Wind turbine generator blade fault detection method and device

Abstract

The invention relates to a wind turbine generator blade fault detection method and device, relates to the technical field of fault detection, and aims to at least solve the problem of high false alarm rate caused by scarcity of blade operation fault data and difficulty in intelligently identifying blade fault conditions. The method comprises the following steps: acquiring historical state data of normal operation of blades of the wind turbine generator; according to the historical state data, constructing a leaf fingerprint identification model; inputting the current state data of the target blade into the blade fingerprint identification model to obtain a current evaluation result; if it is determined that the current evaluation result is smaller than a preset threshold, it is determined that the target blade is abnormal; and if it is determined that the current evaluation result is greater than or equal to a preset threshold, determining that the target blade is normal.

Description

Technical Field

[0001] This application relates to the field of fault detection technology, and in particular to a method and device for detecting faults in wind turbine blades. Background Technology

[0002] Offshore wind power, as an important component of clean energy, has developed rapidly in recent years. Offshore wind turbines operate in complex and harsh marine environments, facing numerous challenges. As a key component of offshore wind turbines, the monitoring and diagnosis of wind turbine blade faults is of great significance for ensuring the stable operation of offshore wind turbines. However, the complex operating environment of the blades and the presence of a large amount of noise interference make it difficult to effectively extract characteristic data, affecting the accuracy of fault diagnosis. Currently, due to the scarcity of fault samples, it is difficult to intelligently and accurately detect blade faults. Summary of the Invention

[0003] This invention provides a method, apparatus, system, and storage medium for classifying electricity users, to at least address the problem of scarce blade operation fault data, making it difficult to intelligently identify blade fault conditions and resulting in a high false alarm rate. The technical solution of this invention is as follows:

[0004] According to a first aspect of the present invention, a method for detecting faults in wind turbine blades is provided. The method includes: acquiring historical state data of normal operation of wind turbine blades; constructing a blade fingerprint recognition model based on the historical state data; using the blade fingerprint recognition model to positively evaluate the state data based on the normal operation trend represented by the historical state data; inputting the current state data of a target blade into the blade fingerprint recognition model to obtain a current evaluation result; determining that the target blade is abnormal if the current evaluation result is less than a preset threshold; and determining that the target blade is normal if the current evaluation result is greater than or equal to the preset threshold.

[0005] As one implementation method, a blade fingerprint recognition model is constructed based on historical state data, including: performing feature fusion processing on the historical state data to obtain multiple historical fusion feature vectors; constructing a preset model; the preset model contains multiple Gaussian distribution functions; the preset model represents the correlation between operating state data and evaluation results; each Gaussian distribution function includes a probability density function; based on the posterior probabilities of each historical fusion feature vector belonging to each Gaussian distribution function, the function parameters of each Gaussian distribution function in the preset model are updated sequentially until the updated function parameters meet a preset condition, thus obtaining the blade fingerprint recognition model; the preset condition is that the difference between the likelihood function value after two consecutive updates and the likelihood function value before the update is less than a preset threshold; the update operation includes: determining the updated posterior probabilities of each historical fusion feature vector belonging to each Gaussian distribution function based on the updated function parameters of each Gaussian distribution function; updating the updated function parameters of each corresponding Gaussian distribution function based on the updated posterior probabilities, thus obtaining the updated function parameters of each Gaussian distribution function; and determining the updated likelihood function value based on the updated function parameters.

[0006] In this implementation, since referenced abnormal data is scarce, a large amount of historical data under normal operating conditions is used as the basis for model training. By iteratively updating the function parameters of the Gaussian distribution function, the accuracy and authenticity of the model fitting the distribution of training data are improved, so as to ensure that the constructed blade fingerprint recognition model can most accurately capture the complex feature distribution patterns of normal operating data under various operating conditions, thereby obtaining more accurate evaluation results.

[0007] As one implementation method, the relationship between the function parameters and the posterior probability is represented by the following formula: Where p(k|X) n ) represents the posterior probability that the nth historical fusion feature vector belongs to the kth Gaussian distribution function; Let k be the weight of the k-th Gaussian distribution function in the t-th iteration update; Let be the mean of the k-th Gaussian distribution function in the t-th iteration update; Let X be the covariance matrix of the k-th Gaussian distribution function in the t-th iteration update; n There are n historical fusion feature vectors; N is the total number of historical fusion feature vectors; K is the number of Gaussian distribution functions; the relationship between the function parameters and the likelihood function values ​​is represented by the following formula; Among them, L (t) X is the likelihood function value in the t-th iteration update; n There are n historical fusion feature vectors; Let k be the weight of the k-th Gaussian distribution function in the t-th iteration update; Let be the mean of the k-th Gaussian distribution function in the t-th iteration update; Let be the covariance matrix of the k-th Gaussian distribution function in the t-th iteration update; N is the total number of historical fusion feature vectors.

[0008] As one implementation method, historical state data represents multi-dimensional feature data generated during the historical operation of wind turbine blades. Feature fusion processing is performed on the historical state data to obtain a historical fusion feature vector. Specifically, this includes: processing the feature data of each dimension in the historical state data to obtain multi-dimensional state processing data; extracting features from the multi-dimensional state processing data to obtain multi-dimensional signal feature vectors; and fusing the multi-dimensional signal feature vectors based on preset feature weights to obtain the historical fusion feature vector.

[0009] In this implementation, the wind turbine blades operate in a complex noise environment. The blade operating status is monitored comprehensively from multiple dimensions. By extracting signal feature vectors from multiple dimensions in historical status data and fusing these feature vectors, the resulting historical fused feature vector can be regarded as the "digital fingerprint" of the blade, providing reliable data support for subsequent model training, thereby improving the accuracy of fault detection.

[0010] As one implementation method, the feature data of each dimension includes sound feature data, vibration feature data, and temperature feature data; the state processing data of multiple dimensions includes sound feature processing data, vibration feature processing data, and temperature feature processing data; data processing is performed on the feature data of each dimension in the historical state data to obtain state processing data of multiple dimensions, including: determining the sound feature data, vibration feature data, and temperature feature data of each measurement position under various operating conditions of the wind turbine blade; various operating conditions include static operating condition, no-load operating condition, and loaded operating condition; based on multiple sound feature data and multiple vibration feature data, establishing the static reference envelope range under various operating conditions; based on the static reference envelope range, eliminating data in the sound feature data and vibration feature data that exceed the static reference envelope range, and eliminating static measurement errors introduced by changes in different operating conditions, to obtain sound feature processing data and vibration feature processing data; eliminating outliers in multiple temperature feature data, and performing data normalization on the processed temperature feature data to obtain temperature feature processing data.

