Wind turbine generator fault diagnosis method

Abstract

The invention relates to a wind turbine generator fault diagnosis method. Comprising the steps of collecting an audio signal of a wind turbine generator in a working operation state; preprocessing the audio signal through a wavelet analysis signal processing module to obtain a noise reduction signal; wherein the wavelet analysis signal processing module adopts different expression functions to process a wavelet coefficient according to a comparison result of the wavelet coefficient and a threshold value; inputting the noise reduction signal into the trained feature extraction model for feature extraction; and performing classification mapping on the signal after feature extraction through a BP neural network, and determining fault diagnosis of the wind turbine generator. According to the method, audio signals are preprocessed through variable threshold wavelet analysis, and the noise reduction process is adjusted, so that part of useful signals in wavelet coefficients are reserved. The characteristics in the audio signals are extracted, so that the change characteristics of the operation state of the wind turbine generator can be accurately captured, and the operation state of the wind turbine generator can be monitored in real time.

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical field of wind turbines, and in particular to a method for diagnosing faults of wind turbines. Background Art

[0002] As a renewable and clean form of energy, wind power generation is of great significance to reducing greenhouse gas emissions and reducing dependence on fossil fuels. Therefore, the development and application of wind turbines have promoted the increase in the proportion of renewable energy in the energy structure, and have played a positive role in addressing climate change and environmental issues. The continuous advancement and application of wind power generation technology have promoted the development of the wind energy industry and provided new growth points for related industrial chains, including the manufacturing, installation, operation and maintenance of wind turbines, as well as the planning, design, and construction of wind farms. The development of the wind energy industry has not only driven employment growth in related industries, but also promoted economic development and transformation and upgrading.

[0003] Wind turbines are a common type of rotating machinery that is essential for ventilation in factories, boilers, and buildings. In order to ensure the normal operation of these devices, fault diagnosis of wind turbines becomes particularly important. In actual operation, equipment managers usually use the "listen, touch, see, and compare" method to regularly inspect equipment and collect various data, including vibration parameters, noise parameters, oil temperature, etc. However, traditional inspection methods have certain limitations, especially in the face of equipment complexity and non-intuitive fault characteristics. Therefore, researchers and engineers have been committed to developing advanced wind turbine fault diagnosis methods. These systems are designed to improve the reliability and safety of wind turbines through real-time monitoring, alarming, and diagnosis.

[0004] In the related art, the environmental noise component contained in the original operating audio of each component is calculated, the actual operating audio of each component is calculated respectively, and the fault diagnosis of the component is performed based on the actual operating audio, and the actual operating audio is equal to the difference between the original operating audio and the environmental noise component contained.

[0005] Regarding the above technical solution, the inventors found that there are at least the following technical problems:

[0006] Since the characteristics of different components may cause variations in raw operating audio and environmental noise, if the device does not take these differences into account, the accuracy and reliability of fault diagnosis may be affected.

[0007] Therefore, it is necessary to improve one or more problems existing in the above-mentioned related technical solutions.

[0008] It should be noted that this section is intended to provide background or context for the technical solutions of the present invention stated in the claims. The description herein is not admitted to be prior art by inclusion in this section. Summary of the invention

[0009] The object of the present invention is to provide a wind turbine fault diagnosis method, thereby solving one or more problems caused by the limitations and defects of the related art at least to a certain extent.

[0010] In a first aspect, the present invention provides a wind turbine fault diagnosis method, comprising:

[0011] Collect audio signals of wind turbines in operation;

[0012] The audio signal is preprocessed by a wavelet analysis signal processing module to obtain a noise reduction signal; wherein the wavelet analysis signal processing module uses different expression functions to process the wavelet coefficients according to the comparison result between the wavelet coefficients and the threshold;

[0013] Inputting the denoised signal into a trained feature extraction model for feature extraction; wherein the feature extraction model uses the denoised signal as a training set and uses an autoencoder neural network for iterative training;

[0014] The signal after feature extraction is classified and mapped through BP neural network, and the fault diagnosis of the wind turbine is determined.

