Fan blade damage identification method and device based on small sample

By converting the audio time domain signal into frequency domain signal and extracting dimensionless octave features, and combining with the MAML algorithm to train the neural network model, the problem of low training accuracy of the blade damage recognition model of wind turbine is solved, and high Accurately and quickly judge the health status of the fan blades, reducing the cost of data acquisition.

CN118964926APending Publication Date: 2024-11-15HUAZHONG UNIV OF SCI & TECH RES INST SHENZHEN
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410944331.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-15
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

In the prior art, the training accuracy of the blade damage recognition model of the wind turbine generator is not high, mainly due to the difficulty in collecting defect signals, resulting in low recognition accuracy.

Method used

The fan blade damage recognition method based on small samples is adopted. By converting the audio time domain signal into frequency domain signal, the dimensionless octave feature is extracted, and the initial neural network model is updated gradiently to improve the model's generalization ability and adaptability.

Benefits of technology

Quickly detect fan blade defects under a small sample size, reduce data acquisition costs, improve the accuracy of fan blade damage recognition and model training efficiency, and enhance the practicality and stability of the model.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118964926A_ABST
    Figure CN118964926A_ABST
Patent Text Reader

Abstract

The invention belongs to the field of neural networks, and particularly discloses a fan blade damage identification method and device based on a small sample, and the method comprises the steps: collecting an audio time domain signal of a fan blade, and converting the audio time domain signal into a frequency domain signal; performing feature extraction on the frequency domain signal to obtain a dimensionless octave feature, and taking the octave feature as a data set for model training; and inputting the data set into an initial neural network model, carrying out gradient updating on the initial neural network model by using an MAML algorithm, and continuously adjusting parameters of the initial neural network model until the model converges to obtain a trained fan blade damage identification model, and performing fan blade damage identification by using the trained fan blade damage identification model to obtain an identification result of the fan blade. According to the invention, the model training efficiency and training precision can be improved, and the recognition accuracy of the fan blade is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application belongs to the field of neural networks, and more specifically, to a method and device for identifying wind turbine blade damage based on a small sample. Background Art

[0002] Due to the global energy crisis and the growing demand for clean energy, wind power has become an important source of electricity for production and life. Wind energy is one of the fastest growing renewable energy sources in the world, with advantages such as clean, renewable, and mature technology. However, compared with traditional energy, wind energy still has problems such as poor reliability and high operation and maintenance costs, resulting in poor availability and economy.

[0003] Wind turbine blades (WTB) are important components of wind turbines (WT), accounting for about 20% of the total cost and one of the components with the highest failure rate. Since commercial wind turbines are usually exposed to harsh working environments and changeable weather conditions, various defects will greatly shorten the operation time of wind turbines.

[0004] Structural Health Monitoring (SHM) technology is a major concern for the wind power industry and academia. Reliable and cost-effective monitoring can reduce long downtime and high maintenance costs, and avoid catastrophic consequences due to undetected failures. In recent years, the size of wind turbines has gradually increased, thereby improving efficiency and power generation, but the probability of failure has also increased.

[0005] At present, many studies are mainly focused on how to accurately identify acoustic signals of various types of damage and how to design acoustic feature extraction algorithms and deep learning models to improve the accuracy of recognition. Deep learning requires the collection of a large amount of data to achieve the diversity of types and degrees of damage to wind turbine blades in order to improve the generalization ability of the model. However, it is very difficult to collect a large number of wind turbine blade defect signals, resulting in poor model training accuracy and thus low recognition accuracy of wind turbine blades. Summary of the invention

[0006] In view of the defects of the prior art, the purpose of this application is to provide a method and device for wind turbine blade damage identification based on small samples, aiming to solve the problem that it is difficult to collect a large number of wind turbine blade defect signals, resulting in poor model training accuracy and thus low wind turbine blade identification accuracy.

