates four-dimensional features: frequency-domain kurtosis (measuring spectral peak sharpness), spectral entropy (measuring energy concentration), harmonic regularity factor H (measuring the integrity of the fundamental-harmonic structure), and time-domain autocorrelation index. These four features complement each other from different perspectives: high frequency-domain kurtosis indicates sharp spectral peaks, low spectral entropy indicates concentrated energy, a large harmonic regularity factor H indicates a clear fundamental-harmonic equidistant structure, and a large time-domain autocorrelation index indicates that the time-domain waveform of the envelope signal has quasi-periodicity. All four conditions must be met simultaneously to obtain a high SQI. total value.

[0198] The method of this disclosure solves the problem of signal quality assessment and extraction caused by wind noise interference in the acquisition of acoustic signals from wind turbine blades. It quantifies the periodic intensity of the blade acoustic signal envelope from the time domain dimension (time domain autocorrelation index), enabling rapid detection of wind noise pollution. By integrating the spectral characteristics (kurtosis, entropy), harmonic regularity (harmonic regularity factor), and time domain periodicity of the frequency domain signal, a comprehensive and highly discriminative signal quality comprehensive scoring index (i.e., frequency domain quality index) is constructed. It achieves adaptive determination of at least one of the preset thresholds for the frequency domain quality index and the time domain autocorrelation index, avoiding manual parameter tuning and making the method of this disclosure robust to different signal-to-noise ratio environments. Based on the time domain autocorrelation index and the frequency domain quality index, it achieves signal segment quality assessment and also realizes automatic location and precise removal of contaminated signal segments, preserving high-quality signal segments (also the target signal segments) for subsequent analysis.

[0199] The method described above will be illustrated using two sets of measured acoustic signals of wind turbine blades as examples, based on embodiments of this disclosure.

[0200] The two sets of measured acoustic signals from the wind turbine blades are designated as the first sample and the second sample, respectively. Both samples were collected in actual wind farm environments and are 48kHz sampling rate, 120-second duration, and dual-channel WAV (Waveform Audio File Format, standard lossless digital audio file format) files. Following the principle of "using only the first channel data," the first channel signal was extracted for analysis. The first and second samples are shown in Table 1, which provides data details for both samples.

[0201]

[0202] Table 1 The first sample exhibits a low RMS value (0.000446), a low crest factor (7.93), and a stable and uniform signal amplitude. These numerical characteristics indicate that the acoustic signal is less affected by wind noise, and the quasi-periodicity of the envelope signal can be fully reflected in the time-domain autocorrelation index. Experimental expectations: Channel 1 will have a high throughput (a large number of signal segments with a time-domain autocorrelation index greater than or equal to the first preset index), and Channel 2 will generally have a high and concentrated frequency domain quality index.

[0203] The second sample exhibits a high RMS value (0.003078, approximately 6.9 times that of a good signal), an extremely high crest factor (19.52), and multiple sudden high-amplitude segments. In this scenario, a high RMS value does not represent a "strong signal," but rather "strong wind noise," which is the fundamental reason for misjudgments by the traditional energy threshold method. Experimental expectations: Channel 1 has a low throughput (time-domain periodicity is disrupted by wind noise), while Channel 2 has a higher second preset index (adapting to the discrete quality distribution).

[0204] The parameter is configured as follows: the length of the sliding window. Sliding step size Both the first and second samples consist of 12 signal segments. First preset index The second preset index is the frequency domain quality index corresponding to the largest inter-class variance among the inter-class variances of each signal segment, which does not require manual setting. The extraction strategy for the target signal segment is: dual-channel intersection decision (AND) + 1 second extension before and after, that is, both the time domain autocorrelation index and the frequency domain quality index of the target signal segment meet the corresponding preset thresholds.

[0205] Figure 4 This is a schematic diagram of a first sample according to an exemplary embodiment. Figure 4Figure (a) shows a schematic diagram of the acoustic signal of the first sample, with the amplitude fluctuating uniformly within ±0.004 at each sampling time. Figure 4 Figure (b) shows a schematic diagram of the acoustic signal of the first sample after bandpass filtering at 8-18 kHz. Bandpass filtering effectively suppresses low-frequency mechanical vibrations and environmental noise below 8 kHz. Figure 4 Figure (c) shows the 12 signal segments of the time-domain autocorrelation index M of the first sample. The light gray signal segments are the 12 signal segments. ac If the value is greater than or equal to the first preset index, i.e., Mac ≥ 0.4, then the time-domain autocorrelation index M of the dark gray signal segment is passed. ac Less than the first preset index, that is, Mac < 0.4, failed. The quality distribution of each signal segment on the 120-second time axis is intuitively displayed. Most of the signal segments of the first sample are light gray. The destruction of time domain periodicity by wind noise is quantitatively verified. The 12 signal segments are marked from left to right as window 1 to window 12.

[0206] Figure 5 This is a schematic diagram of a second sample according to an exemplary embodiment. Figure 5 Figure (a) shows a schematic diagram of the acoustic signal of the second sample, which exhibits sudden high amplitudes (up to ±0.06) in the 30-60 second and 80-110 second regions, corresponding to intermittent strong wind noise events. Figure 5 Figure (b) shows a schematic diagram of the acoustic signal of the second sample after bandpass filtering at 8-18 kHz. Bandpass filtering effectively suppresses low-frequency mechanical vibrations and environmental noise below 8 kHz. Figure 5 Figure (c) shows 12 signal segments representing the time axis distribution of the second sample. The meanings of the light gray and dark gray signal segments are... Figure 4 The same as Figure (c) shows that only one signal segment passed through the second sample, which quantitatively verified the disruption of time-domain periodicity by wind noise.

[0207] Figure 6 This is a schematic diagram of a signal segment of window 7 in a first sample, according to an exemplary embodiment. For example... Figure 6 Figure (a) shows a segment of the acoustic signal in window 7 after the DC component has been removed and the signal has been bandpass filtered from 8 to 18 kHz. Figure 6 Figure (b) shows the relative power at the first instant after squaring the signal segment in window 7. Figure 6 Figure (c) shows the envelope signal obtained after low-pass filtering the relative power of the first instant. The envelope signal exhibits obvious slow-varying periodic fluctuations.

[0208] Figure 7This is a schematic diagram illustrating the autocorrelation index curves of multiple signal segments of a first sample according to an exemplary embodiment. Figure 8 This is a schematic diagram of the autocorrelation index curves of multiple signal segments of a second sample, according to an exemplary embodiment. Figure 7 Figures (a), (b), and (c) in the figure are schematic diagrams of the worst, median, and best windows of the time-domain autocorrelation index in the first sample, respectively. Figure 8 Figures (a), (b), and (c) in the diagrams represent the worst, median, and best windows of the time-domain autocorrelation index in the second sample, respectively. Taking a first preset index of 0.4 as an example, it can be seen that... Figure 7 In Figure (a), there is no autocorrelation index curve that exceeds the first preset index, but... Figure 7 In Figures (b) and (c), the autocorrelation index curves show autocorrelation values ​​exceeding the first preset index. However, overall, the autocorrelation index curves for all three windows exhibit significant peaks (M0) at lag times of 1–3 seconds. ac Although the window with a value of 0.209 did not meet the first preset metric, it was still higher than most windows in the second sample, demonstrating better overall time-domain periodicity. It can be seen that... Figure 8 The autocorrelation index curve in figure (a) is almost monotonically decaying (M). ac =0.012), with no periodic peak characteristics, indicating that wind noise has completely disrupted the periodicity of the envelope. Furthermore... Figure 8 The periodicity of the autocorrelation index curves in Figures (b) and (c) is also not as good as... Figure 7 Figures (b) and (c) in the text are obvious.

