Fan blade diagnosis method and device based on voiceprint perception technology and product

Abstract

The invention discloses a fan blade diagnosis method and device based on the voiceprint perception technology and a product, and the method comprises the steps: firstly obtaining a sound detection signal of a target fan blade, and then carrying out the feature extraction of the sound detection signal, and obtaining a corresponding feature vector; then, based on the feature vector, extracting an effective blade sound signal from the sound detection signal; after the extraction of the effective sound signals is completed, voiceprint feature extraction processing can be carried out on the effective blade sound signals to obtain corresponding voiceprint features; thirdly, constructing a blade damage detection model which takes voiceprint features as input and takes damage detection results as output; on the basis, the voiceprint features of the effective blade sound signals can be directly input into the model, so that the damage detection result of the target fan blade can be quickly obtained; therefore, blade detection is carried out based on the voiceprint technology, the cost is low, and the method can adapt to the complex field environment of the fan.

Description

Technical Field

[0001] The present invention belongs to the technical field of fan blade damage detection, and specifically relates to a fan blade diagnosis method, device and product based on voiceprint sensing technology. Background Art

[0002] With the rapid development of my country's new energy industry, the scale of wind power generation is expanding, and the capacity and installed scale of wind turbines are increasing rapidly; but with the rapid development of the wind power industry, the problem of wind turbine failure is becoming increasingly prominent. Therefore, it is very necessary to carry out status monitoring and fault diagnosis of various parts of wind turbines; among them, the blades work in a complex open-air environment for a long time and are prone to failure. If the failure is not discovered and resolved in time, it may cause personnel safety accidents and huge economic losses; based on this, it is of great significance to detect the operating status of wind turbine blades.

[0003] At present, there are many non-destructive testing technologies for wind turbine blades, including infrared imaging, machine vision and other technologies; however, the equipment required for the aforementioned testing technologies is expensive, and most of them have strict requirements on the monitoring environment. Therefore, it is difficult to complete the testing in a field environment, and its accuracy is low; based on this, how to provide a low-cost diagnosis method for wind turbine blades that can be applied to the complex environment in which the wind turbine blades are located has become a problem that needs to be solved urgently. Summary of the invention

[0004] The purpose of the present invention is to provide a wind turbine blade diagnosis method, device and product based on voiceprint sensing technology, so as to solve the problems of high cost and difficulty in completing blade detection in field environments in the prior art.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] In a first aspect, a wind turbine blade diagnosis method based on voiceprint sensing technology is provided, comprising:

[0007] Acquire the sound detection signal of the target fan blade;

[0008] Performing feature extraction processing on the sound detection signal to obtain a feature vector corresponding to the sound detection signal;

[0009] According to the feature vector, extracting a valid blade sound signal from the sound detection signal as a signal to be detected;

[0010] Performing voiceprint feature extraction processing on the signal to be detected to obtain the voiceprint feature of the target fan blade;

[0011] A blade damage detection model is constructed, wherein the blade damage detection model is trained by taking the soundprint features corresponding to the sample effective blade sound signals of several sample fan blades as input and the damage detection results of each sample fan blade as output;

[0012] The soundprint feature of the target fan blade is input into the blade damage detection model to perform blade damage detection processing, so as to obtain a damage detection result of the target fan blade after the blade damage detection processing.

[0013] Based on the above disclosed content, the present invention first obtains the sound detection signal of the target wind turbine blade, and then performs feature extraction processing on the sound detection signal to obtain its corresponding feature vector; then, based on the feature vector, an effective blade sound signal is extracted from the sound detection signal, so as to perform blade damage detection based on the effective blade sound signal; in this way, by extracting the effective sound signal, the noise generated in the sound due to the complex on-site environment can be removed, thereby reducing the interference of the complex on-site environment on blade detection; after completing the extraction of the effective sound signal, the effective blade sound signal can be subjected to voiceprint feature extraction processing to obtain the corresponding voiceprint feature; then, the present invention constructs a blade damage detection model with voiceprint features as input and damage detection results as output; based on this, the voiceprint features of the effective blade sound signal can be directly input into the aforementioned model to quickly obtain the damage detection results of the target wind turbine blade.

[0014] Through the above design, the present invention performs blade detection based on voiceprint technology, wherein the sound signal is easy to obtain and does not require expensive and complex equipment, thereby reducing the detection cost; at the same time, the present invention uses the effective signal in the collected sound signal to perform blade detection, thereby removing the noise in the audio signal caused by the complex on-site environment, which can reduce the interference of the complex on-site environment on blade detection, thereby improving the accuracy of detection; therefore, the wind turbine blade diagnosis method provided by the present invention is not only low-cost, but also adaptable to the complex on-site environment of the wind turbine; therefore, it is very suitable for large-scale application and promotion.

[0015] In a possible design, performing feature extraction processing on the sound detection signal to obtain a feature vector corresponding to the sound detection signal includes:

[0016] Performing frame processing on the sound detection signal to obtain multiple frames of sound signals;

[0017] Performing feature extraction processing on each frame of sound signal to obtain a feature vector corresponding to each frame of sound signal, so as to use the feature vector corresponding to each frame of sound signal to form a feature vector corresponding to the sound detection signal;

[0018] Correspondingly, extracting a valid blade sound signal from the sound detection signal according to the feature vector includes:

[0019] According to the feature vector corresponding to each frame of the sound signal, each frame of the sound signal is subjected to noise identification processing, so as to filter out the noise signal from the multiple frames of the sound signal;

[0020] According to the noise signal, an effective signal is intercepted from the sound detection signal to serve as the effective blade sound signal.

[0021] In a possible design, feature extraction is performed on each frame of sound signal to obtain a feature vector corresponding to each frame of sound signal, including:

[0022] For any frame of sound signal, preprocessing the frame of sound signal to obtain a preprocessed sound signal, wherein the preprocessing includes pre-emphasis processing and windowing processing;

[0023] Performing wavelet decomposition processing on the preprocessed sound signal to obtain a plurality of wavelet component signals and a wavelet coefficient corresponding to each wavelet component signal;

[0024] Performing discrete fast Fourier transform processing on each wavelet component signal to obtain the frequency spectrum corresponding to each wavelet component signal;

[0025] According to the wavelet coefficients corresponding to the wavelet component signals, the frequency spectra corresponding to the wavelet component signals are spliced ​​to obtain the complete frequency spectra of the pre-processed sound signal after the splicing process;

[0026] Based on the complete spectrum, calculating the spectrum characteristic value of the preprocessed sound signal;

[0027] The eigenvalues ​​of the frequency spectrum are used to generate a eigenvector of the sound signal of any frame.

[0028] In a possible design, calculating the spectrum characteristic value of the preprocessed sound signal based on the complete spectrum includes:

[0029] Acquire several groups of frequency parameters, wherein each group of frequency parameters includes a maximum frequency and a minimum frequency;

[0030] According to the complete spectrum and several groups of frequency parameters, and in accordance with the following formula (1), the spectrum energy of the complete spectrum under each group of frequency parameters is calculated;

[0031]

[0032] In the above formula (1), E(i) represents the spectrum energy of the complete spectrum under the i-th group of frequency parameters, H(j) represents the j-th frequency in the complete spectrum, and F i (j) represents the control parameter corresponding to the jth frequency, f min,i represents the minimum frequency in the i-th group of frequency parameters, f max,i represents the maximum frequency in the i-th group of frequency parameters, where i=1, 2, ..., n, and n represents the total number of groups of frequency parameters;

[0033] in,

[0034] In formula (2), k(i), k(i-1), and k(i+1) all represent frequency thresholds;

[0035] in,

[0036] In formula (3), T represents a constant, f1 represents the sampling frequency, and F mel () represents the Mel scale function, represents the inverse function of the Mel scale function;

[0037] The spectrum characteristic value is formed by using the spectrum energy of the complete spectrum under each set of frequency parameters;

[0038] Correspondingly, using the frequency spectrum feature value to generate a feature vector of any frame of sound signal includes:

[0039] Get the characteristic order;

[0040] Based on each spectrum energy in the spectrum characteristic value, and according to the following formula (4), the sound characteristic value of any frame sound signal at each characteristic order is calculated;

[0041]

[0042] In the above formula (4), C(m) represents the sound feature value of any frame sound signal at the mth feature order, where m=1, 2, ..., M, and M represents the total order of feature orders;

[0043] The feature vector of the sound signal of any frame is formed by using the sound feature value of the sound signal of any frame at each feature order.