[0007] In a first aspect, the present application provides a method for identifying wind turbine blade damage based on a small sample, comprising:

[0008] Collecting the audio time domain signal of the fan blades, converting the audio time domain signal into a frequency domain signal; performing feature extraction on the frequency domain signal to obtain dimensionless octave features, and using the octave features as a data set for model training;

[0009] Converting the audio time domain signal into a frequency domain signal and performing feature extraction to extract dimensionless octave features can more effectively capture key information in the audio signal, better reflect the periodicity and frequency components of the sound, and thus help identify damage to the fan blades. The dimensionless processing of the octave features can make the features consistent in different situations, reduce the impact caused by changes in measurement scale or unit, and improve the generalization ability of the model. The data set is input into the initial neural network model, and the MAML algorithm is used to perform gradient update on the initial neural network model and continuously adjust the parameters of the initial neural network model until the model converges to obtain a trained fan blade damage recognition model, and the trained fan blade damage recognition model is used to perform fan blade damage recognition to obtain fan blade recognition results.

[0010] Using the MAML algorithm to train the neural network model can quickly adapt to new tasks. MAML enables the neural network to quickly adjust parameters in a small amount of data or different tasks through learning. The MAML algorithm uses task training and secondary gradients. Slight changes in the model parameters will update and improve the loss function of all tasks from the training and test data. Gradient updates through the MAML algorithm can find high-quality model parameters more quickly and make the model converge effectively. This can not only improve training efficiency and training accuracy, but also improve the final performance of the model, making it more practical and stable in practical applications.

[0011] In some embodiments, it also includes:

[0012] Dividing the data set into a training set, a test set, and a validation set;

[0013] The data set is input into the initial neural network model, and the initial neural network model is gradient updated by using the MAML algorithm and the parameters of the initial neural network model are continuously adjusted until the model converges to obtain a trained wind turbine blade damage recognition model, including:

[0014] Combining each sample in the training set and the validation set according to different categories to obtain a training set task and a validation set task;

[0015] The initial neural network model is trained based on the training set task to obtain a first-order gradient, and the initial model parameters are adjusted using the first-order gradient to obtain an intermediate model;

[0016] The intermediate model is trained based on the verification set task to obtain a second-order gradient, and the second-order gradient is used as a model parameter for updating the intermediate model to obtain a trained wind turbine blade damage recognition model.

[0017] In some embodiments, extracting features from the frequency domain signal to obtain dimensionless octave features includes:

[0018] Dividing the frequency spectrum corresponding to the frequency domain signal into a plurality of different frequency bands by using an equal-proportional octave algorithm, and obtaining the sound pressure of each frequency band;

[0019] Determine the sound pressure level corresponding to the frequency band based on the sound pressure of each frequency band;

[0020] Each sound pressure level is subjected to maximum normalization and minimum normalization, and the sound pressure level is converted into a dimensionless octave band characteristic.

[0021] In some embodiments, after extracting features from the frequency domain signal to obtain dimensionless octave features, the method further includes:

[0022] Performing dimension reduction processing and principal component analysis on the octave features by using a PCA algorithm, and distinguishing the octave features into normal audio, cracked audio, and unbalanced audio;

[0023] The method of using the octave features as a data set for model training includes:

[0024] The normal audio, crackle audio, and unbalanced audio corresponding to the octave features processed by PCA are used as data sets for model training.

[0025] In some embodiments, the training of the initial neural network model based on the training set task to obtain a first-order gradient includes:

[0026] Inputting the training set task into the initial neural network model to obtain a training prediction value, obtaining a training loss value based on the training prediction value and the identification label, and obtaining a first-order gradient of a model parameter according to the training loss value;

[0027] The step of training the intermediate model based on the validation set task to obtain a second-order gradient includes:

[0028] The validation set task is input into the intermediate model to obtain a validation prediction value, a validation loss value is obtained based on the validation prediction value and the identification label, and the second-order gradient of the model parameter is obtained according to the validation loss value.

[0029] In some embodiments, the fan blade damage identification is performed using a trained fan blade damage identification model to obtain a fan blade damage identification result, including:

[0030] Using the test set as the data to be identified for the wind turbine blades;

[0031] Inputting the data to be identified into a trained fan blade damage identification model to obtain a fan blade damage identification result corresponding to the data to be identified;

[0032] The trained wind turbine blade damage recognition model is obtained by training using the MAML algorithm based on the data set and the recognition labels corresponding to the data set.