[0209] like Figure 4 and Figure 5 As shown in Figure (c), it summarizes the time-domain autocorrelation index (M) of all 12 windows of the first and second samples in a bar chart. ac (Value), light gray bars represent M ac A value greater than or equal to the first preset indicator (i.e., passing) indicates M (dark gray bar). ac Values ​​less than the first preset criterion (i.e., failing) are represented by dashed lines, indicating the first preset criterion. For the first sample, M ac The range is [0.2094, 0.5933], and the pass rate is 8 / 12 = 66.7%. For the second sample, M... ac The range is [0.0117, 0.4044], and the pass rate is 1 / 12 = 8.3%. The M value of the second sample... ac The mean was only about 1 / 5 of that of the first sample, and there was only one window M in the second sample. ac≥0.4 (Window #2: 0.4044, just past the dashed line corresponding to the first preset index). The pass rate difference between the two sets of data reached 8 times, quantitatively demonstrating the strong discriminative ability of the time-domain autocorrelation index to distinguish the quality of acoustic signals.

[0210] Figure 9 This is a window-by-window numerical diagram of the six dimensions of a first sample, according to an exemplary embodiment. Figure 10 This is a window-by-window numerical diagram of the six dimensions of the second sample according to an exemplary embodiment. The six dimensions are spectral kurtosis, spectral entropy, harmonic regularity factor, time-domain autocorrelation index, frequency-domain quality index, and frequency-domain fusion quality index. The numerical ranges of the dimensions are shown in Table 2. Table 2 is a table of numerical ranges for multiple dimensions.

[0211]

[0212] Table 2 according to Figure 9 and Figure 10 Figures (a) and (b) show that the differences in spectral kurtosis and spectral entropy between the first and second samples are relatively small. This verifies the limitation that a single frequency domain feature (such as spectral kurtosis or spectral entropy alone) cannot accurately determine the quality of a signal segment, thus highlighting the necessity of four-dimensional feature fusion in the embodiments of this disclosure. Figure 9 and Figure 10 In Figure (c), it can be seen that the difference in harmonic regularity factors is not obvious. According to... Figure 9 and Figure 10 Figure (e) shows the time-domain autocorrelation index. The differences were greatest; the first sample generally had windows with values ​​above 0.2, while most windows in the second sample were close to 0 (the lowest being 0.0117). The time-domain autocorrelation index... It becomes the single dimension with the highest distinguishing power. According to Figure 9 and Figure 10 Figures (d) and (f) show that after fusing spectral kurtosis, spectral entropy, and harmonic regularity factors, the resulting frequency domain fusion quality indices are relatively distinct, as are the spectral kurtosis, spectral entropy, harmonic regularity factors, and time-domain autocorrelation indices. After fusion, the resulting frequency domain quality indices show relatively clear distinctions. Furthermore, the second preset index differs significantly between the first and second samples (0.0872 vs 0.1217, a difference of 1.4 times), with the second sample's SQI being significantly higher. total The distribution is more discrete (high wind noise window SQI) total Extremely low, with some clean windows having a low SQI. total (Higher), the method in this embodiment automatically increases the threshold to adapt to this bimodal distribution, which demonstrates the superiority of adaptive thresholds over fixed thresholds.

[0213] Figure 11 This is a frequency domain signal diagram of window 8 of a first sample, illustrated according to an exemplary embodiment, wherein the M of the signal segment of window 8 ac maximum. Figure 12 This is a frequency domain signal diagram of window 2 of the second sample, as illustrated in an exemplary embodiment, wherein the M of the signal segment of window 2 ac Maximum. For example... Figure 11 and Figure 12 Figure (a) shows the full spectrum of the corresponding signal segment from 0 to 10 Hz (with the main spectral peak frequencies marked), as shown below. Figure 11 and Figure 12 Figure (b) is a magnified view of the 0–5 Hz area in Figure (a). In this figure, the discrete spectral structure is shown using a stem plot (matchstick plot). It can be seen that the optimal window for the first sample (window #8, M) is... ac =0.5933) The envelope spectrum (i.e., the frequency domain signal) exhibits 2–3 clear spectral peaks with distinct spectral structure; the optimal window for the second sample (window #2, M) ac The frequency domain signal with a value of 0.4044 also exhibits a certain harmonic structure (which precisely explains why the window passes the threshold), but the time domain autocorrelation index M of the second sample... ac Lower than the first sample.

[0214] Figure 13 This is a schematic diagram illustrating the determination label of a first sample according to an exemplary embodiment. Figure 14 This is a schematic diagram illustrating the determination label of a second sample according to an exemplary embodiment. Wherein, Figure 13 and Figure 14 Figure (a) shows the acoustic signal diagrams of the first and second samples, respectively. Figure 13 and Figure 14 Figure (b) in the figure is a schematic diagram of the first judgment label of the first sample and the second sample, respectively. Figure 13 and Figure 14 Figure (c) in the figure is a schematic diagram of the second judgment label for the first sample and the second sample, respectively. Figure 13 and Figure 14 Figure (d) in the figure are schematic diagrams of the judgment labels for the first and second samples, respectively. Figure 13 and Figure 14 Figure (e) shows the target signal segments of the first and second samples, respectively. As can be seen from the figure, based on... The signal segments in the first sample had a higher pass rate (8 / 12), and the judgment was more lenient, based on... The pass rate of signal segments in the first sample was low (6 / 12), requiring stricter judgment. The judgment label was determined by taking the intersection, further reducing the pass rate to 3 / 12 (windows #3, #6, and #10 correspond to 21-29 seconds, 51-59 seconds, and 91-99 seconds, respectively). Nine windows in the first sample were still judged as unusable by the intersection, indicating that the 120-second signal was not of consistently high quality and still required further screening using the method described in this embodiment. In the second sample, only one window (#2, corresponding to 11-19 seconds) passed the intersection judgment; this window is exactly M. ac =0.4044 (exceeding the first preset index) and SQI total The window value is 0.1908 (far higher than the second preset index of 0.1217). Figure 13 and Figure 14 Figure (d) shows the extracted target signal segments. The first sample has three usable segments (21–29s, 51–59s, 91–99s), with a total usable time of 24.0 seconds / 120 seconds (20.0%). The second sample has one usable segment (11–19s), with a total usable time of 8.0 seconds / 120 seconds (6.7%). The usable time ratio between the two sets of data (20.0% vs 6.7%) is approximately three times that of Method 1 (based on...). The pass rate of method 1 (based on screening) is 66.7% vs 8.3% ≈ 8 times, while method 2 (based on screening) has a pass rate of 66.7% vs 8.3% ≈ 8 times. The pass rate of the two selected samples was approximately 3 times that of the other two samples (50.0% vs 16.7%). These three indicators corroborated the quality difference between the two selected samples from different perspectives.

[0215] like Figure 4 and Figure 5 In Figure (a), if the first and second samples are placed side by side on the same Y-axis scale, the difference in quality between the two sets of data can be observed with the naked eye even without any algorithm processing. The amplitude of the first sample is uniform and stable, while the second sample shows violent intermittent amplitude fluctuations (the peak value is about 17 times that of the good signal), which reflects the severe impact of wind noise on the signal amplitude. It clearly shows that the selected first sample has better quality, while the second sample has relatively poor quality.

[0216] Figure 15 The power spectral density diagrams of the first and second samples shown in an exemplary embodiment further show that, in the range of 8-18kHz, the power spectral density curve of the first sample is relatively low and flat, which means that there is no strong airflow turbulence interference in this frequency band. The second sample, on the other hand, shows a wide-band "spicule" or "bulge" state, which directly proves that the wind speed is too high during this period and the high frequency band is completely submerged by wind noise.