[0011] In this implementation, by covering the entire scene from the stationary state to different operating states of the equipment, and combining multi-location data acquisition, multi-dimensional feature data including sound, vibration and temperature under different operating conditions are obtained to comprehensively monitor the blade operating status. The multi-dimensional feature data is then processed to reduce noise and remove abnormal data, removing irrelevant interference signals from the data, improving data quality, and providing a more reliable data foundation for subsequent feature extraction and model training.

[0012] As one implementation method, the multi-dimensional signal feature vector includes sound signal feature vector, vibration signal feature vector, and temperature signal feature vector. Feature extraction is performed on the multi-dimensional state processing data to obtain multi-dimensional signal feature vectors, including: extracting acoustic fingerprint features from the sound feature processing data using a preset extraction method; integrating the acoustic fingerprint features to obtain the sound signal feature vector; the preset extraction method includes one or more of the following: wavelet transform, Fourier transform, and time-frequency analysis; extracting time-domain and frequency-domain statistical features from the vibration feature processing data; integrating the time-domain and frequency-domain statistical features to obtain the vibration signal feature vector; the time-domain statistical features include peak value, root mean square, variance, and kurtosis; the frequency-domain statistical features include frequency centroid and frequency variance; and extracting the temperature change trend features from the temperature feature processing data based on the difference between adjacent temperature data points to obtain the temperature signal feature vector.

[0013] As one implementation method, the current state data of the target blade is input into the blade fingerprint recognition model to obtain the current evaluation result, including: performing feature fusion processing on the current state data of the target blade to obtain the current fused feature vector; inputting the current fused feature vector into the blade fingerprint recognition model to obtain the current evaluation result; the current evaluation result characterizes the degree of deviation between the current operating state of the target blade and the normal operating state.

[0014] In this embodiment, by real-time monitoring and fault analysis of blade operating data, minute changes in the blade operating status can be detected in a timely manner.

[0015] As one implementation method, the current status data is observed to obtain the current status label of the current status data; if the current status label indicates that the target blade is abnormal, the current status data is stored in the abnormal soundprint database; if the current status label indicates that the target blade is normal, the current status data is stored in the normal soundprint database.

[0016] In this implementation, the anomaly detection results of the model are verified by observing the running status of the current state data. The current state data is then labeled according to the observation results to determine the data type. The data is then classified and saved to the abnormal voiceprint library or the normal voiceprint library according to the label, providing more accurate and comprehensive data for subsequent model training.

[0017] As one implementation method, the blade fingerprint recognition model is trained based on normal state data from a normal acoustic fingerprint database; and the blade fingerprint recognition model is validated based on abnormal state data from an abnormal acoustic fingerprint database.

[0018] In this implementation, the state data in the normal and abnormal acoustic fingerprint databases have a clear correlation between the state data and the blade operation. This allows for real-time updates of the model parameters, providing more comprehensive and accurate data support for model training and ensuring the model's accuracy.

[0019] According to a second aspect of the present invention, a wind turbine blade fault detection device is provided, the device comprising:

[0020] The data acquisition unit is configured to acquire historical status data of the wind turbine blades during normal operation.

[0021] The model building unit is configured to construct a blade fingerprint recognition model based on historical state data. The blade fingerprint recognition model is used to positively evaluate the state data based on the normal operating trends represented by historical state data.

[0022] The fault identification unit is configured to input the current status data of the target blade into the blade fingerprint recognition model to obtain the current evaluation result; if the current evaluation result is determined to be less than a preset threshold, the target blade is determined to be abnormal; if the current evaluation result is determined to be greater than or equal to the preset threshold, the target blade is determined to be normal.

[0023] According to a third aspect of the present invention, a wind turbine blade fault detection system is provided, the device being configured to perform a wind turbine blade fault detection method as described in the first aspect and any possible implementation thereof.

[0024] According to a fourth aspect of the present invention, a wind turbine blade fault detection device is provided, the device being configured to perform a wind turbine blade fault detection method as described in the first aspect and any possible implementation thereof.

[0025] According to a fifth aspect of the present invention, a computer-readable storage medium is provided, on which instructions are stored, such that when the instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is able to perform a wind turbine blade fault detection method as described in the first aspect and any possible implementation thereof.

[0026] According to a sixth aspect of the present disclosure, a computer program product is provided, the computer program product including computer instructions, which, when executed on an electronic device, cause the electronic device to perform the wind turbine blade fault detection method described in the first aspect and any possible implementation thereof.

[0027] The technical solution provided by the embodiments of the present invention brings at least the following beneficial effects: With a small amount of fault sample data to support it, in order to ensure the accuracy of operating status identification, a large amount of state data of wind turbine blades under normal operating conditions is used for training to build a blade fingerprint recognition model for wind turbine blades under normal operating conditions. This allows the model to fully learn the complex situations under normal blade conditions, thereby better distinguishing between normal fluctuations caused by changes in environment or operating conditions and abnormal signals of actual structural faults, thus significantly reducing the false alarm rate. Based on this, by identifying the deviation between the evaluation results output by the blade fingerprint recognition model and the preset threshold, minor faults occurring on the blades can be identified in a timely and accurate manner, improving the accuracy of fault detection.

[0028] 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

[0029] 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.

[0030] Figure 1 This is a schematic diagram of a wind turbine blade fault detection system according to an exemplary embodiment;

[0031] Figure 2 This is a flowchart illustrating a wind turbine blade fault detection method according to an exemplary embodiment. Figure 1 ;

[0032] Figure 3 This is a flowchart illustrating a wind turbine blade fault detection method according to an exemplary embodiment. Figure 2 ;

[0033] Figure 4 This is a flowchart illustrating a wind turbine blade fault detection method according to an exemplary embodiment. Figure 3 ;

[0034] Figure 5 This is a block diagram illustrating a wind turbine blade fault detection device according to an exemplary embodiment;

[0035] Figure 6 This is a schematic diagram of a wind turbine blade fault detection device according to an exemplary embodiment. Detailed Implementation

[0036] 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.

[0037] 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.

[0038] Before providing a detailed introduction to the electricity user classification method provided in the embodiments of this application, let's briefly introduce the application scenarios and implementation environment involved in the embodiments of this application.

[0039] First, a brief introduction to the application scenarios involved in this application will be given.

[0040] Offshore wind power, as an important component of clean energy, has developed rapidly in recent years. Offshore wind turbines operate in complex and harsh marine environments, facing numerous challenges. As a key component of offshore wind turbines, the monitoring and diagnosis of wind turbine blade faults is of great significance for ensuring the stable operation of offshore wind turbines. However, the complex operating environment of the blades and the presence of a large amount of noise interference make it difficult to effectively extract characteristic data, affecting the accuracy of fault diagnosis. Currently, due to the scarcity of fault samples, it is difficult to intelligently and accurately detect blade faults.