[0044] In one possible design, a blade damage detection model is constructed, including:

[0045] Obtaining model parameters of the initial neural network model and initializing the model parameters to obtain multiple sets of initial model parameters;

[0046] Based on multiple groups of initial model parameters, a longicorn population is generated, wherein each longicorn individual in the longicorn population is configured with an initial position vector and an initial velocity vector, and the initial position vector of each longicorn individual corresponds to a group of initial model parameters;

[0047] Initialize the number of iterations t to 1, and obtain the position vector and velocity vector of each beetle individual at the t-th iteration, wherein when t is 1, the position vector and velocity vector of any beetle individual at the t-th iteration are the initial position vector and initial velocity vector of any beetle individual;

[0048] Based on the position vector of each longhorn beetle individual at the t-th iteration, the model parameters of each longhorn beetle individual at the t-th iteration are determined;

[0049] According to the model parameters of each longhorn beetle individual at the t-th iteration, a neural network model corresponding to each longhorn beetle individual is constructed, and the neural network model corresponding to each longhorn beetle individual is trained using the training data, so as to obtain the root mean square error of the neural network model corresponding to each longhorn beetle individual after the training is completed;

[0050] The fitness of each individual beetle is calculated by using the root mean square error of the neural network model corresponding to each individual beetle.

[0051] Based on the fitness of each beetle individual, the global optimal solution and the individual optimal solution of each beetle individual are updated, wherein the global optimal solution is the position vector corresponding to the beetle individual with the largest fitness among all beetle individuals in the first t iterations;

[0052] Determine whether an iteration stop condition is met, wherein the iteration stop condition is that the fitness of the longicorn individual corresponding to the global optimal solution is greater than or equal to a fitness threshold, or t reaches a maximum number of iterations;

[0053] If not, the position vector and velocity vector of each longhorn beetle individual at the t-th iteration are used to calculate the left and right antenna positions of each longhorn beetle individual;

[0054] Using the velocity vector, the left antenna position and the right antenna position of each longhorn beetle individual at the t-th iteration, the position vector of each longhorn beetle individual is updated to obtain the updated position vector of each longhorn beetle individual;

[0055] Using the global optimal solution and the individual optimal solution of each longhorn beetle individual, the velocity vector of each longhorn beetle individual is updated to obtain the updated velocity vector of each longhorn beetle individual;

[0056] Add 1 to t, replace the position vector of each longhorn beetle individual at the t-th iteration with the corresponding updated position vector, and replace the velocity vector of each longhorn beetle individual at the t-th iteration with the corresponding updated velocity vector, and re-determine the model parameters of each longhorn beetle individual at the t-th iteration based on the position vector of each longhorn beetle individual at the t-th iteration, until the iteration stop condition is met, and obtain the optimal model parameters based on the global optimal solution when the iteration stop condition is met;

[0057] The initial neural network model is optimized using the optimal model parameters, so that a blade damage detection model is obtained after the optimization.

[0058] In a possible design, the position vector and velocity vector of each longhorn beetle individual at the t-th iteration are used to calculate the left and right antenna positions of each longhorn beetle individual, including:

[0059] For any individual beetle, the following formula (5) is used to calculate the left and right tentacle positions of the individual beetle;

[0060]

[0061] In the above formula (5), represents the position of the right tentacle of any one of the Tianyan individuals, represents the left tentacle position of any one of the Tianya individuals, X t represents the position vector of any individual Tianniu at the tth iteration, V t represents the velocity vector of any one of the Tianbo individuals at the t-th iteration, and d represents the total distance from the left and right tentacles of any one of the Tianbo individuals to the center of mass of any one of the Tianbo individuals.

[0062] In a possible design, the position vector of each beetle individual is updated using the velocity vector, the left antenna position and the right antenna position of each beetle individual at the t-th iteration, and the updated position vector of each beetle individual is obtained, including:

[0063] For any individual Tianniu, generate an update random number;

[0064] Determine whether the individual optimal solution of any one of the Tianniu individuals is greater than the updated random number;

[0065] If not, generate a first random integer, a second random integer, and a third random integer, wherein the value range of the first random integer, the second random integer, and the third random integer is [1, R], and R represents the total number of longicorn individuals;

[0066] According to the first random integer, the second random integer and the third random integer, and in accordance with the following formula (6), the position vector of any one of the individual beetles is updated;

[0067]

[0068] In the above formula (6), X t+1 represents the updated position vector of any of the individual beetles, represents the position vector of the αth individual in the beetle population at the tth iteration, represents the position vector of the βth individual in the beetle population at the tth iteration, represents the position vector of the δth longicorn individual in the longicorn population at the tth iteration, wherein α is equal to the first random integer, β is equal to the second random integer, δ is equal to the third random integer, and γ represents the variation factor.

[0069] In one possible design, the method further includes:

[0070] If yes, then the displacement length of any individual beetle is calculated using the velocity vector, the left tentacle position and the right tentacle position of the individual beetle at the t-th iteration according to the following formula (7);

[0071]

[0072] In the above formula (7), represents the displacement length of any individual beetle, μ t represents the moving step length of any Tianniu individual at the tth iteration, V t represents the velocity vector of any individual strychnine at the tth iteration, represents the position of the right tentacle of any one of the Tianyan individuals, represents the left tentacle position of any one of the individual beetles, f() is the fitness function, and sign is the sign function;

[0073] Based on the displacement length of any Tianniu individual, the position vector and velocity vector of any Tianniu individual at the t-th iteration, and according to the following formula (8), the updated position vector of any Tianniu individual is calculated;

[0074]

[0075] In the above formula (8), Represents the position update factor.

[0076] In a second aspect, a wind turbine blade diagnosis device based on voiceprint sensing technology is provided, comprising:

[0077] An acquisition unit, used for acquiring a sound detection signal of a target fan blade;

[0078] A feature extraction unit, used to perform feature extraction processing on the sound detection signal to obtain a feature vector corresponding to the sound detection signal;

[0079] A sound separation unit, used for extracting a valid blade sound signal from the sound detection signal according to the feature vector as a signal to be detected;

[0080] The feature extraction unit is further used to perform voiceprint feature extraction processing on the signal to be detected to obtain the voiceprint feature of the target fan blade;

[0081] A model building unit is used to build a blade damage detection model, wherein the blade damage detection model is trained by taking the soundprint features corresponding to the sample effective blade sound signals of a number of sample fan blades as input and the damage detection results of each sample fan blade as output;

[0082] The blade detection unit is used to input the soundprint feature of the target fan blade into the blade damage detection model to perform blade damage detection processing, so as to obtain the damage detection result of the target fan blade after the blade damage detection processing.