[0217] Figure 16 The first and second samples are shown according to an exemplary embodiment. and A comparison chart. For example... Figure 16 Figure (a) shows all windows of the first and second samples. The samples are arranged side-by-side, with the first sample overwhelmingly outperforming the second sample in the vast majority of windows. For example... Figure 16 Figure (b) shows all windows of the first and second samples. The dashed lines, arranged side-by-side and differing in size, represent the second preset index of the first sample (0.0872) and the second preset index of the second sample (0.1217). In the first sample... There are more windows with a value greater than or equal to 0.0872.

[0218] Figure 17 The first and second samples shown according to an exemplary embodiment are based on and A comparison chart of the selected signal segments. (e.g.) Figure 17 In Figure (a), the M1 pass count (8) of the first sample is approximately 8 times the M1 pass count (1) of the second sample. The M1 pass count refers to... The number of signal segments greater than or equal to the first preset index; the M2 pass count (6) of the first sample is approximately 3 times the M2 pass count (2) of the second sample; the M2 pass count refers to... The number of signal segments greater than or equal to the second preset index; the intersection pass count of the first sample is approximately three times the intersection pass count of the second sample. The intersection pass count refers to... Greater than or equal to the first preset index and The number of signal segments that are greater than or equal to the second preset index. For example... Figure 17 In Figure (b), the signal availability of the first sample is 20.0% (24.0 seconds), while the signal availability of the second sample is only 6.7% (8.0 seconds). The signal availability is the percentage of the duration of the target signal segment to the duration of the acoustic signal.

[0219] The RMS of the second sample (0.003078) is approximately 6.9 times that of the first sample (0.000446), but the available time ratio of the second sample (6.7%) is only one-third of that of the first sample (20.0%). This indicates that a high RMS does not necessarily represent high signal quality, quantitatively proving that the traditional energy threshold method based on RMS is completely ineffective in the wind noise scenario of this embodiment (a higher RMS actually means stronger wind noise). In addition, the crest factor of the second sample (19.52) is 2.5 times that of the first sample (7.93). A high crest factor reflects the presence of sudden high-amplitude events in the acoustic signal (sudden wind noise), rather than high acoustic signal quality.

[0220] Figure 18This is a flowchart illustrating a fault early warning method according to an exemplary embodiment, see [link to flowchart]. Figure 18 In this embodiment, the method is executed by an electronic device, and the method includes the following steps: In step 1801, multiple signal segments of the acoustic signal of the wind turbine are acquired.

[0221] The principle of this step is the same as that in the above embodiments, and will not be repeated here.

[0222] In step 1802, based on the envelope signal of each signal segment, the time-domain autocorrelation index of each signal segment is obtained. The envelope signal indicates the curve of amplitude change with time at each sampling moment in the signal segment, and the time-domain autocorrelation index indicates the periodicity of the envelope signal.

[0223] In some embodiments, step 1802 includes: obtaining multiple autocorrelation indices between the envelope signal of each signal segment and the corresponding multiple first hysteresis signals, wherein the multiple first hysteresis signals refer to envelope signals that lag behind the envelope signal at different sampling times, and the multiple autocorrelation indices indicate the degree of similarity between the envelope signal and the corresponding first hysteresis signals; and outputting the maximum value among the multiple autocorrelation indices of each signal segment as the time-domain autocorrelation index of each signal segment.

[0224] In step 1803, based on the frequency domain characteristics and time domain autocorrelation index of the envelope signal of each signal segment, the frequency domain quality index of each signal segment is obtained. The frequency domain quality index indicates the frequency domain quality of the signal segment under the time domain autocorrelation index.

[0225] In some embodiments, step 1803 includes: obtaining a frequency domain quality index of the first signal segment based on the frequency domain characteristics and time domain autocorrelation index of the envelope signal of the first signal segment, wherein the time domain autocorrelation index of the first signal segment is greater than or equal to a corresponding preset threshold.

[0226] In some embodiments, step 1803 includes: obtaining a frequency domain fusion quality index for each signal segment based on at least two frequency domain features of the envelope signal of each signal segment, wherein the at least two frequency domain features refer to different frequency domain features obtained based on each envelope signal through different extraction methods, and the frequency domain fusion quality index indicates the frequency domain quality of the signal segment; adjusting the frequency domain fusion quality index based on the time domain autocorrelation index of each signal segment to obtain the frequency domain quality index of each signal segment.

[0227] In some embodiments, the frequency domain characteristics include at least two of spectral kurtosis, spectral entropy, and harmonic regularity factor. Spectral kurtosis indicates the sharpness of the energy distribution of the frequency domain signal corresponding to the envelope signal, spectral entropy indicates the concentration of the energy distribution of the frequency domain signal corresponding to the envelope signal, and harmonic regularity factor indicates the periodicity of the frequency domain signal corresponding to the envelope signal.

[0228] In some embodiments, the method further includes: performing a fast Fourier transform on the amplitude at each sampling time in the envelope signal to obtain a frequency domain signal, wherein the frequency domain signal indicates the amplitude intensity of signals at different frequencies in the envelope signal; acquiring multiple frequency domain autocorrelation indices between the frequency domain signal and multiple second hysteresis signals at different preset fundamental frequencies, wherein the preset fundamental frequency indicates the rotation frequency of the wind turbine, and the multiple second hysteresis signals refer to frequency domain signals that lag behind the frequency domain signal by different integer multiples of the preset fundamental frequency, and the multiple frequency domain autocorrelation indices indicate the similarity between the frequency domain signal and the multiple second hysteresis signals; acquiring the product between the multiple frequency domain autocorrelation indices corresponding to each preset fundamental frequency to obtain the autocorrelation product of each preset fundamental frequency, and outputting the maximum value among the multiple autocorrelation products as the harmonic regularity factor of the envelope signal.

[0229] In some embodiments, the method further includes: squaring the amplitude at all sampling times in each signal segment to obtain a first instantaneous relative power at each sampling time; removing the DC component of the first instantaneous relative power to obtain a second instantaneous relative power; and performing low-pass filtering based on the second instantaneous relative power at each sampling time to obtain the envelope signal of each signal segment.

[0230] In some embodiments, the method further includes: downsampling the envelope signal to obtain a target envelope signal; obtaining a time-domain autocorrelation index for each signal segment based on the envelope signal of each signal segment, including: obtaining a time-domain autocorrelation index for each signal segment based on the target envelope signal of each signal segment; and obtaining a frequency-domain quality index for each signal segment based on the frequency-domain characteristics and time-domain autocorrelation index of the envelope signal of each signal segment, including: obtaining a frequency-domain quality index for each signal segment based on the frequency-domain characteristics and time-domain autocorrelation index of the target envelope signal of each signal segment.

[0231] In some embodiments, the method further includes performing at least one of the following preprocessing steps on the acoustic signal: removing the DC component of the acoustic signal; and bandpass filtering the acoustic signal, wherein the passband of the bandpass filter covers the frequency band of the sound signal generated by the blades of the wind turbine.

[0232] In step 1804, target signal segments are extracted from the acoustic signal based on the time-domain autocorrelation index and frequency-domain quality index of each signal segment. At least one of the time-domain autocorrelation index and frequency-domain quality index of the target signal segment is greater than or equal to the corresponding preset threshold.

[0233] In some embodiments, step 1804 includes: extracting a signal segment from the acoustic signal that satisfies both a first condition and a second condition as a target signal segment, wherein the first condition is that the time-domain autocorrelation index is greater than or equal to a first preset index, and the second condition is that the frequency-domain quality index is greater than or equal to a second preset index; or, extracting a signal segment from the acoustic signal that satisfies both the first and second conditions as a target signal segment; or, performing a weighted summation of the time-domain autocorrelation index and the frequency-domain quality index of each signal segment to obtain a comprehensive quality index of each signal segment, and extracting a signal segment from the acoustic signal whose comprehensive quality index is greater than or equal to a preset threshold as a target signal segment, wherein the comprehensive quality index indicates the comprehensive quality of the signal segment in the frequency and time domains.

