Fan blade fault identification method and device, terminal equipment and medium

Abstract

The invention provides a fan blade fault recognition method and device, terminal equipment and a medium. The method comprises the steps that a vibration signal of a wind turbine generator cabin is collected; performing frequency domain conversion and modal decomposition on the vibration signal to obtain a first signal and a second signal; performing signal stacking on the first signal and the second signal to obtain a third signal with a time-frequency domain characteristic; and inputting the third signal into a blade fault identification model, and outputting a fault identification result of the wind turbine generator cabin by the blade fault identification model. According to the method, the fan blade fault identification precision can be improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of fan blade fault identification, and particularly relates to a fan blade fault identification method, device, terminal equipment and medium. BACKGROUND

[0002] Wind power generation, as an important part of clean energy, has been rapidly developing worldwide in recent years. With the increase of operating time, wind turbine blades are long-term affected by alternating loads and extreme weather conditions, design or manufacturing defects of blades, improper maintenance or lack of regular inspection, resulting in quality imbalance problems of blades. Blade quality imbalance refers to the phenomenon of uneven mass distribution caused by manufacturing defects, external attachments or structural damage, and its main performance is abnormal periodic vibration amplitude. This fault not only affects the efficiency of wind power generation, but also may cause safety accidents, posing a challenge to the reliability and economy of wind power generation.

[0003] Currently, the fault diagnosis methods for wind turbine blade quality mainly include three ways based on physical models, signal features and data-driven. The fault diagnosis method based on physical model simulates the behavior of the blade under normal and abnormal conditions by establishing a mathematical model. This method usually requires professional knowledge and complex calculations to provide accurate location and cause of the fault. The fault diagnosis method based on signal features focuses on analyzing the signals collected by wind turbine blades (such as vibration, sound, temperature, etc.), and extracts features in the signals through signal processing techniques. Since the change of blade mass or radius will cause the rotor to be in an unbalanced state, resulting in periodic lateral oscillation on the rotor shaft, which causes a high false positive rate. With the rapid development of machine learning, data-driven fault diagnosis methods have been widely applied. This method does not rely on physical models, but learns fault patterns directly from historical data, handles complex nonlinear relationships, and adapts to changing operating conditions. Data-driven fault diagnosis methods include machine learning and deep learning techniques. Machine learning techniques include support vector machines, random forests, K-means clustering, and principal component analysis.

[0004] In the data-driven fault diagnosis method, machine learning algorithms often require a large amount of labeled data for model training, which can easily lead to overfitting problems. In the wind farm scenario, due to the highly variable operating environment and complex and diverse fault patterns, it is extremely difficult to obtain a large amount of accurately labeled data, and this problem is more prominent. In contrast, deep learning can reduce the dependence on labeled data through unsupervised or semi-supervised learning. Deep learning models use multi-layer neural network structures to efficiently solve nonlinear problems and automatically learn complex feature representations from raw data - this ability is crucial for detecting subtle changes in wind turbine faults.

[0005] While physical model-based methods can achieve accurate fault location and cause analysis, they require extensive professional knowledge and precise parameters, which are difficult to obtain in actual operation, and these methods cannot cope with complex changes in the operating environment. Signal feature-based methods have made progress in extracting signal features such as vibration and sound, but still face difficulties in extracting weak fault signals (such as slight mass imbalance or aerodynamic imbalance); some studies focus only on a single signal type, failing to fully utilize information from both the time and frequency domains. Machine learning and deep learning methods can handle complex nonlinear relationships, but rely on large amounts of labeled data during training. In environments like wind turbine blades with multiple fault modes and varying operating conditions, obtaining large amounts of high-quality labeled data is difficult and prone to overfitting. Furthermore, while commonly used deep learning methods such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs) are widely used in fault diagnosis, they often neglect the correlation between time-series information and frequency domain features; although some studies have attempted to fuse time and frequency domain features, their network structures are simplistic, and the feature fusion effect is poor, limiting the generalization and robustness of fault diagnosis.

[0006] In summary, traditional wind turbine blade fault identification methods suffer from two fundamental limitations: first, the fragmentation of frequency and time domain features; and second, the homogenization of models due to a single architectural design. These shortcomings will affect the accuracy of wind turbine blade fault identification. Summary of the Invention

[0007] The present invention addresses the technical problem of providing a method, apparatus, terminal equipment, and medium for identifying wind turbine blade faults, thereby improving the accuracy of wind turbine blade fault identification.

[0008] In a first aspect, the present invention provides a method for identifying faults in wind turbine blades, the method comprising the following steps: Collect vibration signals from the wind turbine nacelle; The vibration signal is subjected to frequency domain transformation and mode decomposition to obtain the first signal and the second signal. The first and second signals are stacked to obtain a third signal with time-frequency domain characteristics; The third signal is input into the blade fault identification model, which outputs the fault identification result of the wind turbine nacelle. The blade fault identification model includes a feature extraction module, a feature fusion module, and a fault identification module connected in sequence. The feature extraction module includes a parallel multi-scale convolutional branch and a Transformer branch. The multi-scale convolutional branch is used to extract the multi-dimensional spatial features caused by wind turbine blade faults in the third signal. The Transformer branch is used to extract the temporal features of the third signal. The feature fusion module is used to cross-fuse multi-dimensional spatial features and temporal features to obtain fused features; The fault identification module is used to identify the fault identification results corresponding to the fused features through the fully connected layer.