[0083] In a third aspect, another wind blade diagnostic device based on voiceprint sensing technology is provided. Taking the device as an electronic device as an example, it includes a memory, a processor and a transceiver that are communicatively connected in sequence, wherein the memory is used to store computer programs, the transceiver is used to send and receive messages, and the processor is used to read the computer program to execute the wind blade diagnostic method based on voiceprint sensing technology as described in the first aspect or any possible design of the first aspect.

[0084] In a fourth aspect, a storage medium is provided, on which instructions are stored. When the instructions are executed on a computer, the wind turbine blade diagnosis method based on voiceprint sensing technology as described in the first aspect or any possible design of the first aspect is executed.

[0085] In a fifth aspect, a computer program product comprising instructions is provided, which, when executed on a computer, causes the computer to execute the wind turbine blade diagnosis method based on voiceprint sensing technology as in the first aspect or any possible design of the first aspect.

[0086] Beneficial effects:

[0087] (1) The present invention performs blade detection based on voiceprint technology, wherein the sound signal is easy to obtain and does not require expensive and complex equipment, thereby reducing the detection cost; at the same time, the present invention uses the effective signal in the collected sound signal to perform blade detection, thereby removing the noise in the audio signal caused by the complex on-site environment, which can reduce the interference of the complex on-site environment on the blade detection, thereby improving the accuracy of the detection; therefore, the wind turbine blade diagnosis method provided by the present invention is not only low-cost, but also adaptable to the complex on-site environment of the wind turbine; therefore, it is very suitable for large-scale application and promotion. BRIEF DESCRIPTION OF THE DRAWINGS

[0088] Figure 1 A schematic flow chart of the steps of a wind turbine blade diagnosis method based on voiceprint sensing technology provided by an embodiment of the present invention;

[0089] Figure 2 A schematic diagram of the structure of a wind turbine blade diagnostic device based on voiceprint sensing technology provided by an embodiment of the present invention;

[0090] Figure 3 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0091] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in combination with the drawings and the description of the embodiments or the prior art. Obviously, the following description of the structure of the drawings is only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. It should be noted that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation of the present invention.

[0092] It should be understood that although the terms first, second, etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another unit. For example, a first unit can be referred to as a second unit, and similarly, a second unit can be referred to as a first unit without departing from the scope of the exemplary embodiments of the present invention.

[0093] It should be understood that the term "and / or" that may appear in this article is merely a description of the association relationship between associated objects, indicating that three relationships may exist. For example, A and / or B can represent three situations: A exists alone, B exists alone, and A and B exist at the same time. The term " / and" that may appear in this article describes another type of association object relationship, indicating that two relationships may exist. For example, A / and B can represent two situations: A exists alone, and A and B exist alone. In addition, the character " / " that may appear in this article generally indicates that the previous and next associated objects are in an "or" relationship.

[0094] Example:

[0095] See also Figure 1 As shown, the fan blade diagnosis method based on voiceprint sensing technology provided in this embodiment performs blade detection based on voiceprint technology, wherein the sound signal is easy to obtain and no expensive and complex equipment is required, thereby reducing the detection cost; at the same time, the present invention uses the effective signal in the collected sound signal to perform blade detection, thereby removing the noise generated by the complex on-site environment in the audio signal, and reducing the interference of the complex on-site environment on the blade detection, thereby improving the accuracy of the detection; therefore, the fan blade diagnosis method provided by the present method is not only low-cost, but also adaptable to the complex on-site environment of the fan; therefore, it is very suitable for large-scale application and promotion; wherein, for example, the present method can be but not limited to running on the fan blade diagnosis end side, optionally, the fan blade diagnosis end can be but not limited to a personal computer (PC), it can be understood that the aforementioned execution subject does not constitute a limitation on the embodiments of the present application, and accordingly, the operation steps of the present method can be but not limited to the following steps S1 to S6.

[0096] S1. Obtain the sound detection signal of the target wind turbine blade; in the specific implementation, for example, but not limited to, a sound signal acquisition device is installed on the tower of the target wind turbine, which includes a main control chip, an audio acquisition chip, a communication chip and a power supply, etc.; wherein, the main control chip can adopt the commonly used STM32 chip (can work in an environment of -40° to 60°, which is very suitable for the operating environment of the wind turbine), the audio acquisition chip can adopt the VS1053 chip (sampling frequency up to 16Khz) and the communication chip can adopt a 4G or 5G communication chip, etc.; in this way, the aforementioned devices use commonly used chips, which are low in cost and high in reliability, and can reduce the cost of blade damage detection; based on this, the above-mentioned sound signal acquisition device is used to collect the sound detection signal of the target wind turbine blade.

[0097] At the same time, since the environment in which the target fan is located is very complex and there are various random noise sources, such as wind sound, there is a lot of noise in the collected sound detection signal; therefore, in order to adapt to the complex on-site environment of the target fan and ensure the accuracy of its blade detection in a complex environment, the present embodiment is provided with an effective signal extraction process, that is, to extract the effective blade sound signal from the collected sound detection signal, so as to complete the subsequent blade detection based on the signal; wherein, the extraction process of the effective blade sound signal can be but is not limited to the following steps S2 and S3 as shown.

[0098] S2. Perform feature extraction processing on the sound detection signal to obtain a feature vector corresponding to the sound detection signal; in this embodiment, the sound detection signal is firstly subjected to frame processing to obtain multiple frames of sound signals; then, feature extraction processing is performed on each frame of the sound signal to obtain a feature vector corresponding to each frame of the sound signal, so as to utilize the feature vector corresponding to each frame of the sound signal to form a feature vector corresponding to the sound detection signal.

[0099] Optionally, any frame of sound signal is taken as an example to illustrate the extraction process of its corresponding feature vector; wherein, the extraction process of the feature vector of any frame of sound signal may be but is not limited to the following steps S21 to S26.

[0100] S21. For any frame of sound signal, preprocess the any frame of sound signal to obtain a preprocessed sound signal, wherein the preprocessing includes pre-emphasis processing and windowing processing; in this embodiment, the pre-emphasis processing is to use a high-pass filter to enhance the high-frequency components in the any frame of sound signal; at the same time, the windowing processing is to multiply the any frame of sound signal by a Hamming window to increase the continuity of the signal; of course, the aforementioned pre-emphasis processing and windowing processing are both commonly used techniques for signal processing, and their principles will not be repeated.

[0101] After completing the preprocessing of any frame of sound signal, wavelet decomposition processing can be performed so that the wavelet components obtained by wavelet decomposition can be used to reconstruct the spectrum of any frame of sound signal; wherein the wavelet decomposition process can be but is not limited to the following step S22.

[0102] S22. Perform wavelet decomposition processing on the pre-processed sound signal to obtain a plurality of wavelet component signals and wavelet coefficients corresponding to each wavelet component signal; in this embodiment, the principle of wavelet decomposition is multi-resolution analysis, that is, by scaling and translating a set of wavelet basis functions, the time-frequency window of the signal changes with the frequency, thereby obtaining wavelet components of different frequency bands; of course, wavelet decomposition is a commonly used technology for signal processing, and its principle will not be repeated here.

[0103] After obtaining several wavelet component signals and corresponding wavelet coefficients, a frequency-band discrete Fourier transform processing method can be used to generate the spectrum of each wavelet component signal; wherein, the generation process of the spectrum corresponding to the wavelet component signal can be but is not limited to the following step S23.