[0015] Optionally, the step of preprocessing the audio signal for noise reduction by using a wavelet analysis signal processing module includes:

[0016] Performing analog-to-digital conversion on the audio signal to obtain a digital signal;

[0017] Decomposing the digital signal into wavelet coefficients of different scales and frequencies;

[0018] Comparing the wavelet coefficient with a threshold, and performing threshold processing using different expression functions according to the comparison result;

[0019] The wavelet coefficients after threshold processing are reconstructed by inverse transformation.

[0020] Optionally, the step of comparing the wavelet coefficient with a threshold and performing threshold processing using different expression functions according to the comparison result includes:

[0021] The expression function of threshold processing is:

[0022]

[0023] Among them, W j,k represents the wavelet coefficient before processing; j, k represent the maximum decomposition level of the wavelet coefficient and the length of the wavelet coefficient respectively; represents the processed wavelet coefficient; λ represents the threshold; a represents the amount of expansion and contraction of the wavelet coefficient in the original function; sign(*) represents the sign function.

[0024] Optionally, the step of preprocessing the audio signal for noise reduction by using a wavelet analysis signal processing module includes:

[0025] The denoised audio signal is downsampled to obtain the denoised signal.

[0026] Optionally, the feature extraction model uses the noise reduction signal as a training set and uses an autoencoder neural network for iterative training, including:

[0027] Get the initial input vector from the training set through the input layer;

[0028] Map the input vector to the hidden layer through the encoding network;

[0029] Reconstructing the calculation results in the hidden layer through a decoding network, and generating an updated input vector to feed back to the input layer;

[0030] Iterative training is performed with the goal of minimizing the reconstruction error.

[0031] Optionally, the step of inputting the noise reduction signal into a trained feature extraction model for feature extraction includes:

[0032] After the autoencoder neural network completes iterative training, the encoder part including the input layer and the hidden layer is used as the feature extraction model.

[0033] Optionally, the step of classifying and mapping the signal after feature extraction through a BP neural network and determining the fault diagnosis of the wind turbine generator set includes:

[0034] The noise reduction signals of wind turbines under different working conditions are used as the test set;

[0035] The mapping relationship between the noise reduction signal and different working conditions is automatically learned based on the BP neural network through the test set.

[0036] Optionally, the step of classifying and mapping the signal after feature extraction through a BP neural network and determining the fault diagnosis of the wind turbine generator set includes:

[0037] The weights and biases of BP neural network are optimized by particle swarm optimization algorithm.

[0038] Optionally, the different operating conditions of the wind turbine generator set include at least: normal working state, dynamic and static friction, rotor misalignment and rotor imbalance.

[0039] The technical solution provided by the present invention may include the following beneficial effects:

[0040] In the present invention, the audio signal is preprocessed by variable threshold wavelet analysis, and the noise reduction process is adjusted to retain some useful signals in the wavelet coefficients. By extracting the features in the audio signal, it is helpful to accurately capture the changing features of the unit's operating status and realize real-time monitoring of the wind turbine unit's operating status. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The accompanying drawings herein are incorporated into and constitute a part of the specification, illustrate embodiments consistent with the present invention, and together with the specification are used to explain the principles of the present invention. Obviously, the accompanying drawings described below are only some embodiments of the present invention, and for those of ordinary skill in the art, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0042] Figure 1 A schematic diagram showing a flow chart of a wind turbine fault diagnosis method in an exemplary embodiment of the present invention;

[0043] Figure 2 A schematic diagram showing the judgment logic of a wind turbine fault diagnosis method in an exemplary embodiment of the present invention is shown. DETAILED DESCRIPTION

[0044] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the present invention will be more comprehensive and complete and fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0045] In addition, the accompanying drawings are only schematic illustrations of embodiments of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings represent the same or similar parts, and their repeated descriptions will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities.

[0046] The present invention provides a method for diagnosing faults of a wind turbine generator set. Figure 1 As shown in , the following steps are included:

[0047] Step S100: collecting audio signals of a wind turbine in operation.