[0041] To address the aforementioned issues, this application proposes a method for detecting wind turbine blade faults. Even with limited fault sample data, to ensure accuracy in identifying operating conditions, a large amount of state data from wind turbine blades under normal operating conditions is used for training. This constructs a blade fingerprint recognition model for the wind turbine blades under normal operating conditions, allowing the model to fully learn the complexities of normal blade conditions. This enables it to better distinguish between normal fluctuations caused by environmental or operating condition changes and abnormal signals indicating actual structural faults, thereby significantly reducing the false alarm rate. Based on this, by identifying the deviation between the evaluation results output by the blade fingerprint recognition model and a preset threshold, minor faults in the blades can be identified promptly and accurately, improving the accuracy of fault detection.

[0042] Secondly, the implementation architecture involved in this application will be briefly introduced below.

[0043] Figure 1 This is a schematic diagram of a wind turbine blade fault detection system provided in this application. Figure 1As shown, the wind turbine blade fault detection system includes a multi-dimensional feature fusion module 101, a model training module 102, a fault detection module 103, and a status data labeling module 104.

[0044] The multi-dimensional feature fusion module 101, model training module 102, fault detection module 103, and status data labeling module 104 are connected via communication.

[0045] The multi-dimensional feature fusion module 101 is configured to process the feature data of each dimension in the state data to obtain multi-dimensional state processing data; extract features from the multi-dimensional state processing data to obtain multi-dimensional signal feature vectors; and fuse the multi-dimensional signal feature vectors based on preset feature weights to obtain fused feature vectors.

[0046] The model training module 102 is configured to update the function parameters of each Gaussian distribution function in the preset model one by one according to the posterior probability of each historical fused feature vector belonging to each Gaussian distribution function, until the updated function parameters meet the preset conditions, and thus obtain the leaf fingerprint recognition model; the preset condition is that the function difference between the likelihood function value after two adjacent updates and the likelihood function value before the update is less than a preset threshold.

[0047] The fault detection module 103 is configured to perform feature fusion processing on the current state data of the target blade to obtain the current fused feature vector; input the current fused feature vector into the blade fingerprint recognition model to obtain the current evaluation result; the current evaluation result represents the degree of deviation between the current operating state of the target blade and the normal operating state; if the current evaluation result is determined to be less than a preset threshold, the target blade is determined to be abnormal; if the current evaluation result is determined to be greater than or equal to the preset threshold, the target blade is determined to be normal.

[0048] The status data label module 104 is configured to observe the running status of the current status data and obtain the current status label of the current status data; if the current status label indicates that the target blade is abnormal, the current status data is stored in the abnormal acoustic fingerprint library; if the current status label indicates that the target blade is normal, the current status data is stored in the normal acoustic fingerprint library. The blade fingerprint recognition model is trained based on the normal status data in the normal acoustic fingerprint library; the blade fingerprint recognition model is validated based on the abnormal status data in the abnormal acoustic fingerprint library.

[0049] Figure 2 This is a flowchart illustrating a wind turbine blade fault detection method according to an exemplary embodiment, such as... Figure 2 As shown, the method for detecting wind turbine blade faults includes the following steps.

[0050] S21, acquire historical status data of normal operation of wind turbine blades.

[0051] Historical status data characterizes the multi-dimensional feature data generated during the historical operation of wind turbine blades.

[0052] Multidimensional feature data includes sound feature data, vibration feature data, and temperature feature data.

[0053] To ensure the accuracy of operational status identification, even with limited fault sample data, it is necessary to acquire a large amount of operational status data of wind turbine blades under normal operating conditions, providing data support for subsequent model training.

[0054] In one implementation, an intelligent data acquisition device containing multiple sensors is used to synchronously monitor the sound, vibration, and temperature of wind turbine blades in an integrated manner, thereby obtaining the status data of offshore wind turbine blades during operation.

[0055] Since wind turbine blades generate a lot of noise during operation, feature extraction and feature fusion processing of the acquired historical state data are required before model training to improve data quality and provide a more reliable data foundation for subsequent model training and fault detection.

[0056] Furthermore, feature fusion processing is performed on the historical state data to obtain a historical fused feature vector. For example... Figure 3 As shown, the determination of the historical fusion feature vector is carried out in specific ways according to steps S211 to S213.

[0057] S211, respectively, performs data processing on the feature data of each dimension in the historical state data to obtain state processing data of multiple dimensions.

[0058] The state processing data includes multiple dimensions: sound feature processing data, vibration feature processing data, and temperature feature processing data.

[0059] Specifically, the sound characteristics, vibration characteristics, and temperature characteristics of the wind turbine blades at various measurement locations under multiple operating conditions are determined; these multiple operating conditions include static operating conditions, no-load operating conditions, and loaded operating conditions.

[0060] Based on multiple sound characteristic data and multiple vibration characteristic data, a static reference envelope range is established under various working conditions.

[0061] Based on the static reference envelope range, data exceeding the static reference envelope range in the sound and vibration feature data are eliminated, as well as static measurement errors introduced by changes in different operating conditions, to obtain sound feature processing data and vibration feature processing data.

[0062] Outliers in multiple temperature feature data are eliminated, and the processed temperature feature data is then normalized to obtain processed temperature feature data.

[0063] Among them, the static operating condition represents that the wind turbine blades are not started and are in a stopped state; the no-load operating condition represents that the wind turbine blades are started but not connected to the actual load; and the loaded operating condition represents that the wind turbine blades are connected to the actual working load.

[0064] Each measurement location needs to be optimized in advance to determine the specific key parts of the wind turbine blades and the key nodes of the pitch and yaw systems for testing, in order to obtain accurate and representative data.

[0065] Understandably, based on the state data obtained in step S21 above, by covering the entire scenario of wind turbine blades from static to different operating states, and combining multi-location data acquisition, a comprehensive evaluation of equipment performance and problem localization can be achieved, thereby comprehensively reflecting the overall and local operating status of wind turbine blades. Simultaneously, the acquired multi-dimensional feature data under different operating conditions, including sound, vibration, and temperature, undergoes noise reduction and outlier removal processing to eliminate irrelevant interference signals, improve data quality, and provide a more reliable data foundation for subsequent feature extraction and model training.

[0066] S212 extracts features from the state processing data of multiple dimensions to obtain signal feature vectors of multiple dimensions.

[0067] Multidimensional signal feature vectors include sound signal feature vectors, vibration signal feature vectors, and temperature signal feature vectors.

[0068] In some implementations, firstly, acoustic fingerprint features are extracted from the sound feature processing data using a preset extraction method; then, the acoustic fingerprint features are integrated and processed to obtain a sound signal feature vector; the preset extraction method includes one or more of the following: wavelet transform, Fourier transform, and time-frequency analysis.