[0234] In some embodiments, the method further includes: obtaining the first type of inter-class variance of the frequency domain quality index of each signal segment; and outputting the frequency domain quality index corresponding to the largest first type of inter-class variance among the first type of inter-class variances of each signal segment as a second preset index.

[0235] In some embodiments, the method further includes: outputting a preset value as a first preset index; or, sorting the time-domain autocorrelation indices of each signal segment in ascending order to obtain an ordered dataset, and outputting the quantile of a preset percentage of the ordered dataset as the first preset index; or, obtaining the second-class variance of the time-domain autocorrelation indices of each signal segment, and outputting the time-domain autocorrelation index corresponding to the largest second-class variance among the second-class variances of each signal segment as the first preset index.

[0236] In some embodiments, step 1804 includes: extending each second signal segment in the acoustic signal forward and backward by a preset time period to obtain a third signal segment, wherein the second signal segment is a signal segment in which at least one of the time-domain autocorrelation index and the frequency-domain quality index is less than a corresponding preset threshold; and extracting target signal segments in the acoustic signal other than the third signal segment.

[0237] In step 1805, fault identification is performed based on the target signal segment. If the identification result indicates that a fault exists, an early warning is issued.

[0238] In this step, fault identification refers to determining whether there is a fault in the wind turbine blades based on the target signal segment, and obtaining the identification result. Faults include cracks, breaks, leading edge corrosion, chipping, icing, imbalance (loosening or foreign matter attached), etc. The identification result may be that the wind turbine blades are in normal condition (no fault), or the identification result may be that the wind turbine blades have some kind of fault.

[0239] In some embodiments, fault identification based on a target signal segment includes: inputting the target signal segment into a fault identification model and obtaining the identification result output by the fault identification model. The fault identification model can be an SVM (Support Vector Machine), a deep learning identification model, etc.

[0240] When the identification result indicates a fault, an early warning is issued. The early warning methods include various methods, such as displaying a warning message indicating a fault in the wind turbine blades, issuing an audible and visual alarm (controlling a buzzer to sound and a warning light to flash), or sending a message (using SMS, email, etc. to send the identification result to the maintenance personnel's terminal), etc. The above are just examples of early warning methods. The early warning methods can also be other methods, and this disclosure does not specifically limit them.

[0241] The method of this disclosure identifies faults based on extracted target signal segments. Since these target signal segments are less affected by wind noise and other disturbances, false anomalies not originating from the blades are eliminated at the source, significantly improving the accuracy of fault identification and reducing false alarms and missed alarms at the root. Furthermore, the target signal segments are stable blade periodic signals with a more uniform distribution of fault characteristics, allowing the fault identification model to more accurately capture defect patterns, significantly improving identification accuracy and thus enhancing the reliability of early warnings.

[0242] Figure 19 This is a block diagram illustrating an apparatus for extracting signal segments from an acoustic signal according to an exemplary embodiment. The apparatus includes: The first acquisition module 1901 is configured to acquire multiple signal segments of the acoustic signal of the wind turbine. The second acquisition module 1902 is configured to acquire the time-domain autocorrelation index of each signal segment based on the envelope signal of each signal segment. The envelope signal indicates the curve of amplitude change with time at each sampling moment in the signal segment, and the time-domain autocorrelation index indicates the periodicity of the envelope signal. The third acquisition module 1903 is configured to acquire the frequency domain quality index of each signal segment based on the frequency domain characteristics of the envelope signal of each signal segment and the time domain autocorrelation index. The frequency domain quality index indicates the frequency domain quality of the signal segment under the time domain autocorrelation index. The first extraction module 1904 is configured to extract target signal segments from the acoustic signal based on the time-domain autocorrelation index and frequency-domain quality index of each signal segment, wherein at least one of the time-domain autocorrelation index and the frequency-domain quality index of the target signal segment is greater than or equal to a corresponding preset threshold.

[0243] In some embodiments, the second acquisition module is configured to acquire multiple autocorrelation indices between the envelope signal of each signal segment and the corresponding plurality of first hysteresis signals, wherein the plurality of first hysteresis signals refer to the envelope signal that lags behind the envelope signal by different sampling times, and the plurality of autocorrelation indices indicate the degree of similarity between the envelope signal and the corresponding first hysteresis signals; and output the maximum value among the plurality of autocorrelation indices of each signal segment as the time-domain autocorrelation index of each signal segment.

[0244] In some embodiments, the third acquisition module is configured to acquire a frequency domain quality index of the first signal segment based on the frequency domain characteristics of the envelope signal of the first signal segment and the time domain autocorrelation index, wherein the time domain autocorrelation index of the first signal segment is greater than or equal to a corresponding preset threshold.

[0245] In some embodiments, the third acquisition module is configured to acquire a frequency domain fusion quality index for each signal segment based on at least two frequency domain features of the envelope signal of each signal segment. The at least two frequency domain features refer to different frequency domain features obtained by different extraction methods based on each envelope signal. The frequency domain fusion quality index indicates the frequency domain quality of the signal segment. The frequency domain fusion quality index is adjusted based on the time domain autocorrelation index of each signal segment to obtain the frequency domain quality index of each signal segment.

[0246] In some embodiments, the frequency domain features include at least two of spectral kurtosis, spectral entropy, and harmonic regularity factor. The spectral kurtosis indicates the sharpness of the energy distribution of the frequency domain signal corresponding to the envelope signal, the spectral entropy indicates the concentration of the energy distribution of the frequency domain signal corresponding to the envelope signal, and the harmonic regularity factor indicates the periodicity of the frequency domain signal corresponding to the envelope signal.

[0247] In some embodiments, the apparatus further includes a seventh acquisition module: the seventh acquisition module is configured to perform a fast Fourier transform on the amplitude at each sampling time in the envelope signal to obtain a frequency domain signal, the frequency domain signal indicating the amplitude intensity of different frequency signals in the envelope signal; at different preset fundamental frequencies, acquire multiple frequency domain autocorrelation indices between the frequency domain signal and multiple second hysteresis signals, the preset fundamental frequency indicating the rotation frequency of the wind turbine, the multiple second hysteresis signals referring to the frequency domain signals that lag behind the frequency domain signal by different integer multiples of the preset fundamental frequency, the multiple frequency domain autocorrelation indices indicating the similarity between the frequency domain signal and the multiple second hysteresis signals respectively; acquire the product of the multiple frequency domain autocorrelation indices corresponding to each preset fundamental frequency to obtain the autocorrelation product of each preset fundamental frequency, and output the maximum value among the multiple autocorrelation products as the harmonic regularity factor of the envelope signal.

[0248] In some embodiments, the first extraction module is configured to extract signal segments from the acoustic signal that satisfy both a first condition and a second condition as the target signal segment, wherein the first condition is that the time-domain autocorrelation index is greater than or equal to a first preset index, and the second condition is that the frequency-domain quality index is greater than or equal to a second preset index; or, to extract signal segments from the acoustic signal that satisfy both the first and second conditions as the target signal segment; or, to perform a weighted summation of the time-domain autocorrelation index and the frequency-domain quality index of each signal segment to obtain a comprehensive quality index of each signal segment, and to extract signal segments from the acoustic signal whose comprehensive quality index is greater than or equal to a preset threshold as the target signal segment, wherein the comprehensive quality index indicates the comprehensive quality of the signal segment in the frequency and time domains.

[0249] In some embodiments, the apparatus further includes an eighth acquisition module: the eighth acquisition module is configured to acquire the first type variance of the frequency domain quality index of each signal segment; and output the frequency domain quality index corresponding to the largest first type variance among the first type variances of each signal segment as a second preset index.