[0048] Step S200: pre-processing the audio signal by a wavelet analysis signal processing module to obtain a noise reduction signal, wherein the wavelet analysis signal processing module processes the wavelet coefficients using different expression functions according to the comparison result between the wavelet coefficients and the threshold.

[0049] Step S300: Input the denoised signal into the trained feature extraction model for feature extraction, wherein the feature extraction model uses the denoised signal as a training set and uses an autoencoder neural network for iterative training.

[0050] Step S400: classify and map the signal after feature extraction through the BP neural network, and determine the fault diagnosis of the wind turbine.

[0051] It should be understood that a sound sensor is installed at the position of the wind turbine generator transmission generator to collect audio signals inside and around the wind turbine, including audio signals of each component in normal and abnormal operating conditions, as input data for wind turbine audio signal fault diagnosis; the audio signal is subjected to signal denoising using wavelet analysis signal processing technology, and an improved threshold function is constructed to retain some useful signals in the wavelet coefficients, adjust the denoising process, and reconstruct the signal. Real-time monitoring of the wind turbine operating status is achieved; the denoised signal obtained after preprocessing is used as input to construct an autoencoder neural network model, including an encoder and a decoder, and the training set audio signal data is used for training, the encoder part after training is retained, the audio signal features are extracted, and used as a feature extraction module; fault diagnosis based on intelligent BP neural network. According to the normal and fault audio information collected from the wind turbine operating status obtained by real-time monitoring, the signal features are characterized by the feature extraction module. Including normal working conditions, dynamic and static friction, rotor misalignment and imbalance, an intelligent BP neural network is established to perform fault diagnosis on the model.

[0052] It is also necessary to understand that audio signal acquisition: sound sensors are installed at the wind turbine drive generator to collect audio signals inside and around the wind turbine, including audio signals of each component in normal and abnormal operating conditions, as input data for wind turbine audio signal fault diagnosis.

[0053] It should also be understood that the sound sensor of a wind turbine is a device used to monitor and diagnose the operating status of a wind turbine. By collecting audio signals inside or around the wind turbine, the fault diagnosis method designed by the invention provides data for monitoring the operating status of the wind turbine. orig (t) includes vibration and friction sounds from different components. The sound sensor installed on the wind turbine is connected to the data acquisition system, and the collected audio signal is transmitted to the central processing unit for analysis and processing. Specifically, the sound sensor and the ring network switch are transmitted to the industrial computer, which is connected to the remote control device through the wireless network module, and the audio signal is compressed and stored as the original audio signal.

[0054] It is also necessary to understand the threshold-modified wavelet audio signal preprocessing. In the audio signal processing process, the wavelet analysis signal processing technology method is used to perform signal noise reduction on the audio signal. In order to retain some useful signals in the wavelet coefficients, an improved threshold function is constructed, the noise reduction process is adjusted, and the signal is reconstructed. Real-time monitoring of the operating status of the wind turbine is achieved.

[0055] Another thing that needs to be understood is the audio signal feature extraction of the autoencoder neural network. The noise reduction signal obtained after preprocessing is used as input to build an autoencoder neural network model, including an encoder and a decoder. The training set audio signal data is used for training, and the encoder part after training is retained to extract the audio signal features and use it as a feature extraction module.

[0056] It is also necessary to understand that variable threshold wavelet analysis can effectively extract features from audio signals of wind turbines, including normal working conditions, dynamic and static friction, rotor misalignment and imbalance, which helps to accurately capture the changing characteristics of the unit's operating status; threshold-improved wavelet audio signal preprocessing. In the process of audio signal processing, the audio signal is subjected to signal denoising using wavelet analysis signal processing technology. In order to retain some useful signals in the wavelet coefficients, an improved threshold function is constructed, the denoising process is adjusted, and the signal is reconstructed. Real-time monitoring of the operating status of wind turbines is achieved; the particle swarm algorithm optimizes the weights and bias values ​​of the BP neural network, making the network training faster and more accurate, avoiding the dilemma of the network falling into the local optimal solution, and improving the accuracy and reliability of fault diagnosis.