[0033] In a second aspect, the present application also provides a wind turbine blade damage identification device based on a small sample, comprising:

[0034] A signal conversion unit, used for collecting the audio time domain signal of the fan blade, and converting the audio time domain signal into a frequency domain signal;

[0035] A feature extraction unit, used to extract features from the frequency domain signal to obtain dimensionless octave features, and use the octave features as a data set for model training;

[0036] The model training unit is used to input the data set into the initial neural network model, use the MAML algorithm to perform gradient update on the initial neural network model and continuously adjust the parameters of the initial neural network model until the model converges to obtain a trained wind turbine blade damage recognition model.

[0037] In a third aspect, the present application also provides a wind turbine blade damage identification system, comprising: a signal acquisition module, a communication module, a central server, and a wind turbine blade damage identification device based on a small sample;

[0038] The mobile device is connected to the central server via a communication module to check the health status of the wind turbine blades.

[0039] In a fourth aspect, the present application provides an electronic device, comprising: at least one memory for storing programs; and at least one processor for executing the programs stored in the memory. When the program stored in the memory is executed, the processor is used to execute the method described in the first aspect or any possible implementation of the first aspect.

[0040] In a fifth aspect, the present application provides a computer-readable storage medium, which stores a computer program. When the computer program runs on a processor, the processor executes the method described in the first aspect or any possible implementation of the first aspect.

[0041] In a sixth aspect, the present application provides a computer program product. When the computer program product runs on a processor, the processor executes the method described in the first aspect or any possible implementation of the first aspect.

[0042] It can be understood that the beneficial effects of the second to fifth aspects mentioned above can be found in the relevant description of the first aspect mentioned above, and will not be repeated here.

[0043] In general, the above technical solutions conceived by this application have the following beneficial effects compared with the prior art:

[0044] (1) This application uses the MAML algorithm to perform model training based on small samples, so that wind turbine blade defects can be detected from a smaller sample size, reducing the labor cost of data collection, and can quickly and accurately determine the health status of wind turbine blades, solving the problem of difficulty in collecting a large number of wind turbine blade defect signals, improving the accuracy of model training, and thereby improving the recognition accuracy of wind turbine blades.

[0045] (2) Converting the audio time domain signal into a frequency domain signal and extracting features to extract dimensionless octave features can more effectively capture key information in the audio signal and better reflect the periodicity and frequency components of the sound, thereby helping to identify damage to the wind turbine blades. Dimensionless processing of octave features can make the features consistent in different situations, reduce the impact of changes in measurement scale or units, and improve the generalization ability of the model.

[0046] (3) The neural network model is trained using the MAML algorithm, which can quickly adapt to new tasks. MAML enables the neural network to quickly adjust parameters in a small amount of data or different tasks through learning. The MAML algorithm uses task training and secondary gradients. Slight changes in the model parameters will update and improve the loss function of all tasks from the training and test data. Gradient updates using the MAML algorithm can find high-quality model parameters more quickly and make the model converge effectively. This can not only improve training efficiency and training accuracy, but also improve the final performance of the model, making it more practical and stable in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is one of the flow charts of the wind turbine blade damage identification method based on a small sample provided in the embodiment of the present application;

[0048] Figure 2 It is a MAML training flow chart provided in an embodiment of the present application;

[0049] Figure 3 It is a flowchart of the algorithm design provided by the embodiment of the present application;

[0050] Figure 4 These are time domain and frequency domain diagrams of audio signals of three defect types in an embodiment of the present application;

[0051] Figure 5 It is an octave band characteristic diagram of three defect types in the embodiment of the present application;

[0052] Figure 6 It is a schematic diagram of the confusion matrix of the traditional algorithm and the small sample algorithm under different data amounts in the embodiment of the present application;

[0053] Figure 7 It is a structural schematic diagram of a wind turbine blade damage identification device based on a small sample provided in an embodiment of the present application;

[0054] Figure 8 It is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0056] The term "and / or" in this article is a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. The symbol " / " in this article indicates that the associated objects are in an or relationship, for example, A / B means A or B.