[0250] In some embodiments, the device further includes a ninth acquisition module: the ninth acquisition module is configured to output a preset value as the first preset index; or, to sort the time-domain autocorrelation indices of each signal segment in ascending order to obtain an ordered dataset, and output the quantile of a preset percentage of the ordered dataset as the first preset index; or, to acquire the second-type variance of the time-domain autocorrelation indices of each signal segment, and output the time-domain autocorrelation index corresponding to the largest second-type variance among the second-type variances of each signal segment as the first preset index.

[0251] In some embodiments, the apparatus further includes a tenth acquisition module: the tenth acquisition module is configured to square the amplitude of all sampling times in each signal segment to obtain a first instantaneous relative power quantity at each sampling time; remove the DC component of the first instantaneous relative power quantity to obtain a second instantaneous relative power quantity; and perform low-pass filtering based on the second instantaneous relative power quantity at each sampling time to obtain the envelope signal of each signal segment.

[0252] In some embodiments, the apparatus further includes a first downsampling module: the first downsampling module is configured to downsample the envelope signal to obtain a target envelope signal; the step of obtaining the time-domain autocorrelation index of each signal segment based on the envelope signal of each signal segment includes: obtaining the time-domain autocorrelation index of each signal segment based on the target envelope signal of each signal segment; the step of obtaining the frequency-domain quality index of each signal segment based on the frequency-domain features of the envelope signal of each signal segment and the time-domain autocorrelation index includes: obtaining the frequency-domain quality index of each signal segment based on the frequency-domain features of the target envelope signal of each signal segment and the time-domain autocorrelation index.

[0253] In some embodiments, the apparatus further includes a first preprocessing module: the first preprocessing module is configured to perform at least one of the following preprocessing steps on the acoustic signal: removing the DC component of the acoustic signal; and performing bandpass filtering on the acoustic signal, the passband of the bandpass filtering covering the frequency band of the sound signal generated by the blades of the wind turbine.

[0254] In some embodiments, the first extraction module is configured to extend each second signal segment in the acoustic signal forward and backward by a preset time period to obtain a third signal segment, wherein the second signal segment is a signal segment in which at least one of the time-domain autocorrelation index and the frequency-domain quality index is less than a corresponding preset threshold; and to extract target signal segments in the acoustic signal other than the third signal segment.

[0255] In some embodiments, the apparatus further includes a first value acquisition module; the first value acquisition module is configured to acquire a quality score for each signal segment based on the envelope signal of each signal segment, the quality score indicating the quality of the signal segment; the extraction of a target signal segment from the acoustic signal based on the time-domain autocorrelation index and the frequency-domain quality index of each signal segment includes: extracting a target signal segment from the acoustic signal based on the time-domain autocorrelation index, the frequency-domain quality index, and the quality score of each signal segment, wherein at least one of the time-domain autocorrelation index, the frequency-domain quality index, and the quality score of the target signal segment is greater than or equal to a corresponding preset threshold.

[0256] It should be noted that the signal segment extraction device for acoustic signals provided in the above embodiments is only illustrated by the division of the above functional units. In practical applications, the above functions can be assigned to different functional units as needed, that is, the internal structure of the electronic device can be divided into different functional units to complete all or part of the functions described above. In addition, the signal segment extraction device for acoustic signals and the method embodiment for extracting signal segments from acoustic signals provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiment, which will not be repeated here.

[0257] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0258] Figure 20 This is a block diagram illustrating a fault warning device according to an exemplary embodiment, the device comprising: The fourth acquisition module 2001 is configured to acquire multiple signal segments of the acoustic signal of the wind turbine. The fifth acquisition module 2002 is configured to acquire the time-domain autocorrelation index of each signal segment based on the envelope signal of each signal segment. The envelope signal indicates the curve of amplitude change with time at each sampling moment in the signal segment, and the time-domain autocorrelation index indicates the periodicity of the envelope signal. The sixth acquisition module 2003 is configured to acquire the frequency domain quality index of each signal segment based on the frequency domain characteristics of the envelope signal of each signal segment and the time domain autocorrelation index. The frequency domain quality index indicates the frequency domain quality of the signal segment under the time domain autocorrelation index. The second extraction module 2004 is configured to extract target signal segments from the acoustic signal based on the time-domain autocorrelation index and frequency-domain quality index of each signal segment, wherein at least one of the time-domain autocorrelation index and the frequency-domain quality index of the target signal segment is greater than or equal to a corresponding preset threshold. The identification module 2005 is configured to perform fault identification based on the target signal segment, and to issue an early warning if the identification result indicates that a fault exists.

[0259] In some embodiments, the fifth acquisition module is configured to acquire multiple autocorrelation indices between the envelope signal of each signal segment and the corresponding multiple first hysteresis signals, wherein the multiple first hysteresis signals refer to the envelope signals that lag behind the envelope signals at different sampling times, and the multiple autocorrelation indices indicate the similarity between the envelope signals and the corresponding first hysteresis signals; and output the maximum value among the multiple autocorrelation indices of each signal segment as the time-domain autocorrelation index of each signal segment.

[0260] In some embodiments, the sixth acquisition module is configured to acquire a frequency domain quality index of the first signal segment based on the frequency domain characteristics of the envelope signal of the first signal segment and the time domain autocorrelation index, wherein the time domain autocorrelation index of the first signal segment is greater than or equal to a corresponding preset threshold.

[0261] In some embodiments, the sixth acquisition module is configured to acquire a frequency domain fusion quality index for each signal segment based on at least two frequency domain features of the envelope signal of each signal segment. The at least two frequency domain features refer to different frequency domain features obtained by different extraction methods based on each envelope signal. The frequency domain fusion quality index indicates the frequency domain quality of the signal segment. The frequency domain fusion quality index is adjusted based on the time domain autocorrelation index of each signal segment to obtain the frequency domain quality index of each signal segment.

[0262] In some embodiments, the frequency domain features include at least two of spectral kurtosis, spectral entropy, and harmonic regularity factor. The spectral kurtosis indicates the sharpness of the energy distribution of the frequency domain signal corresponding to the envelope signal, the spectral entropy indicates the concentration of the energy distribution of the frequency domain signal corresponding to the envelope signal, and the harmonic regularity factor indicates the periodicity of the frequency domain signal corresponding to the envelope signal.

[0263] In some embodiments, the device further includes an eleventh acquisition module: the eleventh acquisition module is configured to perform a fast Fourier transform on the amplitude at each sampling time in the envelope signal to obtain a frequency domain signal, the frequency domain signal indicating the amplitude intensity of different frequency signals in the envelope signal; at different preset fundamental frequencies, acquire multiple frequency domain autocorrelation indices between the frequency domain signal and multiple second hysteresis signals, the preset fundamental frequency indicating the rotation frequency of the wind turbine, the multiple second hysteresis signals referring to the frequency domain signals that lag behind the frequency domain signal by different integer multiples of the preset fundamental frequency, the multiple frequency domain autocorrelation indices indicating the similarity between the frequency domain signal and the multiple second hysteresis signals respectively; acquire the product between the multiple frequency domain autocorrelation indices corresponding to each preset fundamental frequency to obtain the autocorrelation product of each preset fundamental frequency, and output the maximum value among the multiple autocorrelation products as the harmonic regularity factor of the envelope signal.

[0264] In some embodiments, the second extraction module is configured to extract signal segments from the acoustic signal that satisfy both a first condition and a second condition as the target signal segment, wherein the first condition is that the time-domain autocorrelation index is greater than or equal to a first preset index, and the second condition is that the frequency-domain quality index is greater than or equal to a second preset index; or, to extract signal segments from the acoustic signal that satisfy both the first and second conditions as the target signal segment; or, to perform a weighted summation of the time-domain autocorrelation index and the frequency-domain quality index of each signal segment to obtain a comprehensive quality index of each signal segment, and to extract signal segments from the acoustic signal whose comprehensive quality index is greater than or equal to a preset threshold as the target signal segment, wherein the comprehensive quality index indicates the comprehensive quality of the signal segment in the frequency and time domains.