[0104] S23. Perform discrete fast Fourier transform processing on each wavelet component signal to obtain the frequency spectrum corresponding to each wavelet component signal; in the present embodiment, different wavelet coefficients obtained by wavelet decomposition belong to different frequency bands, and the wavelet coefficients in the high frequency band represent the fast-varying components of the sound signal of any frame, while the wavelet coefficients in the low frequency band represent the slow-varying components; therefore, discrete fast Fourier transform processing is performed on the wavelet components of different frequency bands respectively, so that the energy of different frequency bands can be distinguished, and at the same time, subsequent frequency spectrum splicing of the divided frequency bands and then feature extraction are performed, so that on the basis of feature extraction, the wavelet sub-band energy can be introduced, thereby improving the comprehensiveness of feature extraction, so that the obtained feature vector can have a stronger noise recognition ability; of course, discrete fast Fourier transform processing is also a commonly used technology in signal processing, and its principle will not be repeated here.

[0105] In this way, after obtaining the spectrum corresponding to each wavelet component signal based on the aforementioned step S23, the spectrum splicing in the same frequency band can be performed based on the wavelet coefficients, and finally merged to obtain the complete spectrum of the aforementioned preprocessed sound signal; wherein, the spectrum splicing process can be but is not limited to as shown in the following step S24.

[0106] S24. According to the wavelet coefficients corresponding to the wavelet component signals, the frequency spectra corresponding to the wavelet component signals are spliced ​​to obtain the complete frequency spectrum of the preprocessed sound signal after the splicing process; in the specific implementation, it is equivalent to first determining the frequency spectra corresponding to the wavelet component signals belonging to the same frequency band according to the wavelet coefficients, and then splicing the frequency spectra corresponding to the wavelet component signals belonging to the same frequency band; finally, the spliced ​​frequency spectra corresponding to the various frequency bands are merged to obtain the aforementioned complete frequency spectrum; and after obtaining the complete frequency spectrum of the aforementioned preprocessed sound signal, feature extraction can be performed, and the process can be but not limited to as shown in the following steps S25 and S26.

[0107] S25. Based on the complete spectrum, the spectrum characteristic values ​​of the preprocessed sound signal are calculated; in specific implementation, for example, but not limited to, several groups of frequency parameters are first obtained; then, the spectrum energy of the complete spectrum under each group of frequency parameters is calculated based on the complete spectrum and the several groups of frequency parameters; finally, the spectrum energy of the complete spectrum under each group of frequency parameters can be used to form the spectrum characteristic values; wherein, for example, each group of frequency parameters includes a maximum frequency and a minimum frequency; based on this, it is equivalent to using the maximum frequency and the minimum frequency in each group of frequency parameters as boundaries to calculate the spectrum energy of the aforementioned complete spectrum under each group of frequency parameters.

[0108] Optionally, the spectrum energy of the complete spectrum under each set of frequency parameters may be calculated by, for example but not limited to, using the following formula (2).

[0109]

[0110] In the above formula (1), E(i) represents the spectrum energy of the complete spectrum under the i-th group of frequency parameters, H(j) represents the j-th frequency in the complete spectrum, and F i (j) represents the control parameter corresponding to the jth frequency, f min,i represents the minimum frequency in the i-th group of frequency parameters, f max,i represents the maximum frequency in the i-th group of frequency parameters, where i=1, 2, ..., n, and n represents the total number of groups of frequency parameters.

[0111] In this embodiment, the aforementioned control parameters can be calculated using, but not limited to, the following formula (2).

[0112]

[0113] In the above formula (2), k(i), k(i-1), and k(i+1) all represent frequency thresholds; in this embodiment, k(i) is taken as an example to illustrate the corresponding calculation process.

[0114] in,

[0115] In the above formula (3), T represents a constant, f1 represents the sampling frequency, and F mel () represents the Mel scale function, represents the inverse function of the Mel scale function; optionally, T is the number of signal points of any of the aforementioned frames of sound signal, and f1 represents the sampling frequency of any of the aforementioned frames of sound signal; at the same time, Similarly,

[0116] Thus, the spectrum energy of the complete spectrum under each set of frequency parameters can be calculated through the aforementioned formulas (1)-(3); and then, each spectrum energy can be used to form the aforementioned frequency characteristic value.

[0117] After the frequency characteristic value is obtained, the characteristic vector of any frame sound signal can be generated based on it. The process can be but not limited to the following step S26.

[0118] S26. Generate a feature vector of any frame of sound signal using the spectral feature values; in specific implementation, for example, but not limited to, first obtaining the feature order, and then, based on the spectral energies in the spectral feature values, calculating the sound feature values ​​of any frame of sound signal at each feature order; finally, the sound feature values ​​of any frame of sound signal at each feature order can be used to form a feature vector of any frame of sound signal.

[0119] Optionally, the following formula (4) may be used, for example but not limited to, to calculate the sound feature value of any frame of sound signal at each feature order.

[0120]

[0121] In the above formula (4), C(m) represents the sound feature value of any frame of the sound signal at the mth feature order, where m=1, 2, ..., M, and M represents the total order of the feature orders.

[0122] After calculating a plurality of sound feature values ​​of any frame of sound signal based on the aforementioned formula (4), a feature vector can be formed based on the values.

[0123] In this way, through the aforementioned steps S21 to S26, the characteristic vector of each frame sound signal can be extracted, and then, based on this, the characteristic vector of the sound detection signal can be formed; and after obtaining the characteristic vector of the sound detection signal, its corresponding characteristic vector can be used to perform noise identification, so as to extract a valid signal from the sound detection signal; wherein, the extraction process of the valid signal (i.e., the effective blade sound signal described below) can be but is not limited to as shown in the following step S3.

[0124] S3. According to the feature vector, extract the effective blade sound signal from the sound detection signal as the signal to be detected; in the specific implementation, first perform noise recognition processing on each frame of the sound signal according to the feature vector corresponding to each frame of the sound signal, so as to filter out the noise signal from the multiple frames of sound signals; then, based on the noise signal, extract the effective signal from the sound detection signal as the effective blade sound signal.

[0125] In specific applications, for example, but not limited to, first obtaining a noise recognition model, wherein the noise recognition model is trained with the feature vectors of several sample sound signals as input and the noise recognition results of each sample sound signal as output; optionally, for example, the aforementioned noise recognition model can be but not limited to an SVM model, that is, several sample sound signals are selected and labeled, with the valid signal labeled as 1 and the noise signal labeled as 0; then, the feature vectors of the aforementioned sample sound signals are input into the SVM model, and the corresponding labels are used as the expected output to train the SVM model, thereby obtaining a noise recognition model; based on this, the feature vectors of the aforementioned frames of sound signals can be directly input into the noise recognition model to identify the noise signal.

[0126] Furthermore, after the noise signal is identified, the starting time and the ending time of each noise signal can be determined; then, the sound segment belonging to the valid signal can be cut out from the sound detection signal to obtain the valid blade sound signal.

[0127] After obtaining the effective blade sound signal of the target wind turbine blade, blade damage detection can be performed; wherein, the present embodiment first extracts the voiceprint features of the effective blade sound signal (i.e., the signal to be detected below); then, the voiceprint features are used to identify blade damage; wherein, the voiceprint feature extraction process can be but is not limited to as shown in the following step S4.

[0128] S4. Perform voiceprint feature extraction processing on the signal to be detected to obtain the voiceprint features of the target fan blades; in this embodiment, for example, but not limited to, extracting MFCC (Mel frequency cepstral coefficient) features of the signal to be detected as the voiceprint features of the target fan blades; of course, MFCC feature extraction is a commonly used technology for audio signal processing, and its principle will not be repeated.

[0129] After obtaining the voiceprint features of the target fan blade, damage identification of the target fan blade can be performed based on the voiceprint features; wherein, this embodiment constructs a blade damage detection model, and uses the model to implement damage detection of the target fan blade; wherein, the construction process of the aforementioned model can be but is not limited to as shown in the following step S5.