[0057] The terms "first" and "second" in the specification and claims herein are used to distinguish different objects rather than to describe a specific order of the objects. For example, a first response message and a second response message are used to distinguish different response messages rather than to describe a specific order of the response messages.

[0058] In the embodiments of the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

[0059] In the description of the embodiments of the present application, unless otherwise specified, "multiple" means two or more than two. For example, multiple processing units refer to two or more processing units, etc.; multiple elements refer to two or more elements, etc.

[0060] Next, the technical solutions provided in the embodiments of the present application are introduced.

[0061] Reference Figure 1 The present application provides a method for identifying wind turbine blade damage based on a small sample, comprising:

[0062] S101. Collecting the audio time domain signal of the fan blade, converting the audio time domain signal into a frequency domain signal;

[0063] S102. Extracting features from the frequency domain signal to obtain dimensionless octave features, and using the octave features as a data set for model training;

[0064] S103. Input the data set into the initial neural network model, use the MAML algorithm to perform gradient update on the initial neural network model and continuously adjust the parameters of the initial neural network model until the model converges to obtain a trained fan blade damage recognition model, and use the trained fan blade damage recognition model to perform fan blade damage recognition to obtain a fan blade recognition result.

[0065] First, through the above S101, an audio time domain signal is obtained from the fan blade. Real-time or periodic collection is performed through a microphone or sensor, and the microphone device is set near the fan blade. The audio time domain signal represents the amplitude of the sound changing over time. For subsequent processing, it is usually necessary to convert the time domain signal into a frequency domain signal. The present application uses fast Fourier transform (FFT) to convert signals between time domain and frequency domain, and the converted frequency domain signal can display the components of the sound at different frequencies.

[0066] Then, features are extracted from the frequency domain signal. In particular, dimensionless octave features are extracted. Octave features are usually used to capture periodicity and frequency information in signals, and dimensionless processing can make the features have a unified scale and comparability.

[0067] Finally, the model is trained and optimized. The extracted octave features are used as a data set and input into the initial neural network model. The initial neural network model is optimized using the meta-learning algorithm (MAML, Model-Agnostic Meta-Learning). MAML aims to improve the generalization and adaptability of the model by quickly adjusting the model parameters to adapt it to new tasks or data sets. During the training process, MAML updates the initial parameters of the model by calculating the gradients on multiple tasks so that it can quickly learn and adapt to future tasks. Through repeated iterative optimization, until the model converges, that is, the model performs stably on the training set and can effectively identify the damage of the wind turbine blades. The trained wind turbine blade damage recognition model is used to identify the wind turbine blade damage on the collected data to be identified (usually the test set), and the recognition result of the wind turbine blade is obtained.

[0068] Optionally, the trained model can be applied to a fan monitoring system in an actual environment to automatically identify the damage of the fan blades. This application realizes intelligent identification of fan blade damage by combining audio signal processing technology and meta-learning algorithm, thereby improving the accuracy and efficiency of diagnosis.

[0069] The wind turbine blade damage identification method based on small samples provided in this application uses the MAML algorithm to train the neural network model, which can quickly adapt to new tasks. MAML learns how to learn, so that the neural network can quickly adjust parameters in a small amount of data or different tasks. The MAML algorithm uses task training and secondary gradients, which is very sensitive to changes in tasks. Therefore, a slight change in the model parameters will produce a substantial improvement in the loss function of all tasks from training and test data. By updating the gradient through the MAML algorithm, high-quality model parameters can be found more quickly, so that the model can converge effectively. Not only can the training efficiency be improved, but also the final performance of the model can be improved, making it more practical and stable in practical applications. Using the MAML algorithm to train the model based on small samples, the defects of wind turbine blades can be detected from a smaller sample size, reducing the labor cost of data collection, and can quickly and accurately determine the health status of wind turbine blades, solving the problem of difficulty in collecting a large number of wind turbine blade defect signals, improving the accuracy of model training, and thus improving the recognition accuracy of wind turbine blades.