[0069] Among them, acoustic fingerprint features include coefficient matrix, acoustic audio domain data, and energy distribution of sound feature processing data in time and frequency.

[0070] Specifically, firstly, the coefficient matrix of acoustic fingerprint features is determined.

[0071] Based on the preset wavelet function and the collected samples of sound feature processing data, wavelet transform is performed on the sound feature processing data to obtain the coefficient matrix of acoustic fingerprint features, as shown in the following formula (1).

[0072]

[0073] Among them, W S(a,b) is the coefficient matrix of the acoustic fingerprint feature; a is the scaling parameter; b is the translation parameter; ψ() is the wavelet function; the acoustic feature processing data is S={s i}, i = 1, 2, ..., H, where H is the number of sample points for sound feature processing data; S i Process the data for the i-th sound feature.

[0074] Secondly, determine the amplitude and phase of the acoustic fingerprint features.

[0075] The acoustic feature processing data is subjected to Fast Fourier Transform to determine the acoustic frequency domain data of the acoustic fingerprint features. The acoustic frequency domain data includes amplitude and phase. Specifically, it is shown in the following formula (2).

[0076]

[0077] Among them, F S (k1) represents the acoustic frequency domain data of the acoustic fingerprint feature; k1 is the frequency index; It is a complex exponential function; j is the imaginary unit, satisfying j 2 =-1; S i H represents the i-th sound feature processing data; H is the number of sample points collected for the sound feature processing data.

[0078] Third, determine the energy distribution of acoustic fingerprint feature processing data in time and frequency.

[0079] Based on the preset window function and the collected samples of sound feature processing data, time-frequency analysis is performed on the sound feature processing data to determine a two-dimensional function. The two-dimensional function represents the energy distribution of the sound feature processing data in time and frequency. Specifically, it is shown in the following formula (3).

[0080] STFT S (m,ω)=∑s i w(nm)e -jωi (3).

[0081] Among them, STFT S (m,ω) is a two-dimensional function representing the energy distribution of sound feature processing data in time and frequency; w() is a window function; m is the shift of the time window, n is the time index; ω is the frequency; S i Process the data for the i-th sound feature.

[0082] Secondly, the time-domain and frequency-domain statistical features of the vibration feature processing data are extracted; the time-domain and frequency-domain statistical features are integrated and processed to obtain the vibration signal feature vector; the time-domain statistical features include peak value, root mean square, variance, and kurtosis; the frequency-domain statistical features include frequency centroid and frequency variance.

[0083] According to formulas (4) to (8), data statistics are performed on the vibration feature processing data to determine the time domain statistical characteristics and obtain the peak value, root mean square, variance, and kurtosis of the vibration signal feature vector.

[0084] According to formulas (9) and (10), the vibration feature processing data is subjected to fast Fourier transform to determine the frequency domain statistical features and obtain the frequency centroid and frequency variance of the vibration signal feature vector.

[0085] First, determine the peak value of the vibration signal feature vector, as shown in formula (4).

[0086] P V =max{v m}(4).

[0087] Among them, P V V = {v} is the peak value of the characteristic vector of the vibration signal. m}, where m = 1, 2, ... is the vibration feature processing data sequence.

[0088] Secondly, determine the root mean square in the time-domain statistical characteristics, as shown in formula (5).

[0089]

[0090] Among them, RMS V V is the root mean square of the vibration signal feature vector; M is the total number of samples collected for vibration feature processing; v m This is the data processing for the m-th vibration feature.

[0091] Third, determine the variance in the time-domain statistical characteristics, as shown in formulas (6) and (7).

[0092]

[0093] Among them, VAR V The variance of the vibration signal's eigenvectors; The mean of the vibration characteristic processing data; v m This represents the m-th vibration feature processing data; M is the total number of samples collected for vibration feature processing data.

[0094]

[0095] in, The mean of the vibration characteristic processing data; v m This represents the m-th vibration feature processing data; M is the total number of samples collected for vibration feature processing data.

[0096] Fourth, determine the kurtosis in the time-domain statistical features, as shown in formula (8).

[0097]

[0098] Among them, K V The kurtosis of the vibration signal feature vector; The mean of the vibration feature processing data; VAR V V is the variance of the vibration signal feature vector; M is the total number of samples collected for vibration feature processing data; v m This is the data processing for the m-th vibration feature.

[0099] Fifth, determine the frequency centroid in the frequency domain statistical characteristics, as shown in formula (9).

[0100]

[0101] Among them, FC V F is the centroid of the frequency of the characteristic vector of the vibration signal; V (k2) represents the vibration frequency domain data; k2 is the frequency index.

[0102] Sixth, determine the frequency variance in the frequency domain statistical characteristics, as shown in formula (10).

[0103]

[0104] Among them, FV v FC is the frequency variance of the eigenvector of the vibration signal. V F is the centroid of the frequency of the vibration signal characteristic vector; V (k2) represents the vibration frequency domain data; k2 is the frequency index.

[0105] Finally, based on the difference between adjacent temperature data points, the temperature change trend features of the temperature feature processing data are extracted to obtain the temperature signal feature vector.

[0106] S213, based on preset feature weights, performs feature fusion on signal feature vectors of multiple dimensions to obtain historical fused feature vectors.

[0107] Specifically, the sound signal feature vector, vibration signal feature vector, and temperature signal feature vector are weighted according to preset feature weights and then fused using a weighted summation method to obtain a multi-dimensional historical fused feature vector.

[0108] The preset feature weights are determined based on experience or data analysis to ascertain the importance of sound, vibration, and temperature features in characterizing blade failures.

[0109] S22, Construct a blade fingerprint recognition model based on historical state data; the blade fingerprint recognition model is used to positively evaluate the state data based on the normal operation trend represented by historical state data.

[0110] In one implementation, based on the above step S21, feature fusion processing is performed on the historical state data to obtain a historical fusion feature vector.

[0111] Among them, there are multiple historical fusion feature vectors.

[0112] First, construct a pre-defined model.

[0113] The preset model contains multiple Gaussian distribution functions. Each Gaussian distribution function includes a probability density function. Each Gaussian distribution function corresponds to a different weight.

[0114] The pre-defined model represents the correlation between operational status data and evaluation results.

[0115] Secondly, based on the posterior probabilities of each historical fusion feature vector belonging to each Gaussian distribution function, the function parameters of each Gaussian distribution function in the preset model are updated one by one until the updated function parameters meet the preset conditions, thus obtaining the leaf fingerprint recognition model; the preset condition is that the function difference between the likelihood function value after two adjacent updates and the likelihood function value before the update is less than a preset threshold.