[0265] In some embodiments, the apparatus further includes a twelfth acquisition module: the twelfth acquisition module is configured to acquire the first type variance of the frequency domain quality index of each signal segment; and output the frequency domain quality index corresponding to the largest first type variance among the first type variances of each signal segment as a second preset index.

[0266] In some embodiments, the device further includes a thirteenth acquisition module: the thirteenth acquisition module is configured to output a preset value as the first preset index; or, to sort the time-domain autocorrelation indices of each signal segment in ascending order to obtain an ordered dataset, and output the quantile of a preset percentage of the ordered dataset as the first preset index; or, to acquire the second-type variance of the time-domain autocorrelation indices of each signal segment, and output the time-domain autocorrelation index corresponding to the largest second-type variance among the second-type variances of each signal segment as the first preset index.

[0267] In some embodiments, the apparatus further includes a fourteenth acquisition module: the fourteenth acquisition module is configured to square the amplitude of all sampling times in each signal segment to obtain a first instantaneous relative power quantity at each sampling time; remove the DC component of the first instantaneous relative power quantity to obtain a second instantaneous relative power quantity; and perform low-pass filtering based on the second instantaneous relative power quantity at each sampling time to obtain the envelope signal of each signal segment.

[0268] In some embodiments, the apparatus further includes a second downsampling module: the second downsampling module is configured to downsample the envelope signal to obtain a target envelope signal; the step of obtaining the time-domain autocorrelation index of each signal segment based on the envelope signal of each signal segment includes: obtaining the time-domain autocorrelation index of each signal segment based on the target envelope signal of each signal segment; the step of obtaining the frequency-domain quality index of each signal segment based on the frequency-domain features of the envelope signal of each signal segment and the time-domain autocorrelation index includes: obtaining the frequency-domain quality index of each signal segment based on the frequency-domain features of the target envelope signal of each signal segment and the time-domain autocorrelation index.

[0269] In some embodiments, the apparatus further includes a second preprocessing module: the second preprocessing module is configured to perform at least one of the following preprocessing steps on the acoustic signal: removing the DC component of the acoustic signal; and performing bandpass filtering on the acoustic signal, the passband range of the bandpass filtering covering the frequency band of the sound signal generated by the blades of the wind turbine.

[0270] In some embodiments, the second extraction module is configured to extend each second signal segment in the acoustic signal forward and backward by a preset time period to obtain a third signal segment, wherein the second signal segment is a signal segment in which at least one of the time-domain autocorrelation index and the frequency-domain quality index is less than a corresponding preset threshold; and to extract target signal segments in the acoustic signal other than the third signal segment.

[0271] In some embodiments, the apparatus further includes a second value acquisition module; the second value acquisition module is configured to acquire a quality score for each signal segment based on the envelope signal of each signal segment, the quality score indicating the quality of the signal segment; the extraction of a target signal segment from the acoustic signal based on the time-domain autocorrelation index and the frequency-domain quality index of each signal segment includes: extracting a target signal segment from the acoustic signal based on the time-domain autocorrelation index, the frequency-domain quality index, and the quality score of each signal segment, wherein at least one of the time-domain autocorrelation index, the frequency-domain quality index, and the quality score of the target signal segment is greater than or equal to a corresponding preset threshold.

[0272] It should be noted that the fault warning device provided in the above embodiments is only illustrated by the division of the above functional units when providing fault warnings. In practical applications, the above functions can be assigned to different functional units as needed, that is, the internal structure of the electronic device can be divided into different functional units to complete all or part of the functions described above. In addition, the fault warning device and the fault warning method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0273] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0274] When an electronic device is provided as a terminal, Figure 21 This is a block diagram illustrating a terminal according to an exemplary embodiment. The diagram shows a structural block diagram of a terminal 2100 provided in an exemplary embodiment of this disclosure. The terminal 2100 may be a smartphone, tablet computer, MP3 player (Moving Picture Experts Group Audio Layer III), MP4 player (Moving Picture Experts Group Audio Layer IV), laptop computer, or desktop computer. The terminal 2100 may also be referred to as a user device, portable terminal, laptop terminal, desktop terminal, or other names.

[0275] Typically, terminal 2100 includes a processor 2101 and a memory 2102.

[0276] Processor 2101 may include one or more processing cores, such as a quad-core processor, an octa-core processor, etc. Processor 2101 may be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). Processor 2101 may also include a main processor and a coprocessor. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, processor 2101 may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content required to be displayed on the screen. In some embodiments, processor 2101 may also include an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.

[0277] The memory 2102 may include one or more computer-readable storage media, which may be non-transitory. The memory 2102 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In some embodiments, the non-transitory computer-readable storage media in the memory 2102 are used to store at least one computer program, which is executed by the processor 2101 to implement the method provided in the method embodiments of this application.

[0278] In some embodiments, the terminal 2100 may also optionally include a peripheral device interface 2103 and at least one peripheral device. The processor 2101, memory 2102, and peripheral device interface 2103 can be connected via a bus or signal line. Each peripheral device can be connected to the peripheral device interface 2103 via a bus, signal line, or circuit board. Specifically, the peripheral device includes at least one of the following: a radio frequency circuit 2104, a display screen 2105, a camera assembly 2106, an audio circuit 2107, and a power supply 2108.

[0279] Peripheral device interface 2103 can be used to connect at least one I / O (Input / Output) related peripheral device to processor 2101 and memory 2102. In some embodiments, processor 2101, memory 2102 and peripheral device interface 2103 are integrated on the same chip or circuit board; in some other embodiments, any one or two of processor 2101, memory 2102 and peripheral device interface 2103 can be implemented on separate chips or circuit boards, which is not limited in this embodiment.

[0280] The radio frequency (RF) circuit 2104 is used to receive and transmit RF (Radio Frequency) signals, also known as electromagnetic signals. The RF circuit 2104 communicates with communication networks and other communication devices via electromagnetic signals. The RF circuit 2104 converts electrical signals into electromagnetic signals for transmission, or converts received electromagnetic signals back into electrical signals. In some embodiments, the RF circuit 2104 includes: an antenna system, an RF transceiver, one or more amplifiers, a tuner, an oscillator, a digital signal processor, a codec chipset, a user identity module card, etc. The RF circuit 2104 can communicate with other terminals through at least one wireless communication protocol. This wireless communication protocol includes, but is not limited to: the World Wide Web, metropolitan area networks, intranets, various generations of mobile communication networks (2G, 3G, 4G, and 5G), wireless local area networks, and / or WiFi (Wireless Fidelity) networks. In some embodiments, the RF circuit 2104 may also include circuitry related to NFC (Near Field Communication), which is not limited in this application.

[0281] Display screen 2105 is used to display a UI (User Interface). This UI may include graphics, text, icons, videos, and any combination thereof. When display screen 2105 is a touch display screen, it also has the ability to collect touch signals on or above its surface. These touch signals can be input as control signals to processor 2101 for processing. In this case, display screen 2105 can also be used to provide virtual buttons and / or a virtual keyboard, also known as soft buttons and / or a soft keyboard. In some embodiments, there may be one display screen 2105, disposed on the front panel of terminal 2100; in other embodiments, there may be at least two display screens, disposed on different surfaces of terminal 2100 or in a folded design; in still other embodiments, display screen 2105 may be a flexible display screen, disposed on a curved or folded surface of terminal 2100. Furthermore, display screen 2105 may be configured as a non-rectangular, irregular shape, i.e., a non-rectangular screen. The display screen 2105 can be made of materials such as LCD (Liquid Crystal Display) and OLED (Organic Light-Emitting Diode).