[0130] S5. Construct a blade damage detection model, wherein the blade damage detection model is trained by taking the soundprint features corresponding to the sample effective blade sound signals of several sample fan blades as input and the damage detection results of each sample fan blade as output; in the specific implementation, for example, the blade damage detection model may also adopt the SVM (support vector machine) model; wherein, the support vector machine (SVM) is a machine learning algorithm based on statistics, which realizes the classification of data by constructing an optimal hyperplane. Therefore, a label will be set for each sample effective sound blade during training, and the label is 1 for a normal signal and the label is 0 for a fault signal. Therefore, it is equivalent to taking the soundprint features corresponding to each sample effective blade sound signal as input and the label corresponding to each sample effective blade sound signal as the expected output to train the SVM model, thereby obtaining the corresponding blade damage detection model; of course, this embodiment only considers whether the blade is faulty, but does not consider the specific fault type.

[0131] At the same time, the training of the SVM model mainly requires determining the optimal penalty factor of the model and the bandwidth of the radial basis kernel function of the model; therefore, this embodiment provides a new model parameter optimization algorithm, the process of which can be but is not limited to the following steps S501 to S513.

[0132] S501. Obtain model parameters of the initial neural network model, and initialize the model parameters to obtain multiple sets of initial model parameters; in the specific implementation, the initial neural network model is an untrained SVM model, and its corresponding model parameters are the penalty factor and the bandwidth of the model radial basis kernel function; therefore, it is initialized multiple times to obtain multiple sets of initial model parameters; after completing the initialization of the model parameters, the longicorn population can be introduced to complete the optimization of the model parameters, and the process can be but not limited to the following steps S502 to S512.

[0133] S502. Generate a longicorn population based on multiple groups of initial model parameters, wherein each longicorn individual in the longicorn population is configured with an initial position vector and an initial velocity vector, and the initial position vector of each longicorn individual corresponds to a group of initial model parameters; in the specific implementation, assuming that there are 10 groups of initial model parameters, then 10 longicorn individuals are generated, and the initial position vectors corresponding to each longicorn individual correspond to a group of initial model parameters; based on this, it is equivalent to continuously iterating the initial position vectors of the 10 longicorn individuals to optimize and obtain the optimal model parameters; wherein, the optimization process can be but is not limited to the following steps S503 to S512 as shown.

[0134] S503. Initialize the number of iterations t to 1, and obtain the position vector and velocity vector of each longhorn beetle individual at the tth iteration, wherein when t is 1, the position vector and velocity vector of any longhorn beetle individual at the tth iteration are the initial position vector and initial velocity vector of any longhorn beetle individual; in this embodiment, at the first iteration, the position vector and velocity vector corresponding to each longhorn beetle individual are the initial values, that is, the initial position vector and initial velocity vector; then, based on this, the position and velocity are updated so as to find the optimal model parameters during the position and velocity update process.

[0135] Optionally, the specific optimization process may be but is not limited to the following steps S504 to S512.

[0136] S504. Based on the position vector of each longhorn beetle individual at the tth iteration, determine the model parameters of each longhorn beetle individual at the tth iteration; in the present embodiment, it has been explained above that the initial position vector of each longhorn beetle individual corresponds to an initial model parameter, therefore, each element in the position vector represents each model parameter of the SVM model, based on this, the model parameters of each longhorn beetle individual at the tth iteration can be determined according to the position vector, and then, the model parameters at the tth iteration can be used to construct a neural network model corresponding to each longhorn beetle individual, so as to calculate the fitness of each longhorn beetle individual at the tth iteration based on the neural network model; wherein, the model construction process is shown in the following step S505.

[0137] S505. According to the model parameters of each longhorn beetle individual at the tth iteration, a neural network model corresponding to each longhorn beetle individual is constructed, and the training data is used to train the neural network model corresponding to each longhorn beetle individual, so as to obtain the root mean square error of the neural network model corresponding to each longhorn beetle individual after the training is completed; in this embodiment, the training data is the voiceprint features corresponding to the sample effective blade sound signals of the sample fan blades in the aforementioned step S5; in this way, after the training is completed, the root mean square error of each neural network model can be calculated, and then, based on the root mean square error, the fitness of each longhorn beetle individual can be calculated; wherein, the calculation process of the fitness of the longhorn beetle individual can be but is not limited to as shown in the following step S506.

[0138] S506. The fitness of each longhorn beetle individual is calculated using the root mean square error of the neural network model corresponding to each longhorn beetle individual; in specific applications, the smaller the root mean square error, the smaller the deviation between the predicted value and the actual value of the model, and the more accurate the model prediction result. Therefore, this embodiment uses the inverse of the root mean square error as the fitness function; in this way, the larger the fitness function, the smaller the root mean square error, and the higher the prediction accuracy of the model.

[0139] After calculating the fitness of each longhorn beetle individual at the t-th iteration, the global optimal solution and the individual optimal solution may be updated, and the process may be but is not limited to that shown in the following step S507.

[0140] S507. Based on the fitness of each longhorn beetle individual, update the global optimal solution and the individual optimal solutions of each longhorn beetle individual, wherein the global optimal solution is the position vector corresponding to the longhorn beetle individual with the largest fitness among all longhorn beetle individuals during the first t iterations; in specific applications, the global optimal solution is the position vector corresponding to the longhorn beetle individual with the largest fitness during the first t iterations, such as in the first five iterations, the fitness of longhorn beetle individual 4 in the third iteration is the largest, then the position vector corresponding to longhorn beetle individual 4 in the third iteration is the global optimal solution during the first five iterations; at the same time, assuming that in the fifth iteration, the fitness of longhorn beetle individual 3 is the largest and is greater than the previous fitness of longhorn beetle individual 4, then the global optimal solution at this time is updated to the position vector corresponding to longhorn beetle individual 3; similarly, the individual optimal solution of any longhorn beetle individual is the position vector corresponding to the fitness of any longhorn beetle individual when it is the largest during the 1st to tth iterations.

[0141] In this way, after the global optimal solution and the individual optimal solution are updated, it can be determined whether the iteration stop condition is met, and the determination process can be but is not limited to the following step S508.

[0142] S508. Determine whether an iteration stopping condition is met, wherein the iteration stopping condition is that the fitness of the longicorn individual corresponding to the global optimal solution is greater than or equal to a fitness threshold, or t reaches a maximum number of iterations.

[0143] In this embodiment, if the iteration stop condition is met, the elements in the position vector corresponding to the global optimal solution can be used as the optimal model parameters; otherwise, it is necessary to continue iteration, that is, to update the position and speed of each individual longicorn beetle, and the update process can be but is not limited to the following steps S509 to S511.

[0144] S509. If not, the position vector and velocity vector of each longhorn beetle individual at the t-th iteration are used to calculate the position of the left and right antennae of each longhorn beetle individual. In the specific implementation, any longhorn beetle individual is taken as an example for specific explanation. Among them, the example can be but not limited to using the following formula (1) to calculate the position of its left and right antennae.

[0145]

[0146] In the above formula (5), represents the position of the right tentacle of any one of the Tianyan individuals, represents the left tentacle position of any one of the Tianya individuals, X t represents the position vector of any individual Tianniu at the tth iteration, V t represents the velocity vector of any one of the beetles at the tth iteration, and d represents the total distance from the left and right tentacles of any one of the beetles to the center of mass of any one of the beetles. In this embodiment, d is a preset value.

[0147] In this way, the positions of the left and right whiskers of each longhorn beetle individual can be calculated based on the aforementioned formula (5); then, the positions of each longhorn beetle individual can be updated in combination with the velocity vectors of the aforementioned longhorn beetle individuals at the tth iteration; wherein, the position refinement process can be but is not limited to as shown in the following step S510.