[0116] Finally, the function parameters that meet the preset conditions are determined as the final target function parameters, thereby obtaining the blade fingerprint recognition model.

[0117] The update operation includes: determining the updated posterior probability of each historical fused feature vector belonging to each Gaussian distribution function based on the function parameters of each Gaussian distribution function before the update; updating the function parameters of each corresponding Gaussian distribution function before the update based on the updated posterior probabilities to obtain the updated function parameters of each Gaussian distribution function; and determining the updated likelihood function value based on the updated function parameters.

[0118] In one implementation, such as Figure 4 As shown, the update operation is specifically implemented according to steps S221 to S225.

[0119] Multiple historical fusion feature vectors are used as training samples and labeled as X={X n}, n = 1, 2, ..., N, where N is the total number of historical fused feature vectors used as training samples.

[0120] The default model is a Gaussian mixture model, which contains K Gaussian distribution functions, and the initial function parameters of each Gaussian distribution function are randomly selected. The function parameters include weights, covariance matrix and mean.

[0121] S221, determine the posterior probability of each training sample belonging to each Gaussian distribution function.

[0122] The posterior probability is determined according to the following formulas (11) and (12).

[0123] Based on the dimensions of the mean, covariance matrix, and fused eigenvector, the probability density function of the Gaussian distribution function is determined as shown in the following formula (11).

[0124]

[0125] in, Let be the probability density function of the k-th Gaussian distribution function at the n-th historical fused feature vector; d is the dimension of the fused feature vector; Let be the mean of the k-th Gaussian distribution function at the t-th iteration; Let be the covariance matrix of the k-th Gaussian distribution function at the t-th iteration; is the determinant of the covariance matrix; exp() is an exponential function with the natural constant e as its base; This is the transpose of the matrix.

[0126] Based on the weights and the probability density function of the Gaussian distribution function, the posterior probability is determined as shown in the following formula (12).

[0127]

[0128] Where p(k|X) n ) represents the posterior probability that the nth fused feature vector belongs to the kth Gaussian distribution function; K is the number of Gaussian distribution functions; t indicates that the current iteration is the tth iteration; Let be the probability density function of the k-th Gaussian distribution function at the n-th historical fusion feature vector; represents the weight of the k-th Gaussian distribution function at the t-th iteration; N is the total number of historical fusion feature vectors used as training samples.

[0129] S222, based on the posterior probability, update the function parameters of the Gaussian distribution function to obtain the updated function parameters.

[0130] Based on formulas (13) to (15), determine the updated function parameters, including the updated weights, the updated mean, and the updated covariance matrix.

[0131] First, determine the updated weights.

[0132] The updated weights are determined by the ratio of the sum of the posterior probabilities of the kth Gaussian distribution function to the total number of training samples, as shown in the following formula (13).

[0133]

[0134] in, The weights of the k-th Gaussian distribution function at the (t+1)-th iteration are the updated weights; p(k|X) n X is the posterior probability that the nth fused feature vector belongs to the kth Gaussian distribution function; N is the total number of historical fused feature vectors used as training samples; X n This is the nth fused feature vector.

[0135] Secondly, determine the updated mean.

[0136] Based on the posterior probability, a weighted average is calculated on all training samples to determine the updated mean, as shown in the following formula (14).

[0137]

[0138] in, p(k|X) is the mean of the k-th Gaussian distribution function at the (t+1)-th iteration, i.e., the updated mean; n X is the posterior probability that the nth fused feature vector belongs to the kth Gaussian distribution function; n This is the nth fused feature vector.

[0139] Third, determine the updated covariance matrix.

[0140] Based on the difference between the training sample and the updated mean, and the posterior probability, the updated covariance matrix is ​​determined as shown in the following formula (15).

[0141]

[0142] in, Let be the covariance matrix of the k-th Gaussian distribution function at the (t+1)-th iteration, i.e., the updated covariance matrix; p(k|X) is the updated mean; n X is the posterior probability that the nth fused feature vector belongs to the kth Gaussian distribution function; n This is the nth fused feature vector.

[0143] S223, Determine the likelihood function value based on the updated function parameters.

[0144] According to formula (16), the likelihood function value is determined based on the updated weights, the updated mean, and the updated covariance matrix.

[0145]

[0146] Among them, L (t) Let be the likelihood function value of the model corresponding to the t-th iteration; Let be the mean of the k-th Gaussian distribution function at the (t+1)-th iteration; Let be the covariance matrix of the k-th Gaussian distribution function at the (t+1)-th iteration; X represents the weight of the k-th Gaussian distribution function at the (t+1)-th iteration; n Let N be the nth fused feature vector; N is the total number of historical fused feature vectors used as training samples.

[0147] S224, iteratively update the function parameters and determine the updated likelihood function value.

[0148] Repeat step S221 above, using the updated weights, updated mean, and updated covariance matrix as function parameters of the Gaussian distribution function to determine the updated posterior probability.

[0149] Repeat step S222 above, and update the function parameters of the Gaussian distribution function again based on the updated posterior probability.

[0150] Repeat step S223 above to obtain the updated likelihood function value based on the updated function parameters.

[0151] The iterative update continues until the difference between the updated likelihood function value and the previous likelihood function value is less than a preset threshold, as shown in formula (17).

[0152] ||L (t+1) -L (t) ||<ε (17).

[0153] Among them, L (t) Let L be the likelihood function value of the model corresponding to the t-th iteration, i.e., the likelihood function value before the update; (t+1) ε is the likelihood function value of the model corresponding to the (t+1)th iteration; that is, the updated likelihood function value; ε is the change threshold.

[0154] A change threshold ε>0 indicates that the parameter changes very little after one iteration, and is usually set to a small value.

[0155] S225, determine the function parameters corresponding to the iteration stop as the target parameters, and obtain the acoustic fingerprint recognition model.

[0156] In this implementation, since referenced abnormal data is scarce, a large amount of historical data under normal operating conditions is used as the basis for model training. By iteratively updating the function parameters of the Gaussian distribution function, the accuracy and authenticity of the model fitting the distribution of training data are improved, so as to ensure that the constructed blade fingerprint recognition model can most accurately capture the complex feature distribution patterns of normal operating data under various operating conditions, thereby obtaining more accurate evaluation results.

[0157] S23, input the current state data of the target blade into the blade fingerprint recognition model to obtain the current evaluation result.

[0158] Specifically, the current state data of the target blade is subjected to feature fusion processing to obtain a current fused feature vector. This current fused feature vector is then input into the blade fingerprint recognition model to obtain the current evaluation result. The current evaluation result characterizes the degree of deviation between the current operating state of the target blade and its normal operating state.