[0057] The above wind turbine fault diagnosis method is adopted to pre-process the audio signal through variable threshold wavelet analysis and adjust the noise reduction process to retain some useful signals in the wavelet coefficients. By extracting the features in the audio signal, it is helpful to accurately capture the changing characteristics of the unit's operating status and realize real-time monitoring of the wind turbine's operating status.

[0058] Next, we will refer to Figure 1 to Figure 2 Each step of the wind turbine fault diagnosis method in this example embodiment is described in more detail.

[0059] In some embodiments, reference Figure 2 As shown in , step S200 includes:

[0060] Step S210: performing analog-to-digital conversion on the audio signal to obtain a digital signal.

[0061] Step S220: decompose the digital signal into wavelet coefficients of different scales and frequencies.

[0062] Step S230: Compare the wavelet coefficients with the threshold, and perform threshold processing using different expression functions according to the comparison result.

[0063] Step S240: reconstructing the wavelet coefficients after threshold processing by inverse transformation.

[0064] It should be understood that the audio signal collected from the sound sensor is transmitted to the audio signal processing in the remote control device, and the signal noise reduction processing method is used to clean the audio data, remove abnormal values, and retain valid audio information. orig (t) Perform a series of preprocessing steps to improve signal quality and reduce noise.

[0065] Analog-to-digital conversion: Since the original audio signal is an analog signal, it needs to be converted into a digital signal to obtain a discrete digital audio signal when it is transmitted to the computer remote control device for signal noise reduction. The conversion of analog signals to digital signals is achieved through the ADC digital-to-analog converter.

[0066] Filtering and denoising: Filtering and denoising the audio signal to reduce the background noise in the signal. The present invention adopts an improved wavelet threshold audio signal denoising method for filtering. Wavelet signal denoising is an effective method for removing noise from a signal, which can retain the main features of the signal. The method first decomposes the signal into wavelet coefficients of different scales and frequencies, then performs threshold processing on these coefficients, and finally performs inverse wavelet transform to reconstruct the signal to obtain a signal with noise removed.

[0067] First, determine the basic wavelet function ψ(t), translate the basic wavelet function ψ(t) according to the time scale axis, and obtain the wavelet sequence:

[0068]

[0069] Among them, a represents the expansion of the original function, and b is the displacement of the original function on the time scale. Secondly, the wavelet transform is as follows:

[0070]

[0071] Among them, W x (a,b) represents a function associated with a and b.

[0072] Finally, the processed wavelet coefficients are reconstructed according to the wavelet reconstruction algorithm. The wavelet reconstruction process includes inverse wavelet transform, which restores the wavelet coefficients processed by threshold value to the signal x(t) in the time domain.

[0073]

[0074] Among them, Cψ is the conversion coefficient. That is, x(t) is the reconstructed audio signal.

[0075] In some embodiments, reference Figure 2 As shown in , step S230 includes:

[0076] The expression function of threshold processing is:

[0077]

[0078] Among them, W j,k represents the wavelet coefficient before processing; j, k represent the maximum decomposition level of the wavelet coefficient and the length of the wavelet coefficient respectively; represents the processed wavelet coefficient; λ represents the threshold; a represents the amount of expansion and contraction of the wavelet coefficient in the original function; sign(*) represents the sign function.

[0079] It is important to understand that a noisy signal contains the superposition of the original signal and the noise signal. Since the original signal and the noise signal have different characteristics at different decomposition levels, after the noisy frequency signal is subjected to wavelet transform, the wavelet coefficients obtained correspond to the original signal and the noise signal respectively. Through the coefficient selection method, the wavelet coefficients corresponding to the noise are eliminated, and the wavelet coefficients corresponding to the original signal are retained to reconstruct the signal.

[0080] In order to remove the noise coefficient and retain the main features of the signal, threshold processing is performed. The threshold λ in formula (4) is calculated by the following formula:

[0081]

[0082] Where N represents the signal sampling length and σ represents the standard deviation of the noise.