[0070] In some embodiments, it also includes:

[0071] Dividing the data set into a training set, a test set, and a validation set;

[0072] The data set is input into the initial neural network model, and the initial neural network model is gradient updated by using the MAML algorithm and the parameters of the initial neural network model are continuously adjusted until the model converges to obtain a trained wind turbine blade damage recognition model, including:

[0073] Combining each sample in the training set and the validation set according to different categories to obtain a training set task and a validation set task;

[0074] The initial neural network model is trained based on the training set task to obtain a first-order gradient, and the initial model parameters are adjusted using the first-order gradient to obtain an intermediate model;

[0075] The intermediate model is trained based on the verification set task to obtain a second-order gradient, and the second-order gradient is used as a model parameter for updating the intermediate model to obtain a trained wind turbine blade damage recognition model.

[0076] Furthermore, the first-order gradient and the second-order gradient are obtained as follows:

[0077] Inputting the training set task into the initial neural network model to obtain a training prediction value, obtaining a training loss value based on the training prediction value and the identification label, and obtaining a first-order gradient of a model parameter according to the training loss value;

[0078] The validation set task is input into the intermediate model to obtain a validation prediction value, a validation loss value is obtained based on the validation prediction value and the identification label, and the second-order gradient of the model parameter is obtained according to the validation loss value.

[0079] Reference Figure 2 Specifically, this embodiment first combines samples from the overall data set according to different categories to form multiple training set tasks and validation set tasks. The specific model training process is as follows:

[0080] First-order gradient adjustment:

[0081] For each training set task, the corresponding sample data is used to train the initial neural network model. In this process, the first-order gradient of the loss function with respect to the model parameters is calculated by back propagation.

[0082] Intermediate model generation:

[0083] Apply first-order gradients to the initial model parameters to adjust the model to obtain an intermediate model, which can be achieved through gradient descent or other optimization algorithms.

[0084] Second-order gradient training:

[0085] The validation set task is used to further train the intermediate model, and the second-order gradient of the loss function with respect to the intermediate model parameters is calculated through back propagation. The second-order gradient takes into account the impact of changes in model parameter updates on the loss function, and is therefore more sophisticated and accurate.

[0086] Final model generation:

[0087] The second-order gradient is applied to the parameters of the intermediate model, and then the model is updated to obtain the final trained wind blade damage identification model.

[0088] This embodiment uses task decomposition and meta-learning techniques to enable the model to continuously optimize its own structure and parameters during the learning process to adapt to various wind blade damage detection tasks. Meta-learning can significantly improve the performance of the model on new tasks, improve the training efficiency and training accuracy of the wind blade damage model, and thus improve the effect of wind blade damage identification.

[0089] In some embodiments, extracting features from the frequency domain signal to obtain dimensionless octave features includes:

[0090] Dividing the frequency spectrum corresponding to the frequency domain signal into a plurality of different frequency bands by using an equal-proportional octave algorithm, and obtaining the sound pressure of each frequency band;

[0091] Determine the sound pressure level corresponding to the frequency band based on the sound pressure of each frequency band;

[0092] Each sound pressure level is subjected to maximum normalization and minimum normalization, and the sound pressure level is converted into a dimensionless octave band characteristic.

[0093] Reference Figure 3 ,After converting the extracted time domain signal into frequency domain signal, the entire spectrum is divided into octaves.

[0094] It should be noted that the octave is a method of division according to frequency ratio. Usually the center frequency of each octave band is twice the center frequency of the previous band, for example: 60Hz, 120Hz, 250Hz, 500Hz, etc.

[0095] For each octave band, the sound pressure level of the frequency components in the band is calculated. Sound pressure level (SPL) is a common unit of measurement for sound intensity, used to describe the relative intensity of sound. Furthermore, the sound pressure level calculated in each octave band is subjected to maximum and minimum normalization to ensure that the sound pressure level values ​​of all bands are mapped to a dimensionless range, usually between 0 and 1. Finally, the normalized sound pressure level is used as the octave feature.

[0096] The octave feature in this embodiment can reflect the sound energy distribution of the frequency domain signal in different frequency bands, can provide energy information of the frequency domain signal in different frequency bands, and can also effectively reduce the amount of data and improve analysis efficiency.