[0282] The camera assembly 2106 is used to acquire images or videos. In some embodiments, the camera assembly 2106 includes a front-facing camera and a rear-facing camera. Typically, the front-facing camera is located on the front panel of the terminal, and the rear-facing camera is located on the back of the terminal. In some embodiments, there are at least two rear-facing cameras, which are any one of a main camera, a depth-sensing camera, a wide-angle camera, and a telephoto camera, to achieve background blurring by fusion of the main camera and the depth-sensing camera, panoramic shooting by fusion of the main camera and the wide-angle camera, VR (Virtual Reality) shooting, or other fusion shooting functions. In some embodiments, the camera assembly 2106 may also include a flash. The flash can be a single-color temperature flash or a dual-color temperature flash. A dual-color temperature flash refers to a combination of a warm-light flash and a cool-light flash, which can be used for light compensation at different color temperatures.

[0283] The audio circuit 2107 may include a microphone and a speaker. The microphone is used to collect sound waves from the user and the environment, converting them into electrical signals that are input to the processor 2101 for processing, or to the radio frequency circuit 2104 for voice communication. For stereo sound acquisition or noise reduction purposes, multiple microphones may be used, each positioned at a different location on the terminal 2100. The microphone may also be an array microphone or an omnidirectional microphone. The speaker is used to convert electrical signals from the processor 2101 or the radio frequency circuit 2104 into sound waves. The speaker may be a conventional diaphragm speaker or a piezoelectric ceramic speaker. When the speaker is a piezoelectric ceramic speaker, it can convert electrical signals not only into audible sound waves but also into inaudible sound waves for purposes such as distance measurement. In some embodiments, the audio circuit 2107 may also include a headphone jack.

[0284] Power supply 2108 is used to power the various components in terminal 2100. Power supply 2108 can be AC ​​power, DC power, a disposable battery, or a rechargeable battery. When power supply 2108 includes a rechargeable battery, the rechargeable battery can be a wired rechargeable battery or a wireless rechargeable battery. A wired rechargeable battery is a battery that is charged via a wired line, and a wireless rechargeable battery is a battery that is charged via a wireless coil. The rechargeable battery can also be used to support fast charging technology.

[0285] Those skilled in the art will understand that Figure 21 The structure shown does not constitute a limitation on terminal 2100 and may include more or fewer components than shown, or combine certain components, or use different component arrangements.

[0286] The aforementioned electronic devices can also be implemented as servers. The structure of a server is described below: Figure 22This is a schematic diagram of a server structure provided in an embodiment of this application. The server 2200 can vary significantly due to different configurations or performance. It may include one or more Central Processing Units (CPUs) 2201 and one or more memories 2202. The one or more memories 2202 store at least one computer program, which is loaded and executed by the one or more processors 2201 to implement the methods provided in the above-described method embodiments. Of course, the server 2200 may also have wired or wireless network interfaces, a keyboard, and input / output interfaces for input and output. The server 2200 may also include other components for implementing device functions, which will not be elaborated upon here.

[0287] In an exemplary embodiment, a computer-readable storage medium is also provided, such as a memory including a computer program that can be executed by a processor to perform the methods in the above embodiments. For example, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, floppy disk, and optical data storage device, etc.

[0288] In an exemplary embodiment, a computer program product or computer program is also provided, which includes program code stored in a computer-readable storage medium. A processor of an electronic device reads the program code from the computer-readable storage medium and executes the program code, causing the electronic device to perform the method described above.

[0289] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0290] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A method for extracting signal segments from acoustic signals, characterized in that, The method includes: Acquire multiple signal segments of the acoustic signal from the wind turbine; Based on the envelope signal of each signal segment, the time-domain autocorrelation index of each signal segment is obtained. The envelope signal indicates the curve of amplitude change with time at each sampling moment in the signal segment, and the time-domain autocorrelation index indicates the periodicity of the envelope signal. Based on the frequency domain characteristics of the envelope signal of each signal segment and the time domain autocorrelation index, the frequency domain quality index of each signal segment is obtained, and the frequency domain quality index indicates the frequency domain quality of the signal segment under the time domain autocorrelation index. Based on the time-domain autocorrelation index and frequency-domain quality index of each signal segment, a target signal segment is extracted from the acoustic signal. At least one of the time-domain autocorrelation index and the frequency-domain quality index of the target signal segment is greater than or equal to a corresponding preset threshold.

2. The method for extracting signal segments from acoustic signals according to claim 1, characterized in that, The process of obtaining the time-domain autocorrelation index of each signal segment based on the envelope signal of each signal segment includes: Multiple autocorrelation indices are obtained between the envelope signal of each signal segment and the corresponding multiple first hysteresis signals. The multiple first hysteresis signals refer to the envelope signal that lags behind the envelope signal at different sampling times. The multiple autocorrelation indices indicate the degree of similarity between the envelope signal and the corresponding first hysteresis signals. The maximum value among multiple autocorrelation indices for each signal segment is output as the time-domain autocorrelation index for each signal segment.

3. The method for extracting signal segments from acoustic signals according to claim 1, characterized in that, The process of obtaining the frequency domain quality index of each signal segment based on the frequency domain characteristics of the envelope signal of each signal segment and the time domain autocorrelation index includes: Based on the frequency domain characteristics of the envelope signal of the first signal segment and the time domain autocorrelation index, the frequency domain quality index of the first signal segment is obtained, and the time domain autocorrelation index of the first signal segment is greater than or equal to the corresponding preset threshold.

4. The method for extracting signal segments from acoustic signals according to claim 1, characterized in that, The process of obtaining the frequency domain quality index of each signal segment based on the frequency domain characteristics of the envelope signal of each signal segment and the time domain autocorrelation index includes: Based on at least two frequency domain features of the envelope signal of each signal segment, a frequency domain fusion quality index is obtained for each signal segment. The at least two frequency domain features refer to different frequency domain features obtained by different extraction methods based on each envelope signal. The frequency domain fusion quality index indicates the frequency domain quality of the signal segment. The frequency domain fusion quality index is adjusted based on the time domain autocorrelation index of each signal segment to obtain the frequency domain quality index of each signal segment.

5. The method for extracting signal segments from acoustic signals according to claim 4, characterized in that, The frequency domain features include at least two of spectral kurtosis, spectral entropy, and harmonic regularity factor. The spectral kurtosis indicates the sharpness of the energy distribution of the frequency domain signal corresponding to the envelope signal, the spectral entropy indicates the concentration of the energy distribution of the frequency domain signal corresponding to the envelope signal, and the harmonic regularity factor indicates the periodicity of the frequency domain signal corresponding to the envelope signal.

6. The method for extracting signal segments from acoustic signals according to claim 5, characterized in that, The method further includes: A fast Fourier transform is performed on the amplitude at each sampling time in the envelope signal to obtain a frequency domain signal, which indicates the amplitude intensity of signals at different frequencies in the envelope signal. Under different preset base frequencies, multiple frequency domain autocorrelation indices are obtained between the frequency domain signal and multiple second hysteresis signals respectively. The preset base frequency indicates the rotation frequency of the wind turbine. The multiple second hysteresis signals refer to the frequency domain signals that lag behind the frequency domain signal by different integer multiples of the preset base frequency respectively. The multiple frequency domain autocorrelation indices indicate the similarity between the frequency domain signal and the multiple second hysteresis signals respectively. The product of multiple frequency domain autocorrelation indices corresponding to each preset fundamental frequency is obtained to obtain the autocorrelation product of each preset fundamental frequency. The maximum value among the multiple autocorrelation products is output as the harmonic regularity factor of the envelope signal.