[0083] In the related art, according to the set fixed threshold λ, when the wavelet coefficient is less than λ, it is considered that the coefficient is composed of noise, and it is set to 0 or eliminated according to the threshold; if the wavelet coefficient is greater than λ, it is considered that the coefficient is composed of useful signals, and the audio signal is retained. Such a threshold function selection affects the effect of noise reduction. The improved threshold function such as formula (4) improves the shortcomings of the traditional threshold function noise reduction.

[0084] In some embodiments, reference Figure 2 As shown in , step S300 includes:

[0085] The denoised audio signal is downsampled to obtain a denoised signal.

[0086] It should be understood that downsampling a high-sampling-rate audio signal reduces the sampling rate of the digital audio signal, that is, reduces the sampling frequency of the signal. Downsampling is usually used to reduce the amount of signal data, thereby saving storage space and transmission bandwidth, and can simplify the complexity of subsequent signal processing.

[0087] In some embodiments, reference Figure 2 As shown in , step S300 includes:

[0088] The initial input vector is obtained from the training set through the input layer.

[0089] The input vector is mapped to the hidden layer through the encoding network.

[0090] The calculation results in the hidden layer are reconstructed through the decoding network, and an updated input vector is generated and fed back to the input layer.

[0091] Iterative training is performed with the goal of minimizing the reconstruction error.

[0092] It is important to understand that the main task of an autoencoder is to encode and decode data. The process involves compressing the input data to extract key features, and then reconstructing the original data through a decoding process. An autoencoder can be considered a combination of an encoder and a decoder, where the input layer and the hidden layer constitute the encoder, which maps the input vector to the hidden layer, and the output value reconstructs the input vector based on the information received by the decoder from the hidden layer.

[0093] The encoding maps the input vector to the hidden layer:

[0094] p i =f(w e x i +q e ) (6)

[0095] Among them, w e ,q e are the weights and biases in the encoding network model respectively; i is the number of layers.

[0096] Decoding reshapes the hidden layer calculation results:

[0097]

[0098] Among them, w d ,q d Represent the weight and bias of the decoding part respectively. i It represents the features extracted by the encoder part that can represent the input data, and f(·) is the activation function.

[0099] Set the cost function:

[0100] The cost function is used to calculate the input and output errors of the autoencoder neural network, and the mean square error function is used as the cost function:

[0101]

[0102] During the training process, the autoencoder reconstructs the input vector and minimizes the reconstruction error by minimizing the loss function and optimizing the parameters.

[0103] In some embodiments, reference Figure 2 As shown in , step S300 includes:

[0104] After the autoencoder neural network completes iterative training, the encoder part including the input layer and the hidden layer is used as the feature extraction model.

[0105] It should be understood that the pre-processed noise reduction signal x(t) is used as input to establish an autoencoder neural network model, which includes an encoder and a decoder. The model is trained using the training set audio signal data, and the encoder part is retained after the training is completed. Then, the encoder part is used to extract the features of the audio signal and used as a feature extraction module.

[0106] Feature extraction module: After the training, only the encoder part is retained as the data feature extraction module. This module is responsible for extracting key features from the audio signals in the training set to determine the local features of the audio signals in the normal operation and fault state of the wind turbine, thus enhancing the fault detection capability of the model.

[0107] In some embodiments, reference Figure 2 As shown in , step S400 includes:

[0108] The weights and biases of BP neural network are optimized by particle swarm optimization algorithm.

[0109] It should be understood that the PSOBP neural network model is established by using the particle swarm algorithm to optimize the BP neural network. The particle swarm algorithm is used to search for the optimal weights and thresholds within a certain range and input them into the BP neural network. This accelerates the training speed of the network, improves the convergence accuracy, and prevents the network from falling into local extreme values. Through the intelligent BP neural network, the fault information of the wind turbine can be quickly diagnosed. The feature model obtained by combining the variable threshold autoencoder neural network is analyzed with the test set data, and the operation fault analysis of the wind turbine is realized based on the intelligent BP neural network after the feature model.

[0110] Particle swarm optimization is a heuristic optimization algorithm inspired by the predation behavior of bird flocks. In PSO, candidate solutions are regarded as "particles" in the group, and these particles search for the optimal solution by moving in the solution space. The PSO algorithm searches for the optimal solution by simulating the iterative movement of particles in the solution space.