[0097] In some embodiments, after extracting features from the frequency domain signal to obtain dimensionless octave features, the method further includes:

[0098] Performing dimension reduction processing and principal component analysis on the octave features by using a PCA algorithm, and distinguishing the octave features into normal audio, cracked audio, and unbalanced audio;

[0099] The method of using the octave features as a data set for model training includes:

[0100] The normal audio, crackle audio, and unbalanced audio corresponding to the octave features processed by PCA are used as data sets for model training.

[0101] Reference Figure 4 and Figure 5 , Figure 4 These are the time domain and frequency domain diagrams of audio signals for three defect types (normal, crack, and unbalanced). Figure 5 These are the octave band characteristic diagrams of three defect types.

[0102] Specifically, the normalized sound pressure level feature vectors form a feature matrix. Each row of this matrix represents a sample (such as an audio file), and each column represents the sound pressure level of an octave band. The PCA algorithm is applied to the feature matrix, and PCA maps the original feature space to a new low-dimensional feature space, namely the principal component. The principal components generated by PCA are then selected, and the number of principal components to be retained is usually determined based on the proportion of explained variance. The principal components processed by PCA are used as new feature vectors to construct the training data set for the model.

[0103] The principal component feature vector after PCA processing is used as input to classify normal audio, crackle audio and unbalanced audio. In this embodiment, PCA is used to process octave features and establish a classification model, which can effectively analyze and classify the features of different audio signals.

[0104] In some embodiments, the fan blade damage identification is performed using a trained fan blade damage identification model to obtain a fan blade damage identification result, including:

[0105] Using the test set as the data to be identified for the wind turbine blades;

[0106] Inputting the data to be identified into a trained fan blade damage identification model to obtain a fan blade damage identification result corresponding to the data to be identified;

[0107] The trained wind turbine blade damage recognition model is obtained by training using the MAML algorithm based on the data set and the recognition labels corresponding to the data set.

[0108] Specifically, this embodiment is a test process of a data set, that is, an application process of a model. The trained fan blade damage recognition model is used to detect and recognize fan blades, and the fan blade damage recognition results of the poems to be recognized in the test set are obtained.

[0109] Reference Figure 6 , Figure 6 It is the confusion matrix of the traditional algorithm and the small sample algorithm under different data amounts.

[0110] Reference Figure 7 The present application also provides a wind turbine blade damage identification device based on a small sample, comprising:

[0111] The signal conversion unit 710 is used to collect the audio time domain signal of the fan blade and convert the audio time domain signal into a frequency domain signal;

[0112] A feature extraction unit 720 is used to extract features from the frequency domain signal to obtain dimensionless octave features, and use the octave features as a data set for model training;

[0113] The model training unit 730 is used to input the data set into the initial neural network model, use the MAML algorithm to gradient update the initial neural network model and continuously adjust the parameters of the initial neural network model until the model converges to obtain a trained wind turbine blade damage recognition model.

[0114] In some embodiments, the data set is input into an initial neural network model, and the initial neural network model is gradient updated using a MAML algorithm and the parameters of the initial neural network model are continuously adjusted until the model converges to obtain a trained wind turbine blade damage recognition model, including:

[0115] Combining each sample in the training set and the validation set according to different categories to obtain a training set task and a validation set task;

[0116] The initial neural network model is trained based on the training set task to obtain a first-order gradient, and the initial model parameters are adjusted using the first-order gradient to obtain an intermediate model;

[0117] The intermediate model is trained based on the verification set task to obtain a second-order gradient, and the second-order gradient is used as a model parameter for updating the intermediate model to obtain a trained wind turbine blade damage recognition model.

[0118] In some embodiments, extracting features from the frequency domain signal to obtain dimensionless octave features includes:

[0119] Dividing the frequency spectrum corresponding to the frequency domain signal into a plurality of different frequency bands by using an equal-proportional octave algorithm, and obtaining the sound pressure of each frequency band;

[0120] Determine the sound pressure level corresponding to the frequency band based on the sound pressure of each frequency band;

[0121] Each sound pressure level is subjected to maximum normalization and minimum normalization, and the sound pressure level is converted into a dimensionless octave band characteristic.