[0148] S510. Use the velocity vector, left antenna position and right antenna position of each longhorn beetle individual at the tth iteration to update the position vector of each longhorn beetle individual to obtain the updated position vector of each longhorn beetle individual; in the specific implementation, any longhorn beetle individual is taken as an example for specific explanation, wherein this embodiment is provided with different update strategies, and the process is shown in the following steps S510a to S510d.

[0149] S510a. For any individual beetle, generate an update random number; in this embodiment, the update random number is greater than or equal to 1; thus, the update strategy of any individual beetle can be determined based on the generated update random number; wherein the specific determination process of the update strategy can be but is not limited to as shown in the following step S510b.

[0150] S510b. Determine whether the individual optimal solution of any of the individual beetles is greater than the updated random number; in the specific implementation, it has been explained above that the individual optimal solution is the position vector corresponding to the individual when the fitness is maximum during the 1st to tth iterations; therefore, it is to determine whether the modulus of the aforementioned individual optimal solution is greater than the updated random number, wherein, if it is greater, a mutation update strategy is adopted, and the process is as shown in the following steps SS510c to S510d.

[0151] S510c. If not, generate a first random integer, a second random integer and a third random integer, wherein the value range of the first random integer, the second random integer and the third random integer is [1, R], and R represents the total number of longicorn individuals; in this embodiment, it is equivalent to randomly selecting the position vectors of three longicorn individuals at the tth iteration among the R longicorn individuals to update the position of any longicorn individual, and the process is shown in the following step S510d.

[0152] S510d. Update the position vector of any one of the individual beetles according to the first random integer, the second random integer and the third random integer and in accordance with the following formula (6).

[0153]

[0154] In the above formula (6), X t+1 represents the updated position vector of any of the individual beetles, represents the position vector of the αth individual in the beetle population at the tth iteration, represents the position vector of the βth individual in the beetle population at the tth iteration, represents the position vector of the δth individual in the beetle population at the tth iteration, wherein α is equal to the first random integer, β is equal to the second random integer, δ is equal to the third random integer, and γ represents the variation factor; in this embodiment, the variation factor is taken in the range of [0,2].

[0155] In this way, when the generated update random number is less than or equal to the individual optimal solution of any of the individual Tianniu, the above formula (6) is used to complete the update of the position vector; at the same time, if the update random number is greater than the individual optimal solution of any of the individual Tianniu, the following steps S510e and S510f are used to perform the position update.

[0156] S510e. Using the velocity vector, the left tentacle position and the right tentacle position of the any one beetle individual at the t-th iteration, and according to the following formula (7), calculate the displacement length of the any one beetle individual.

[0157]

[0158] In the above formula (7), represents the displacement length of any individual beetle, μ t represents the moving step length of any Tianniu individual at the tth iteration, V t represents the velocity vector of any individual strychnine at the tth iteration, represents the position of the right tentacle of any one of the Tianyan individuals, represents the position of the left tentacle of any individual beetle, f() is the fitness function, and sign is the sign function; in this embodiment, the fitness of the left and right tentacle is the model parameter corresponding to the position vector of the left tentacle, to construct a neural network model, so as to perform model training to obtain the root mean square error of the model; in this way, the fitness value corresponding to the left tentacle can be obtained based on the root mean square error; similarly, the same is true for the right tentacle, which will not be repeated here.

[0159] At the same time, for examplet =μ t-1 ×eta, where μ t-1 represents the movement compensation of any one of the Tianniu individuals at the t-1th iteration, eta represents the compensation control factor, and the value of this embodiment is 0.96.

[0160] Based on this, after calculating the displacement length of any individual beetle based on the aforementioned formula (7), the position can be updated based on the displacement length, and the process can be but is not limited to the following step S510f.

[0161] S510f. Based on the displacement length of any Tianniu individual, the position vector and velocity vector of any Tianniu individual at the t-th iteration, and according to the following formula (8), calculate the updated position vector of any Tianniu individual.

[0162]

[0163] In the above formula (8), represents a location update factor; in this embodiment, the location update factor is taken as a constant greater than 0.

[0164] In this way, based on the aforementioned steps S510a to S510f, the position vector of each longhorn beetle individual can be updated; among them, this embodiment introduces a mutation update strategy, that is, when the iterative process enters the later stage, the position of the current optimal individual is mutated and updated, which can minimize the step size of the longhorn beetle individual, so that the longhorn beetle individual moves in a small range near the optimal solution, carefully searches for the current optimal position, and finally accurately finds the optimal solution.

[0165] After completing the position update of each individual longicorn beetle, the speed update may be performed, and the process may be but not limited to the process shown in the following step S511.

[0166] S511. Use the global optimal solution and the individual optimal solution of each longhorn beetle individual to update the velocity vector of each longhorn beetle individual to obtain the updated velocity vector of each longhorn beetle individual. In the specific implementation, any longhorn beetle individual is taken as an example for specific explanation. Among them, the example can be but is not limited to using the following formula (9) to update the velocity of any longhorn beetle individual.

[0167]

[0168] In the above formula (9), V t+1 represents the updated velocity vector of any individual stellate beetle, V t represents the velocity vector of any individual strychnine at the tth iteration, represents the individual optimal solution of any of the above mentioned Tianniu individuals, represents the global optimal solution at the tth iteration, X t represents the position vector of any Tianniu individual at the tth iteration, c1, c2 both represent learning factors, with values ​​of [0, 4], τ1, τ2 represent random numbers in the range of [0, 1], is the learning rate.

[0169] In this way, based on the aforementioned formula (9), the speed of each individual longhorn beetle can be updated; then, the updated position and speed vector can be used as the initial value for the next iteration, and the aforementioned optimization process can be repeated until the iteration stop condition is met to obtain the optimal model parameters of the SVM model; wherein, the iterative process can be but is not limited to the following step S512.

[0170] S512. Add 1 to t, replace the position vector of each longhorn beetle individual at the tth iteration with the corresponding updated position vector, and replace the velocity vector of each longhorn beetle individual at the tth iteration with the corresponding updated velocity vector, and re-determine the model parameters of each longhorn beetle individual at the tth iteration based on the position vector of each longhorn beetle individual at the tth iteration, until the iteration stop condition is met, and obtain the optimal model parameters based on the global optimal solution when the iteration stop condition is met.

[0171] Therefore, based on the aforementioned steps S501 to S512, the optimal model parameters of the SVM can be found and then, the model can be optimized based on the optimal model parameters to obtain a blade damage detection model, and the process can be but is not limited to as shown in the following step S513.

[0172] S513. Optimize the initial neural network model using the optimal model parameters to obtain a blade damage detection model after optimization; in this embodiment, it is equivalent to updating the model parameters of the initial neural network model to the aforementioned optimal model parameters; then, use the test set for testing, and after the test passes, use it as a blade damage detection model.

[0173] Thus, based on the aforementioned steps S501 to S513, a blade damage detection model can be constructed and trained; then, based on the detection model, fault detection of the target wind turbine blade can be implemented, and the process can be but not limited to as shown in the following step S6.

[0174] S6. Input the voiceprint feature of the target fan blade into the blade damage detection model for blade damage detection processing, so as to obtain the damage detection result of the target fan blade after the blade damage detection processing; in this embodiment, when the classification result output by the blade damage detection model is 0, it means that the target fan blade is damaged and is a faulty blade, otherwise, it is a normal blade; wherein, if the target fan blade is detected to be a faulty blade, at this time, a fault warning can be performed so that maintenance personnel can perform maintenance in time, thereby avoiding safety accidents.