7. The method for extracting signal segments from acoustic signals according to claim 1, characterized in that, The extraction of the target signal segment from the acoustic signal includes: From the acoustic signal, signal segments that satisfy both a first condition and a second condition are extracted as the target signal segments. The first condition is that the time-domain autocorrelation index is greater than or equal to a first preset index, and the second condition is that the frequency-domain quality index is greater than or equal to a second preset index; or... From the acoustic signal, the signal segments that satisfy the first condition and the signal segments that satisfy the second condition are extracted as the target signal segments; or, The time-domain autocorrelation index and frequency-domain quality index of each signal segment are weighted and summed to obtain the comprehensive quality index of each signal segment. From the acoustic signal, the signal segments with a comprehensive quality index greater than or equal to a preset threshold are extracted as the target signal segments. The comprehensive quality index indicates the comprehensive quality of the signal segment in the frequency domain and time domain.

8. The method for extracting signal segments from acoustic signals according to claim 7, characterized in that, The method further includes: Obtain the first-order inter-type variance of the frequency domain quality index for each signal segment; The frequency domain quality index corresponding to the largest inter-type variance among the inter-type variances of each signal segment is output as the second preset index.

9. The method for extracting signal segments from acoustic signals according to claim 7, characterized in that, The method further includes: Output the preset value as the first preset index; or... The time-domain autocorrelation indices of each signal segment are sorted in ascending order to obtain an ordered dataset. The quantiles of a preset percentage of the ordered dataset are then output as the first preset index; or... Obtain the second type variance of the time-domain autocorrelation index for each signal segment, and output the time-domain autocorrelation index corresponding to the largest second type variance among the second type variances of each signal segment as the first preset index.

10. The method for extracting signal segments from acoustic signals according to claim 1, characterized in that, The method further includes: The amplitude at each sampling time in each signal segment is squared to obtain the first instantaneous relative power at each sampling time. Remove the DC component of the first instantaneous relative power to obtain the second instantaneous relative power. Based on the second instantaneous relative power at each sampling time, low-pass filtering is performed to obtain the envelope signal of each signal segment.

11. The method for extracting signal segments from acoustic signals according to claim 10, characterized in that, The method further includes: The target envelope signal is obtained by downsampling the envelope signal. The process of obtaining the time-domain autocorrelation index of each signal segment based on the envelope signal of each signal segment includes: Based on the target envelope signal of each signal segment, the time-domain autocorrelation index of each signal segment is obtained; The process of obtaining the frequency domain quality index of each signal segment based on the frequency domain characteristics of the envelope signal of each signal segment and the time domain autocorrelation index includes: Based on the frequency domain characteristics of the target envelope signal of each signal segment and the time domain autocorrelation index, the frequency domain quality index of each signal segment is obtained.

12. The method for extracting signal segments from acoustic signals according to claim 1, characterized in that, The method further includes performing at least one of the following preprocessing steps on the acoustic signal: Remove the DC component of the acoustic signal; The acoustic signal is subjected to bandpass filtering, and the passband range of the bandpass filter covers the frequency band of the sound signal generated by the blades of the wind turbine.

13. The method for extracting signal segments from acoustic signals according to claim 1, characterized in that, The extraction of the target signal segment from the acoustic signal includes: Each second signal segment in the acoustic signal is extended forward and backward by a preset time period to obtain a third signal segment. The second signal segment is a signal segment in which at least one of the time domain autocorrelation index and the frequency domain quality index is less than the corresponding preset threshold. Extract the target signal segment from the acoustic signal, excluding the third signal segment.

14. The method for extracting signal segments from acoustic signals according to claim 1, characterized in that, The method further includes: Based on the envelope signal of each signal segment, the quality score of each signal segment is obtained, and the quality score indicates the quality of the signal segment; The extraction of target signal segments from the acoustic signal based on the time-domain autocorrelation index and frequency-domain quality index of each signal segment includes: Based on the time-domain autocorrelation index, frequency-domain quality index, and quality score of each signal segment, a target signal segment is extracted from the acoustic signal. At least one of the time-domain autocorrelation index, frequency-domain quality index, and quality score of the target signal segment is greater than or equal to a corresponding preset threshold.

15. A fault early warning method, characterized in that, The method includes: Acquire multiple signal segments of the acoustic signal from the wind turbine; Based on the envelope signal of each signal segment, the time-domain autocorrelation index of each signal segment is obtained. The envelope signal indicates the curve of amplitude change with time at each sampling moment in the signal segment, and the time-domain autocorrelation index indicates the periodicity of the envelope signal. Based on the frequency domain characteristics of the envelope signal of each signal segment and the time domain autocorrelation index, the frequency domain quality index of each signal segment is obtained, and the frequency domain quality index indicates the frequency domain quality of the signal segment under the time domain autocorrelation index. Based on the time-domain autocorrelation index and frequency-domain quality index of each signal segment, the target signal segment in the acoustic signal is extracted, and at least one of the time-domain autocorrelation index and the frequency-domain quality index of the target signal segment is greater than or equal to the corresponding preset threshold. Fault identification is performed based on the target signal segment, and if the identification result indicates that a fault exists, an early warning is issued.

16. A device for extracting signal segments from acoustic signals, characterized in that, The device includes: The first acquisition module is configured to acquire multiple signal segments of the acoustic signal of the wind turbine. The second acquisition module is configured to acquire the time-domain autocorrelation index of each signal segment based on the envelope signal of each signal segment. The envelope signal indicates the curve of amplitude change with time at each sampling moment in the signal segment, and the time-domain autocorrelation index indicates the periodicity of the envelope signal. The third acquisition module is configured to acquire the frequency domain quality index of each signal segment based on the frequency domain characteristics of the envelope signal of each signal segment and the time domain autocorrelation index. The frequency domain quality index indicates the frequency domain quality of the signal segment under the time domain autocorrelation index. The first extraction module is configured to extract target signal segments from the acoustic signal based on the time-domain autocorrelation index and frequency-domain quality index of each signal segment, wherein at least one of the time-domain autocorrelation index and the frequency-domain quality index of the target signal segment is greater than or equal to a corresponding preset threshold.

17. A fault early warning device, characterized in that, The device includes: The fourth acquisition module is configured to acquire multiple signal segments of the acoustic signal of the wind turbine. The fifth acquisition module is configured to acquire the time-domain autocorrelation index of each signal segment based on the envelope signal of each signal segment. The envelope signal indicates the curve of amplitude change with time at each sampling moment in the signal segment, and the time-domain autocorrelation index indicates the periodicity of the envelope signal. The sixth acquisition module is configured to acquire the frequency domain quality index of each signal segment based on the frequency domain characteristics of the envelope signal of each signal segment and the time domain autocorrelation index. The frequency domain quality index indicates the frequency domain quality of the signal segment under the time domain autocorrelation index. The second extraction module is configured to extract target signal segments from the acoustic signal based on the time-domain autocorrelation index and the frequency-domain quality index of each signal segment, wherein at least one of the time-domain autocorrelation index and the frequency-domain quality index of the target signal segment is greater than or equal to a corresponding preset threshold. The identification module is configured to identify faults based on the target signal segment, and to issue an early warning if the identification result indicates that a fault exists.

18. An electronic device, characterized in that, The electronic device includes: One or more processors; Memory used to store the executable program code of the processor; The processor is configured to execute the program code to implement the method for extracting signal segments from acoustic signals as described in any one of claims 1 to 14 or the fault warning method as described in claim 15.

19. A computer-readable storage medium, characterized in that, When the instructions in the computer-readable storage medium are executed by the processor of the electronic device, the electronic device is able to perform the method for extracting signal segments from an acoustic signal as described in any one of claims 1 to 14 or the fault warning method as described in claim 15.

20. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the method for extracting signal segments from acoustic signals as described in any one of claims 1 to 14 or the fault warning method as described in claim 15.

Images