[0122] In some embodiments, after extracting features from the frequency domain signal to obtain dimensionless octave features, the method further includes:

[0123] Performing dimension reduction processing and principal component analysis on the octave features by using a PCA algorithm, and distinguishing the octave features into normal audio, cracked audio, and unbalanced audio;

[0124] The method of using the octave features as a data set for model training includes:

[0125] The normal audio, crackle audio, and unbalanced audio corresponding to the octave features processed by PCA are used as data sets for model training.

[0126] In some embodiments, the training of the initial neural network model based on the training set task to obtain a first-order gradient includes:

[0127] Inputting the training set task into the initial neural network model to obtain a training prediction value, obtaining a training loss value based on the training prediction value and the identification label, and obtaining a first-order gradient of a model parameter according to the training loss value;

[0128] The step of training the intermediate model based on the validation set task to obtain a second-order gradient includes:

[0129] The validation set task is input into the intermediate model to obtain a validation prediction value, a validation loss value is obtained based on the validation prediction value and the identification label, and the second-order gradient of the model parameter is obtained according to the validation loss value.

[0130] In some embodiments, the invention further includes an identification unit, which is specifically used to:

[0131] Using the test set as the data to be identified for the wind turbine blades;

[0132] Inputting the data to be identified into a trained fan blade damage identification model to obtain a fan blade damage identification result corresponding to the data to be identified;

[0133] The trained wind turbine blade damage recognition model is obtained by training using the MAML algorithm based on the data set and the recognition labels corresponding to the data set.

[0134] The present application also provides a wind turbine blade damage identification system, comprising: a signal acquisition module, a communication module, a central server, and a wind turbine blade damage identification device based on a small sample;

[0135] The mobile device is connected to the central server via a communication module to check the health status of the wind turbine blades.

[0136] It can be understood that the detailed functional implementation of each of the above-mentioned units / modules can be found in the introduction of the aforementioned method embodiment, and will not be repeated here.

[0137] It should be understood that the above-mentioned device is used to execute the method in the above-mentioned embodiment. The implementation principle and technical effect of the corresponding program module in the device are similar to those described in the above-mentioned method. The working process of the device can refer to the corresponding process in the above-mentioned method, which will not be repeated here.

[0138] Reference Figure 8 Based on the method in the above embodiment, the embodiment of the present application provides an electronic device, which may include: a processor (processor) 810, a communication interface (Communications Interface) 820, a memory (memory) 830 and a communication bus 840, wherein the processor 810, the communication interface 820, and the memory 830 communicate with each other through the communication bus 840. The processor 810 can call the logic instructions in the memory 830 to execute the method in the above embodiment.

[0139] In addition, the logic instructions in the above-mentioned memory 830 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application.

[0140] Based on the method in the above embodiment, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program runs on a processor, the processor executes the method in the above embodiment.

[0141] Based on the method in the above embodiment, an embodiment of the present application provides a computer program product. When the computer program product runs on a processor, the processor executes the method in the above embodiment.

[0142] It is understandable that the processor in the embodiment of the present application can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. The general-purpose processor can be a microprocessor or any conventional processor.

[0143] The method steps in the embodiments of the present application can be implemented by hardware or by a processor executing software instructions. The software instructions can be composed of corresponding software modules, and the software modules can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, mobile hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to a processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an ASIC.

[0144] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer instructions may be transmitted from a website site, a computer, a server or a data center to another website site, a computer, a server or a data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or a data center that includes one or more available media integrated. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid state disk (SSD)), etc.

[0145] It should be understood that the various numerical numbers involved in the embodiments of the present application are only used for the convenience of description and are not used to limit the scope of the embodiments of the present application.

[0146] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A method for identifying wind turbine blade damage based on small samples, characterized in that: include: Collecting the audio time domain signal of the fan blade, and converting the audio time domain signal into a frequency domain signal; Extracting features from the frequency domain signal to obtain dimensionless octave features, and using the octave features as a data set for model training; The data set is input into the initial neural network model, and the MAML algorithm is used to perform gradient update on the initial neural network model and continuously adjust the parameters of the initial neural network model until the model converges to obtain a trained fan blade damage recognition model, and the trained fan blade damage recognition model is used to perform fan blade damage recognition to obtain a fan blade recognition result.