[0175] Therefore, through the wind turbine blade diagnosis method based on voiceprint sensing technology described in detail in the aforementioned steps S1 to S6, the present invention performs blade detection based on voiceprint technology, wherein the sound signal is easy to obtain and does not require expensive and complex equipment, thereby reducing the detection cost; at the same time, the present invention uses the effective signal in the collected sound signal to perform blade detection, thereby removing the noise in the audio signal caused by the complex on-site environment, which can reduce the interference of the complex on-site environment on blade detection, thereby improving the accuracy of detection; therefore, the wind turbine blade diagnosis method provided by the present invention is not only low-cost, but also can adapt to the complex on-site environment of the wind turbine; therefore, it is very suitable for large-scale application and promotion.

[0176] like Figure 2 As shown, the second aspect of this embodiment provides a hardware device for implementing the wind turbine blade diagnosis method based on voiceprint sensing technology described in the first aspect of the embodiment, including:

[0177] The acquisition unit is used to acquire the sound detection signal of the target fan blade.

[0178] The feature extraction unit is used to perform feature extraction processing on the sound detection signal to obtain a feature vector corresponding to the sound detection signal.

[0179] The sound separation unit is used to extract the effective blade sound signal from the sound detection signal according to the feature vector as the signal to be detected.

[0180] The feature extraction unit is further used to perform voiceprint feature extraction processing on the signal to be detected to obtain the voiceprint feature of the target wind turbine blade.

[0181] The model building unit is used to build a blade damage detection model, wherein the blade damage detection model is trained by taking the soundprint features corresponding to the sample effective blade sound signals of several sample fan blades as input and the damage detection results of each sample fan blade as output.

[0182] The blade detection unit is used to input the soundprint feature of the target fan blade into the blade damage detection model to perform blade damage detection processing, so as to obtain the damage detection result of the target fan blade after the blade damage detection processing.

[0183] The working process, working details and technical effects of the device provided in this embodiment can be found in the first aspect of the embodiment and will not be described in detail here.

[0184] like Figure 3 As shown, the third aspect of this embodiment provides another wind blade diagnosis device based on voiceprint sensing technology. Taking the device as an electronic device as an example, it includes: a memory, a processor and a transceiver that are communicatively connected in sequence, wherein the memory is used to store computer programs, the transceiver is used to send and receive messages, and the processor is used to read the computer program to execute the wind blade diagnosis method based on voiceprint sensing technology as described in the first aspect of the embodiment.

[0185] For example, the memory may include, but is not limited to, random access memory (RAM), read only memory (ROM), flash memory, first input first output (FIFO) and / or first in last out (FILO), etc. Specifically, the processor may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). At the same time, the processor may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state.

[0186] In some embodiments, the processor may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the display screen. For example, the processor may not be limited to a microprocessor of the STM32F105 series, a reduced instruction set computer (RISC) microprocessor, an X86 or other architecture processor, or a processor with an integrated embedded neural network processing unit (NPU); the transceiver may be, but is not limited to, a wireless fidelity (WIFI) wireless transceiver, a Bluetooth wireless transceiver, a general packet radio service technology (General Packet Radio Service, GPRS) wireless transceiver, a ZigBee protocol (a low-power LAN protocol based on the IEEE802.15.4 standard, ZigBee) wireless transceiver, a 3G transceiver, a 4G transceiver and / or a 5G transceiver, etc. In addition, the device may also include, but is not limited to, a power module, a display screen and other necessary components.

[0187] The working process, working details and technical effects of the electronic device provided in this embodiment can be found in the first aspect of the embodiment and will not be described in detail here.

[0188] The fourth aspect of this embodiment provides a storage medium storing instructions for the wind blade diagnosis method based on voiceprint sensing technology as described in the first aspect of the embodiment, that is, the storage medium stores instructions, and when the instructions are run on a computer, the wind blade diagnosis method based on voiceprint sensing technology as described in the first aspect of the embodiment is executed.

[0189] The storage medium refers to a carrier for storing data, which may include but is not limited to a floppy disk, a CD, a hard disk, a flash memory, a USB flash drive and / or a memory stick, and the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0190] The working process, working details and technical effects of the storage medium provided in this embodiment can be found in the first aspect of the embodiment and will not be described in detail here.

[0191] The fifth aspect of this embodiment provides a computer program product comprising instructions, which, when executed on a computer, enables the computer to execute the wind blade diagnosis method based on voiceprint sensing technology as described in the first aspect of the embodiment, wherein the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0192] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A wind turbine blade diagnosis method based on voiceprint sensing technology, characterized in that: include: Acquire the sound detection signal of the target fan blade; Performing feature extraction processing on the sound detection signal to obtain a feature vector corresponding to the sound detection signal; According to the feature vector, extracting a valid blade sound signal from the sound detection signal as a signal to be detected; Performing voiceprint feature extraction processing on the signal to be detected to obtain the voiceprint feature of the target fan blade; A blade damage detection model is constructed, wherein the blade damage detection model is trained by taking the soundprint features corresponding to the sample effective blade sound signals of several sample fan blades as input and the damage detection results of each sample fan blade as output; The soundprint feature of the target fan blade is input into the blade damage detection model to perform blade damage detection processing, so as to obtain a damage detection result of the target fan blade after the blade damage detection processing.

2. The method according to claim 1, characterized in that Performing feature extraction processing on the sound detection signal to obtain a feature vector corresponding to the sound detection signal includes: Performing frame processing on the sound detection signal to obtain multiple frames of sound signals; Performing feature extraction processing on each frame of sound signal to obtain a feature vector corresponding to each frame of sound signal, so as to use the feature vector corresponding to each frame of sound signal to form a feature vector corresponding to the sound detection signal; Correspondingly, extracting a valid blade sound signal from the sound detection signal according to the feature vector includes: According to the feature vector corresponding to each frame of the sound signal, each frame of the sound signal is subjected to noise identification processing, so as to filter out the noise signal from the multiple frames of the sound signal; According to the noise signal, an effective signal is intercepted from the sound detection signal to serve as the effective blade sound signal.

3. The method according to claim 2, characterized in that Perform feature extraction on each frame of sound signal to obtain the feature vector corresponding to each frame of sound signal, including: For any frame of sound signal, preprocessing the frame of sound signal to obtain a preprocessed sound signal, wherein the preprocessing includes pre-emphasis processing and windowing processing; Performing wavelet decomposition processing on the preprocessed sound signal to obtain a plurality of wavelet component signals and a wavelet coefficient corresponding to each wavelet component signal; Performing discrete fast Fourier transform processing on each wavelet component signal to obtain the frequency spectrum corresponding to each wavelet component signal; According to the wavelet coefficients corresponding to the wavelet component signals, the frequency spectra corresponding to the wavelet component signals are spliced ​​to obtain the complete frequency spectra of the pre-processed sound signal after the splicing process; Based on the complete spectrum, calculating the spectrum characteristic value of the preprocessed sound signal; The eigenvalues ​​of the frequency spectrum are used to generate a eigenvector of the sound signal of any frame.