[0159] If the current evaluation result is determined to be less than the preset threshold, then the target blade is determined to be abnormal.

[0160] If the current evaluation result is determined to be greater than or equal to the preset threshold, then the target blade is determined to be normal.

[0161] In one implementation, the preset threshold is set to 3a by statistically analyzing the comprehensive scores of all historical fusion feature vectors used as training samples and determining their mean as a.

[0162] The comprehensive score is obtained by using the probability density function of each fused feature vector under the Gaussian subvector K as the comprehensive scoring function. This comprehensive score is used as the evaluation result output by the leaf fingerprint recognition model.

[0163] The real-time fused feature vector will be obtained and input into the acoustic fingerprint recognition model for scoring to obtain the current evaluation result.

[0164] Optionally, observe the running status of the current status data to obtain the current status label of the current status data.

[0165] If the current status label indicates that the target blade is abnormal, then the current status data is stored in the abnormal acoustic signature database. If the current status label indicates that the target blade is normal, then the current status data is stored in the normal acoustic signature database.

[0166] In this implementation, the accuracy of the anomaly detection result determined by the evaluation result output by the acoustic fingerprint recognition model is verified by observing the current state data. The current state data is then labeled according to the observation results to determine the data type. The data is then classified and saved to the abnormal voiceprint library or the normal voiceprint library according to the label, providing more accurate and comprehensive data for subsequent model training.

[0167] Optionally, the blade fingerprint recognition model is trained based on normal state data from a normal acoustic fingerprint database. The blade fingerprint recognition model is then validated based on abnormal state data from an abnormal acoustic fingerprint database.

[0168] In this implementation, the state data in the normal and abnormal acoustic fingerprint databases have a clear correlation between the state data and the blade operation. This allows for real-time updates to the model parameters, providing more comprehensive and accurate data support for model training and thus improving the model's accuracy.

[0169] In one implementation, a blade acoustic signature profile of the wind turbine blade is established by integrating normal state data from a normal acoustic signature database and abnormal state data from an abnormal acoustic signature database. By comparing the state data generated during operation with the blade acoustic signature profile, acoustic signature anomalies such as those occurring during start-up, shutdown, and stable operation are detected and alarms are issued.

[0170] The overall accuracy of the current blade acoustic signature is determined by comparing the fault detection results with the real-time status labels obtained periodically.

[0171] The overall accuracy is compared with a standard threshold. If the overall accuracy is greater than or equal to the standard threshold, the fault detection method meets the requirements for detecting abnormal operating conditions. If the overall accuracy is less than the standard threshold, the fault detection method does not meet the requirements for detecting abnormal operating conditions, and the blade fingerprint recognition model parameters need to be adjusted or the equipment acoustic print profile data needs to be supplemented. The fault detection method characterizes the detection of wind turbine blade faults using a blade fingerprint recognition model.

[0172] The testing cycle is set at one year. The standard threshold is set at 85%.

[0173] Specifically, the overall accuracy of the leaf acoustic image is determined based on real-time status comparison data, as shown in the following formula (18).

[0174]

[0175] Among them, ZQ is the overall accuracy of the blade acoustic print profile; TP is the real-time status data that is actually abnormal and judged as abnormal; FP is the real-time status data that is actually normal but judged as abnormal; TN is the real-time status data that is actually normal and judged as normal; and FN is the real-time status data that is actually abnormal but judged as normal.

[0176] In this embodiment, the accuracy of the fault detection method is verified by periodically testing the overall accuracy of the blade acoustic signature.

[0177] To achieve the above functions, the electricity user classification device includes hardware structures and / or software modules corresponding to the execution of each function. Those skilled in the art will readily recognize that, based on the algorithmic steps of the examples described in conjunction with the embodiments disclosed herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0178] This disclosure also provides an embodiment such as Figure 5 The wind turbine blade fault detection device shown includes: a data acquisition unit 301, a model building unit 302, and a fault identification unit 303.

[0179] The data acquisition unit 301 is configured to acquire historical status data of the normal operation of the wind turbine blades.

[0180] The model building unit 302 is configured to build a blade fingerprint recognition model based on historical state data; the blade fingerprint recognition model is used to positively evaluate the state data based on the normal operation trend represented by the historical state data.

[0181] The fault identification unit 303 is configured to input the current status data of the target blade into the blade fingerprint identification model to obtain the current evaluation result; if the current evaluation result is determined to be less than a preset threshold, the target blade is determined to be abnormal; if the current evaluation result is determined to be greater than or equal to the preset threshold, the target blade is determined to be normal.

[0182] As one implementation method, the model building unit 302 is specifically configured to construct a blade fingerprint recognition model based on historical state data, including: performing feature fusion processing on the historical state data to obtain multiple historical fusion feature vectors; constructing a preset model; the preset model contains multiple Gaussian distribution functions; the preset model represents the correlation between operating state data and evaluation results; each Gaussian distribution function includes a probability density function; based on the posterior probabilities of each historical fusion feature vector belonging to each Gaussian distribution function, the function parameters of each Gaussian distribution function in the preset model are updated sequentially until the updated function parameters meet a preset condition; the preset condition is that the difference between the likelihood function value after two consecutive updates and the likelihood function value before the update is less than a preset threshold; the update operation includes: determining the updated posterior probabilities of each historical fusion feature vector belonging to each Gaussian distribution function based on the updated function parameters of each Gaussian distribution function; updating the updated function parameters of each corresponding Gaussian distribution function based on the updated posterior probabilities to obtain the updated function parameters of each Gaussian distribution function; and determining the updated likelihood function value based on the updated function parameters.

[0183] As one implementation method, the model building unit 302 is specifically configured such that the relationship between the function parameters and the posterior probability is represented by the following formula: Where p(k|X) n ) represents the posterior probability that the nth historical fusion feature vector belongs to the kth Gaussian distribution function; Let k be the weight of the k-th Gaussian distribution function in the t-th iteration update; Let be the mean of the k-th Gaussian distribution function in the t-th iteration update; Let X be the covariance matrix of the k-th Gaussian distribution function in the t-th iteration update; n There are n historical fusion feature vectors; N is the total number of historical fusion feature vectors; K is the number of Gaussian distribution functions; the relationship between the function parameters and the likelihood function values ​​is represented by the following formula; Among them, L (t) X is the likelihood function value in the t-th iteration update; n There are n historical fusion feature vectors; Let k be the weight of the k-th Gaussian distribution function in the t-th iteration update; Let be the mean of the k-th Gaussian distribution function in the t-th iteration update; Let be the covariance matrix of the k-th Gaussian distribution function in the t-th iteration update; N is the total number of historical fusion feature vectors.