2. The method for identifying wind turbine blade damage based on a small sample according to claim 1 is characterized in that: Also includes: Dividing the data set into a training set, a test set, and a validation set; The data set is input into the initial neural network model, and the initial neural network model is gradient updated by using the MAML algorithm and the parameters of the initial neural network model are continuously adjusted until the model converges to obtain a trained wind turbine blade damage recognition model, including: Combining each sample in the training set and the validation set according to different categories to obtain a training set task and a validation set task; The initial neural network model is trained based on the training set task to obtain a first-order gradient, and the initial model parameters are adjusted using the first-order gradient to obtain an intermediate model; The intermediate model is trained based on the verification set task to obtain a second-order gradient, and the second-order gradient is used as a model parameter for updating the intermediate model to obtain a trained wind turbine blade damage recognition model.

3. The method for identifying wind turbine blade damage based on a small sample according to claim 1, characterized in that: The extracting features of the frequency domain signal to obtain dimensionless octave features includes: Dividing the frequency spectrum corresponding to the frequency domain signal into a plurality of different frequency bands by using an equal-proportional octave algorithm, and obtaining the sound pressure of each frequency band; Determine the sound pressure level corresponding to the frequency band based on the sound pressure of each frequency band; Each sound pressure level is subjected to maximum normalization and minimum normalization, and the sound pressure level is converted into a dimensionless octave band characteristic.

4. According to the small sample-based wind turbine blade damage identification method of claim 3, after extracting features from the frequency domain signal to obtain dimensionless octave features, the method further comprises: Performing dimension reduction processing and principal component analysis on the octave features by using a PCA algorithm, and distinguishing the octave features into normal audio, cracked audio, and unbalanced audio; The method of using the octave features as a data set for model training includes: The normal audio, crackle audio, and unbalanced audio corresponding to the octave features processed by PCA are used as data sets for model training.

5. According to the small sample-based wind turbine blade damage identification method of claim 2, the initial neural network model is trained based on the training set task to obtain a first-order gradient, comprising: Inputting the training set task into the initial neural network model to obtain a training prediction value, obtaining a training loss value based on the training prediction value and the identification label, and obtaining a first-order gradient of a model parameter according to the training loss value; The step of training the intermediate model based on the validation set task to obtain a second-order gradient includes: The validation set task is input into the intermediate model to obtain a validation prediction value, a validation loss value is obtained based on the validation prediction value and the identification label, and the second-order gradient of the model parameter is obtained according to the validation loss value.

6. The method for identifying wind turbine blade damage based on small samples according to claim 2 is characterized in that: The fan blade damage identification is performed using the trained fan blade damage identification model to obtain the fan blade damage identification result, including: Using the test set as the data to be identified for the wind turbine blades; Inputting the data to be identified into a trained fan blade damage identification model to obtain a fan blade damage identification result corresponding to the data to be identified; The trained wind turbine blade damage recognition model is obtained by training using the MAML algorithm based on the data set and the recognition labels corresponding to the data set.

7. A wind turbine blade damage identification device based on a small sample, characterized in that: include: A signal conversion unit, used for collecting the audio time domain signal of the fan blade, and converting the audio time domain signal into a frequency domain signal; A feature extraction unit, used to extract features from the frequency domain signal to obtain dimensionless octave features, and use the octave features as a data set for model training; The model training unit is used to input the data set into the initial neural network model, use the MAML algorithm to perform gradient update on the initial neural network model and continuously adjust the parameters of the initial neural network model until the model converges to obtain a trained wind turbine blade damage recognition model.

8. A fan blade damage identification system, characterized in that: include: A signal acquisition module, a communication module, a central server and a wind turbine blade damage identification device based on a small sample as described in claim 7; The mobile device is connected to the central server via a communication module to check the health status of the wind turbine blades.

9. An electronic device, characterized in that: include: at least one memory for storing a computer program; At least one processor is used to execute the program stored in the memory. When the program stored in the memory is executed, the processor is used to execute the method according to any one of claims 1 to 6.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program runs on a processor, the processor is caused to execute the method according to any one of claims 1 to 6.