4. The method according to claim 3, characterized in that Calculating the frequency spectrum characteristic value of the preprocessed sound signal based on the complete frequency spectrum includes: Acquire several groups of frequency parameters, wherein each group of frequency parameters includes a maximum frequency and a minimum frequency; According to the complete spectrum and several groups of frequency parameters, and in accordance with the following formula (1), the spectrum energy of the complete spectrum under each group of frequency parameters is calculated; In the above formula (1), E(i) represents the spectrum energy of the complete spectrum under the i-th group of frequency parameters, H(j) represents the j-th frequency in the complete spectrum, and F i (j) represents the control parameter corresponding to the jth frequency, f min,i represents the minimum frequency in the i-th group of frequency parameters, f max,i represents the maximum frequency in the i-th group of frequency parameters, where i=1, 2, ..., n, and n represents the total number of groups of frequency parameters; in, In formula (2), k(i), k(i-1), and k(i+1) all represent frequency thresholds; in, In formula (3), T represents a constant, f1 represents the sampling frequency, and F mel () represents the Mel scale function, represents the inverse function of the Mel scale function; The spectrum characteristic value is formed by using the spectrum energy of the complete spectrum under each set of frequency parameters; Correspondingly, using the frequency spectrum feature value to generate a feature vector of any frame of sound signal includes: Get the characteristic order; Based on each spectrum energy in the spectrum characteristic value, and according to the following formula (4), the sound characteristic value of any frame sound signal at each characteristic order is calculated; In the above formula (4), C(m) represents the sound feature value of any frame sound signal at the mth feature order, where m=1, 2, ..., M, and M represents the total order of feature orders; The feature vector of the sound signal of any frame is formed by using the sound feature value of the sound signal of any frame at each feature order.

5. The method according to claim 1, characterized in that A blade damage detection model was constructed, including: Obtaining model parameters of the initial neural network model and initializing the model parameters to obtain multiple sets of initial model parameters; Based on multiple groups of initial model parameters, a longicorn population is generated, wherein each longicorn individual in the longicorn population is configured with an initial position vector and an initial velocity vector, and the initial position vector of each longicorn individual corresponds to a group of initial model parameters; Initialize the number of iterations t to 1, and obtain the position vector and velocity vector of each beetle individual at the t-th iteration, wherein when t is 1, the position vector and velocity vector of any beetle individual at the t-th iteration are the initial position vector and initial velocity vector of any beetle individual; Based on the position vector of each longhorn beetle individual at the t-th iteration, the model parameters of each longhorn beetle individual at the t-th iteration are determined; According to the model parameters of each longhorn beetle individual at the t-th iteration, a neural network model corresponding to each longhorn beetle individual is constructed, and the neural network model corresponding to each longhorn beetle individual is trained using the training data, so as to obtain the root mean square error of the neural network model corresponding to each longhorn beetle individual after the training is completed; The fitness of each individual beetle is calculated by using the root mean square error of the neural network model corresponding to each individual beetle. Based on the fitness of each beetle individual, the global optimal solution and the individual optimal solution of each beetle individual are updated, wherein the global optimal solution is the position vector corresponding to the beetle individual with the largest fitness among all beetle individuals in the first t iterations; Determine whether an iteration stop condition is met, wherein the iteration stop condition is that the fitness of the longicorn individual corresponding to the global optimal solution is greater than or equal to a fitness threshold, or t reaches a maximum number of iterations; If not, the position vector and velocity vector of each longhorn beetle individual at the t-th iteration are used to calculate the left and right antenna positions of each longhorn beetle individual; Using the velocity vector, the left antenna position and the right antenna position of each longhorn beetle individual at the t-th iteration, the position vector of each longhorn beetle individual is updated to obtain the updated position vector of each longhorn beetle individual; Using the global optimal solution and the individual optimal solution of each longhorn beetle individual, the velocity vector of each longhorn beetle individual is updated to obtain the updated velocity vector of each longhorn beetle individual; Add 1 to t, replace the position vector of each longhorn beetle individual at the t-th iteration with the corresponding updated position vector, and replace the velocity vector of each longhorn beetle individual at the t-th iteration with the corresponding updated velocity vector, and re-determine the model parameters of each longhorn beetle individual at the t-th iteration based on the position vector of each longhorn beetle individual at the t-th iteration, until the iteration stop condition is met, and obtain the optimal model parameters based on the global optimal solution when the iteration stop condition is met; The initial neural network model is optimized using the optimal model parameters, so that a blade damage detection model is obtained after the optimization.

6. The method according to claim 5, characterized in that Using the position vector and velocity vector of each longhorn beetle individual at the tth iteration, the left and right antenna positions of each longhorn beetle individual are calculated, including: For any individual beetle, the following formula (5) is used to calculate the left and right tentacle positions of the individual beetle; In the above formula (5), represents the position of the right tentacle of any one of the Tianyan individuals, represents the left tentacle position of any one of the Tianya individuals, X t represents the position vector of any individual Tianniu at the tth iteration, V t represents the velocity vector of any one of the Tianbo individuals at the t-th iteration, and d represents the total distance from the left and right tentacles of any one of the Tianbo individuals to the center of mass of any one of the Tianbo individuals.

7. The method according to claim 5, characterized in that Using the velocity vector, the left antenna position and the right antenna position of each beetle individual at the t-th iteration, the position vector of each beetle individual is updated to obtain the updated position vector of each beetle individual, including: For any individual Tianniu, generate an update random number; Determine whether the individual optimal solution of any one of the Tianniu individuals is greater than the updated random number; If not, generate a first random integer, a second random integer, and a third random integer, wherein the value range of the first random integer, the second random integer, and the third random integer is [1, R], and R represents the total number of longicorn individuals; According to the first random integer, the second random integer and the third random integer, and in accordance with the following formula (6), the position vector of any one of the individual beetles is updated; In the above formula (6), X t+1 represents the updated position vector of any of the individual beetles, represents the position vector of the αth individual in the beetle population at the tth iteration, represents the position vector of the βth individual in the beetle population at the tth iteration, represents the position vector of the δth longicorn individual in the longicorn population at the tth iteration, wherein α is equal to the first random integer, β is equal to the second random integer, δ is equal to the third random integer, and γ represents the variation factor.

8. The method according to claim 7, characterized in that The method further comprises: If yes, then the displacement length of any individual beetle is calculated using the velocity vector, the left tentacle position and the right tentacle position of the individual beetle at the t-th iteration according to the following formula (7); In the above formula (7), represents the displacement length of any individual beetle, μ t represents the moving step length of any Tianniu individual at the tth iteration, V t represents the velocity vector of any individual strychnine at the tth iteration, represents the position of the right tentacle of any one of the Tianyan individuals, represents the left tentacle position of any one of the individual beetles, f() is the fitness function, and sign is the sign function; Based on the displacement length of any Tianniu individual, the position vector and velocity vector of any Tianniu individual at the t-th iteration, and according to the following formula (8), the updated position vector of any Tianniu individual is calculated; In the above formula (8), Represents the position update factor.

9. A wind turbine blade diagnostic device based on voiceprint sensing technology, characterized in that: include: An acquisition unit, used for acquiring a sound detection signal of a target fan blade; A feature extraction unit, used to perform feature extraction processing on the sound detection signal to obtain a feature vector corresponding to the sound detection signal; A sound separation unit, used for extracting a valid blade sound signal from the sound detection signal according to the feature vector as a signal to be detected; The feature extraction unit is further used to perform voiceprint feature extraction processing on the signal to be detected to obtain the voiceprint feature of the target fan blade; A model building unit is used to build a blade damage detection model, wherein the blade damage detection model is trained by taking the soundprint features corresponding to the sample effective blade sound signals of a number of sample fan blades as input and the damage detection results of each sample fan blade as output; The blade detection unit is used to input the soundprint feature of the target fan blade into the blade damage detection model to perform blade damage detection processing, so as to obtain the damage detection result of the target fan blade after the blade damage detection processing.

10. A computer program product comprising instructions, characterized in that When the instruction is executed on a computer, the computer is caused to execute a wind turbine blade diagnosis method based on voiceprint sensing technology as described in any one of claims 1 to 8.

Images