[0184] As one implementation method, the model building unit 302 is specifically configured as follows: historical state data represents multi-dimensional feature data generated by the wind turbine blades during historical operation; feature fusion processing is performed on the historical state data to obtain a historical fusion feature vector, specifically including: data processing of feature data of each dimension in the historical state data to obtain multi-dimensional state processing data; feature extraction of the multi-dimensional state processing data to obtain multi-dimensional signal feature vectors; and feature fusion of the multi-dimensional signal feature vectors based on preset feature weights to obtain a historical fusion feature vector.

[0185] As one implementation method, the model building unit 302 is specifically configured as follows: feature data of each dimension includes sound feature data, vibration feature data, and temperature feature data; state processing data of multiple dimensions includes sound feature processing data, vibration feature processing data, and temperature feature processing data; data processing is performed on the feature data of each dimension in the historical state data to obtain state processing data of multiple dimensions, including: determining the sound feature data, vibration feature data, and temperature feature data of each measurement position under various operating conditions of the wind turbine blades; various operating conditions include static operating condition, no-load operating condition, and loaded operating condition; based on multiple sound feature data and multiple vibration feature data, establishing the static reference envelope range under various operating conditions; based on the static reference envelope range, eliminating data in the sound feature data and vibration feature data that exceeds the static reference envelope range, and eliminating static measurement errors introduced by changes in different operating conditions to obtain sound feature processing data and vibration feature processing data; eliminating outliers in multiple temperature feature data, and performing data normalization processing on the processed temperature feature data to obtain temperature feature processing data.

[0186] As one implementation method, the model building unit 302 is specifically configured as follows: Multiple-dimensional signal feature vectors include sound signal feature vectors, vibration signal feature vectors, and temperature signal feature vectors; feature extraction is performed on the multiple-dimensional state processing data to obtain multiple-dimensional signal feature vectors, including: extracting acoustic fingerprint features from the sound feature processing data using a preset extraction method; integrating the acoustic fingerprint features to obtain the sound signal feature vector; the preset extraction method includes one or more of the following: wavelet transform, Fourier transform, and time-frequency analysis; extracting time-domain and frequency-domain statistical features from the vibration feature processing data; integrating the time-domain and frequency-domain statistical features to obtain the vibration signal feature vector; the time-domain statistical features include peak value, root mean square, variance, and kurtosis; the frequency-domain statistical features include frequency centroid and frequency variance; and extracting the temperature change trend features from the temperature feature processing data based on the difference between adjacent temperature data points to obtain the temperature signal feature vector.

[0187] As one implementation method, the fault identification unit 303 is specifically configured to input the current state data of the target blade into the blade fingerprint recognition model to obtain the current evaluation result, including: performing feature fusion processing on the current state data of the target blade to obtain the current fused feature vector; inputting the current fused feature vector into the blade fingerprint recognition model to obtain the current evaluation result; the current evaluation result characterizes the degree of deviation between the current operating state of the target blade and the normal operating state.

[0188] As one implementation method, the fault identification unit 303 is specifically configured to observe the operating status of the current status data and obtain the current status label of the current status data; if the current status label indicates that the target blade is abnormal, the current status data is stored in the abnormal soundprint library; if the current status label indicates that the target blade is normal, the current status data is stored in the normal soundprint library.

[0189] As one implementation method, the fault identification unit 303 is specifically configured to train the blade fingerprint recognition model based on normal state data from the normal acoustic fingerprint library; and to verify the blade fingerprint recognition model based on abnormal state data from the abnormal acoustic fingerprint library.

[0190] Regarding the apparatus in the above embodiments, the specific manner in which each unit module performs its operations has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0191] Figure 6 This is a schematic diagram of a wind turbine blade fault detection device provided in this application. Figure 6 The electronic device 50 may include at least one processor 501 and a memory 503 for storing processor-executable instructions. The processor 501 is configured to execute the instructions in the memory 503 to implement the wind turbine blade fault detection method in the following embodiments.

[0192] In addition, electronic device 50 may also include communication bus 502, at least one communication interface 504, input device 506 and output device 505.

[0193] The processor 501 may be a processor (central processing unit, CPU), a microprocessor unit, an ASIC, or one or more integrated circuits for controlling the execution of the program of the present application.

[0194] The communication bus 502 may include a path for transmitting information between the aforementioned components.

[0195] Communication interface 504 uses any transceiver-like device for communicating with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0196] Input device 506 is used to receive input signals and output device 505 is used to output signals.

[0197] Memory 503 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital versatile optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. Memory may exist independently and be connected to the processing unit via a bus. Memory may also be integrated with the processing unit.

[0198] The memory 503 stores instructions for executing the scheme of this application, and the processor 501 controls the execution. The processor 501 executes the instructions stored in the memory 503 to implement the functions of the method of this application.

[0199] In a specific implementation, as one example, the processor 501 may include one or more CPUs, for example... Figure 6 CPU0 and CPU1 in the CPU.

[0200] In a specific implementation, as one example, the electronic device 50 may include multiple processors, such as... Figure 6 Processors 501 and 507 are shown in the diagram. Each of these processors can be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor. A processor here can refer to one or more devices, circuits, and / or processing cores used to process data (such as computer program instructions).

[0201] The wind turbine blade fault detection equipment, such as Figure 6 The diagram includes a processor 501 and a memory 503 for storing executable instructions of the processor 501. The processor 501 is configured to execute the executable instructions to implement the wind turbine blade fault detection method as described in any of the possible embodiments above. Since the same technical effects can be achieved, further details are omitted here to avoid repetition.

[0202] This application also provides a computer-readable storage medium. When the instructions in the computer-readable storage medium are executed by the processor of the wind turbine blade fault detection device or the wind turbine blade fault detection equipment, the wind turbine blade fault detection device or the wind turbine blade fault detection equipment can perform the wind turbine blade fault detection method as described in any of the above possible embodiments. And it can achieve the same technical effect; to avoid repetition, it will not be described again here.

[0203] This application also provides a computer program product, including a computer program or instructions, which are executed by a processor as described in any of the possible implementations of the wind turbine blade fault detection method above. This achieves the same technical effect, and to avoid repetition, it will not be described again here.

[0204] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application 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 application are indicated by the following claims.

[0205] It should be understood that this application is not limited to the precise structure 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 application is limited only by the appended claims.

Claims

1. A method for detecting faults in wind turbine blades, characterized in that, The method includes: Acquire historical status data of wind turbine blades during normal operation; Based on the historical state data, a blade fingerprint recognition model is constructed; the blade fingerprint recognition model is used to positively evaluate the state data based on the normal operation trend represented by the historical state data. The current state data of the target blade is input into the blade fingerprint recognition model to obtain the current evaluation result; If the current evaluation result is determined to be less than a preset threshold, the target blade is determined to be abnormal; if the current evaluation result is determined to be greater than or equal to the preset threshold, the target blade is determined to be normal.