[0111] In the search space dimension D, the number of particle swarms is set to N. In one iteration t, the speed of each particle i is recorded Location Optimal location And the global optimal position g t . Then the particle speed and position update formula are as follows:

[0112]

[0113] In the formula, c 1 ,c 2 is the learning factor, r 1 ,r 2 is a uniform random number in the interval [0,1]. w represents the inertia weight:

[0114]

[0115] Among them, w takes values ​​in the interval [w min ,w max ], t represents the current iteration number, T max Represents the maximum number of iterations. The optimal position and the optimal solution of the population are obtained by setting the fitness function, that is, the objective function. The fitness function represents the mean square error between the output value of the BP neural network and the target output:

[0116]

[0117] Among them, y ij ,t ij are the actual output value and expected output value of the neural network after training the test sample.

[0118] Specifically, the execution process of the PSO algorithm is as follows:

[0119] 1) Initialize population size, particle speed and position information;

[0120] 2) Calculate the objective function value of each particle according to formula 12;

[0121] 3) Compare the particle objective function with the individual extreme value and update it.

[0122] 4) Compare the objective function value with the global optimal solution and update it;

[0123] 5) Update particle velocity and position according to formula 9 and formula 10;

[0124] In some embodiments, reference Figure 2 As shown in , step S400 includes:

[0125] The noise reduction signals of wind turbines under different working conditions are used as the test set.

[0126] The mapping relationship between the noise reduction signal and different working conditions is automatically learned based on the BP neural network through the test set.

[0127] It is important to understand that the BP neural network is used to solve classification and regression problems. It consists of an input layer, a hidden layer, and an output layer, and each layer contains multiple neurons. The basic idea of ​​the BP neural network is to automatically learn the mapping relationship between input and output through training sample data to achieve classification of unknown data.

[0128] Specifically, the testing process is concretized as fault diagnosis, and sample data under normal operation and fault conditions of the wind turbine are tested to verify the fault diagnosis results of the network model.

[0129] Let X = {x 1 ,...,x t} is the input value of the j input nodes of the input layer of the BP neural network, j = 1, 2, ..., m, the hidden layer contains p neurons, w ij represents the weight calculation result of the i-th neuron in the hidden layer and the j-th node in the input layer. The bias of the i-th node in the hidden layer is θ i Indicates. k represents the actual output of the output layer, a k Indicates the actual offset. The input is a signal of four different working conditions including normal working state, dynamic and static friction, rotor misalignment and imbalance, and the output is the corresponding fault diagnosis result.

[0130] Input N to the hidden layer of the BP neural network i available:

[0131]

[0132] The calculated result N i Substitute it into the activation function h of the hidden layer and calculate the output of the i-th node in the hidden layer:

[0133]

[0134] Combined with formula (14), the kth node of the output layer is N k represents the input value, then:

[0135]

[0136] In the back propagation of the model, starting from the output layer, each node is calculated by the error ladder descent method to adjust the connection value of each layer. The weight adjustment formula is as follows:

[0137]

[0138] Among them, Δw ij Represents the weight adjustment of the hidden layer, and η represents the network learning rate. Fault diagnosis based on BP neural network is essentially the classification of audio signals of wind turbine operation. By combining feature data with corresponding categories, and then using test samples to test the model, the fault diagnosis results of wind turbine operation are obtained.

[0139] When the BP algorithm trains the network model, the weights and biases of each training are randomly assigned by the system within a certain range, resulting in great differences in the results of each training. Sometimes it is difficult to converge to the optimal solution, and even if it converges to the optimal solution, it takes a long number of steps.

[0140] Therefore, the particle swarm algorithm is used to find the optimal weight and bias value within a certain range and apply it to the BP neural network. This can speed up the network training, improve the convergence accuracy, and avoid the network falling into local extreme values. This enables the fan fault to be diagnosed quickly and accurately.