2. The method for detecting wind turbine blade faults according to claim 1, characterized in that, The step of constructing a blade fingerprint recognition model based on the historical state data includes: The historical state data is subjected to feature fusion processing to obtain historical fusion feature vectors, and there are multiple historical fusion feature vectors. A preset model is constructed; the preset model contains multiple Gaussian distribution functions; the preset model represents the correlation between operating status data and evaluation results; each Gaussian distribution function includes a probability density function; Based on the posterior probabilities of each of the historical fusion feature vectors belonging to each of the Gaussian distribution functions, the function parameters of each Gaussian distribution function in the preset model are updated one by one until the updated function parameters meet the preset conditions, thereby obtaining the leaf fingerprint recognition model; the preset conditions are that the function difference between the likelihood function value after two consecutive updates and the likelihood function value before the update is less than a preset threshold. The update operation includes: determining the updated posterior probability of each historical fused feature vector belonging to each Gaussian distribution function based on the function parameters of each Gaussian distribution function before the update; updating the function parameters of each corresponding Gaussian distribution function before the update based on the updated posterior probabilities to obtain the updated function parameters of each Gaussian distribution function; and determining the updated likelihood function value based on the updated function parameters.

3. The wind turbine blade fault detection method according to claim 2, characterized in that, The relationship between the function parameters and the posterior probability is represented by the following formula: Where p(k|X) n ) represents the posterior probability that the nth historical fusion feature vector belongs to the kth Gaussian distribution function; Let k be the weight of the k-th Gaussian distribution function in the t-th iteration update; Let be the mean of the k-th Gaussian distribution function in the t-th iteration update; Let X be the covariance matrix of the k-th Gaussian distribution function in the t-th iteration update; n There are n historical fusion feature vectors; N is the total number of historical fusion feature vectors; K is the number of Gaussian distribution functions; The relationship between the function parameters and the likelihood function values ​​is represented by the following formula; Where L(t) is the likelihood function value in the t-th iteration update; X n There are n historical fusion feature vectors; Let k be the weight of the k-th Gaussian distribution function in the t-th iteration update; Let be the mean of the k-th Gaussian distribution function in the t-th iteration update; Let be the covariance matrix of the k-th Gaussian distribution function in the t-th iteration update; N is the total number of historical fusion feature vectors.

4. The wind turbine blade fault detection method according to claim 2, characterized in that, The historical status data represents the multi-dimensional feature data generated by the wind turbine blades during operation. The step of performing feature fusion processing on the historical state data to obtain a historical fused feature vector specifically includes: The feature data of each dimension in the historical state data are processed to obtain state processing data of multiple dimensions. The state processing data of the multiple dimensions are subjected to feature extraction to obtain signal feature vectors of multiple dimensions. Based on preset feature weights, the feature vectors of the multiple-dimensional signals are fused to obtain a historical fused feature vector.

5. The wind turbine blade fault detection method according to claim 4, characterized in that, The feature data for each dimension includes sound feature data, vibration feature data, and temperature feature data; the state processing data for multiple dimensions includes sound feature processing data, vibration feature processing data, and temperature feature processing data. The process involves processing the feature data of each dimension in the historical state data to obtain multi-dimensional state processing data, including: The sound characteristic data, vibration characteristic data, and temperature characteristic data of the wind turbine blades at various measurement locations under multiple operating conditions are determined; the multiple operating conditions include static operating condition, no-load operating condition, and loaded operating condition; Based on multiple sound feature data and multiple vibration feature data, a static reference envelope range is established under various working conditions. Based on the static reference envelope range, data exceeding the static reference envelope range in the sound feature data and the vibration feature data are eliminated, as well as static measurement errors introduced by changes in different working conditions are eliminated, to obtain the sound feature processing data and the vibration feature processing data. The outliers in the multiple temperature feature data are eliminated, and the processed temperature feature data is normalized to obtain the processed temperature feature data.

6. The wind turbine blade fault detection method according to claim 5, characterized in that, The multi-dimensional signal feature vectors include sound signal feature vectors, vibration signal feature vectors, and temperature signal feature vectors; The step of extracting features from the multi-dimensional state processing data to obtain a multi-dimensional signal feature vector includes: An acoustic fingerprint feature is extracted from the sound feature processing data using a preset extraction method; the acoustic fingerprint feature is then integrated and processed to obtain the sound signal feature vector; the preset extraction method includes one or more of the following: wavelet transform, Fourier transform, and time-frequency analysis; Extract the time-domain and frequency-domain statistical features of the vibration feature processing data; integrate the time-domain and frequency-domain statistical features to obtain the vibration signal feature vector; the time-domain statistical features include peak value, root mean square, variance, and kurtosis; the frequency-domain statistical features include frequency centroid and frequency variance; Based on the difference between adjacent temperature data points, the temperature change trend features of the temperature feature processing data are extracted to obtain the temperature signal feature vector.

7. The wind turbine blade fault detection method according to claim 2, characterized in that, The step of inputting the current state data of the target blade into the blade fingerprint recognition model to obtain the current evaluation result includes: The current state data of the target blade is subjected to feature fusion processing to obtain the current fused feature vector; The current fused feature vector is input into the blade fingerprint recognition model to obtain the current evaluation result; the current evaluation result characterizes the degree of deviation between the current target blade's operating state and its normal operating state.

8. The method for detecting wind turbine blade faults according to any one of claims 1 to 7, characterized in that, The method shown also includes: Observe the running status of the current status data to obtain the current status label of the current status data; If the current state label is determined to represent an anomaly in the target blade, the current state data is stored in the anomaly voiceprint database. The current status label is determined to indicate that the target blade is normal; the current status data is stored in the normal voiceprint library.

9. The method for detecting wind turbine blade faults according to claim 8, characterized in that, The method further includes: The leaf fingerprint recognition model is trained based on the normal state data of the normal voiceprint library; The blade fingerprint recognition model is verified based on the abnormal state data of the abnormal voiceprint library.

10. A wind turbine blade fault detection device, characterized in that, The device includes: The data acquisition unit is configured to acquire historical status data of the normal operation of wind turbine blades; The model building unit is configured to build a blade fingerprint recognition model based on the historical state data; the blade fingerprint recognition model is used to positively evaluate the state data based on the normal operation trend represented by the historical state data. The fault identification unit is configured to input the current status data of the target blade into the blade fingerprint recognition model to obtain the current evaluation result; if the current evaluation result is determined to be less than a preset threshold, the target blade is determined to be abnormal; if the current evaluation result is determined to be greater than or equal to the preset threshold, the target blade is determined to be normal.

Images