[0141] In some embodiments, reference Figure 2 As shown in , different working conditions of wind turbines include at least: normal working state, dynamic and static friction, rotor misalignment and rotor imbalance. It should be understood that fault diagnosis is based on intelligent BP neural network. According to the normal and fault audio information collected from the operating state of the wind turbine obtained by real-time monitoring, the characteristics of the signal are characterized by a feature extraction module. Including normal working state, dynamic and static friction, rotor misalignment and imbalance, several different working conditions, an intelligent BP neural network is established to perform fault diagnosis on the model.

[0142] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example" or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification.

[0143] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art that are not disclosed by the present invention. The specification and examples are to be considered exemplary only, and the true scope and spirit of the present invention are indicated by the appended claims.

Claims

1. A wind turbine fault diagnosis method, characterized in that: include: Collect audio signals of wind turbines in operation; The audio signal is preprocessed by a wavelet analysis signal processing module to obtain a noise reduction signal; wherein the wavelet analysis signal processing module uses different expression functions to process the wavelet coefficients according to the comparison result between the wavelet coefficients and the threshold; Inputting the denoised signal into a trained feature extraction model for feature extraction; wherein the feature extraction model uses the denoised signal as a training set and uses an autoencoder neural network for iterative training; The signal after feature extraction is classified and mapped through BP neural network, and the fault diagnosis of the wind turbine is determined.

2. The wind turbine fault diagnosis method according to claim 1, characterized in that: The step of pre-processing the audio signal for noise reduction by using a wavelet analysis signal processing module comprises: Performing analog-to-digital conversion on the audio signal to obtain a digital signal; Decomposing the digital signal into wavelet coefficients of different scales and frequencies; Comparing the wavelet coefficient with a threshold, and performing threshold processing using different expression functions according to the comparison result; The wavelet coefficients after threshold processing are reconstructed by inverse transformation.

3. The wind turbine fault diagnosis method according to claim 2, characterized in that: The step of comparing the wavelet coefficient with the threshold and using different expression functions to perform threshold processing according to the comparison result comprises: The expression function of threshold processing is: Among them, W j,k represents the wavelet coefficient before processing; j, k represent the maximum decomposition level of the wavelet coefficient and the length of the wavelet coefficient respectively; represents the processed wavelet coefficient; λ represents the threshold; a represents the amount of expansion and contraction of the wavelet coefficient in the original function; sign(*) represents the sign function.

4. The wind turbine fault diagnosis method according to claim 3, characterized in that: The step of pre-processing the audio signal for noise reduction by using a wavelet analysis signal processing module comprises: The denoised audio signal is downsampled to obtain the denoised signal.

5. The wind turbine fault diagnosis method according to claim 1, characterized in that: The feature extraction model uses the noise reduction signal as a training set and uses the autoencoder neural network for iterative training, including the following steps: Get the initial input vector from the training set through the input layer; Map the input vector to the hidden layer through the encoding network; Reconstructing the calculation results in the hidden layer through a decoding network, and generating an updated input vector to feed back to the input layer; Iterative training is performed with the goal of minimizing the reconstruction error.

6. The wind turbine fault diagnosis method according to claim 5, characterized in that: The step of inputting the noise reduction signal into the trained feature extraction model for feature extraction comprises: After the autoencoder neural network completes iterative training, the encoder part including the input layer and the hidden layer is used as the feature extraction model.

7. The wind turbine fault diagnosis method according to any one of claims 1 to 6, characterized in that: The step of classifying and mapping the signal after feature extraction through a BP neural network and determining the fault diagnosis of the wind turbine generator set comprises: The noise reduction signals of wind turbines under different working conditions are used as the test set; The mapping relationship between the noise reduction signal and different working conditions is automatically learned based on the BP neural network through the test set.

8. The wind turbine fault diagnosis method according to claim 7, characterized in that: The step of classifying and mapping the signal after feature extraction through a BP neural network and determining the fault diagnosis of the wind turbine generator set comprises: The weights and biases of BP neural network are optimized by particle swarm optimization algorithm.

9. The wind turbine fault diagnosis method according to claim 8, characterized in that: The different working conditions of the wind turbine generator set include at least: normal working state, dynamic and static friction, rotor misalignment and rotor imbalance.

Images