[0009] Optionally, the expression for the third signal is:

[0010]

[0011]

[0012] in, Indicates the third signal. Indicates the first signal. Indicates the second signal. Indicates signal stacking. Indicates a stacking operation. Indicates frequency The first signal at a certain point characterizes the vibration signal at a certain frequency. Amplitude and phase information at the location, , This represents the total number of sampling points of the signal. Indicates the index of the time-domain sampling point. Indicates the first The vibration signal amplitude at each sampling point The complex rotation coefficients used to construct the butterfly operation in the Fast Fourier Transform are represented. Indicates the first One eigenmode function express The signal amplitude of each modal component Indicates the first The time-varying phase of each modal component.

[0013] Optionally, the multi-scale convolutional branch includes a first convolutional branch, a second convolutional branch, and a third convolutional branch in parallel. The kernel sizes of the different convolutional branches are different. The first convolutional branch is used to capture the high-frequency transient impact characteristics in the third signal caused by the uneven mass of the wind turbine blades. The second convolutional branch is used to capture the periodic harmonic characteristics caused by the gradient of the mass distribution of the wind turbine blades in the third signal. The third convolutional branch is used to capture the low-frequency oscillation mode of the whole machine caused by the mass eccentricity of the wind turbine blades in the third signal and its time-varying trend.

[0014] Optionally, the kernel size in the first convolutional branch is ; The kernel size in the second convolutional branch is... ; The kernel size in the third convolutional branch is... .

[0015] Optionally, extract multi-dimensional spatial features from the third signal caused by wind turbine blade failure, including: The third signal is convolved through the first, second, and third convolution branches respectively, yielding the first, second, and third convolution results. The expression for the convolution operation on the third signal is as follows: , Indicates the first The output signal of each neuron , Indicates the kernel size. Indicates the first The input signal of each neuron, Indicates the first The input signal of the first neuron and the second neuron The weights of each convolutional kernel, , This indicates the bias of the convolution kernel. express Activation function; The results of the first, second, and third convolutions are concatenated to obtain the multi-dimensional spatial features; the expression for the multi-dimensional spatial features is as follows: .

[0016] Optionally, the Transformer branch is a lightweight Transformer; the lightweight Transformer removes the position encoder and decoder.

[0017] Optionally, multi-dimensional spatial features and temporal features can be cross-fused to obtain fused features, including: Multidimensional spatial features are used as query sequences This is used to define the length of the output sequence; , Indicates query weight; Using time-domain features as key-value pair parameters ; , Indicates key weight. Representing time-domain characteristics, , Indicates the value weight; Through calculation formula

[0018] Obtain fusion features ;in, Representing feature dimension, Key features The transpose of .

[0019] In a second aspect, the present invention provides a wind turbine blade fault identification device, comprising: The data acquisition module is used to collect vibration signals from the wind turbine nacelle. The first data processing module is used to perform frequency domain conversion and mode decomposition on the vibration signal to obtain the first signal and the second signal. The second data processing module is used to stack the first signal and the second signal to obtain a third signal with time-frequency domain characteristics. The fault identification module is used to input the third signal into the blade fault identification model, which then outputs the fault identification result of the wind turbine nacelle. The blade fault identification model includes a feature extraction module, a feature fusion module, and a fault identification module connected in sequence. The feature extraction module includes parallel multi-scale convolutional branches and Transformer branches; the multi-scale convolutional branches are used to extract multi-dimensional spatial features caused by wind turbine blade failures in the third signal; the Transformer branches are used to extract temporal features of the third signal. The feature fusion module is used to cross-fuse multi-dimensional spatial features and temporal features to obtain fused features; The fault identification module is used to identify the fault identification results corresponding to the fused features through the fully connected layer.

[0020] Thirdly, the present invention provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described method.

[0021] Fourthly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0022] The present invention has at least the following beneficial effects: By performing frequency domain transformation and mode decomposition on the nacelle vibration signal, a third signal containing both time and frequency domain information is constructed. This overcomes the fragmentation of time and frequency domain features in traditional methods from the data source, thus improving the accuracy of wind turbine blade fault identification. Through parallel multi-scale convolutional and Transformer branches, deep feature extraction is performed on the third signal from both spatial and temporal dimensions, enabling comprehensive and coordinated capture of fault features. This significantly enhances the sensitivity of the blade fault identification model to weak fault signals, achieving high-precision and high-reliability identification of wind turbine blade faults. The blade fault identification model employs a parallel model architecture and cross-fusion mechanism, allowing the model to adaptively focus on the features most relevant to the current fault and suppress noise interference. This design ensures stable diagnostic performance even when facing variable operating conditions not covered by the training data, with significantly better generalization ability than single-architecture models, further improving the accuracy of wind turbine blade fault identification. Attached Figure Description

[0023] The accompanying drawings are provided to further understand the technical solutions of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the technical solutions of the present invention, and do not constitute a limitation on the technical solutions of the present invention.

[0024] Figure 1 This is a flowchart of a wind turbine blade fault identification method in one embodiment of this application; Figure 2 This is a structural diagram of a blade fault identification model in one embodiment of this application; Figure 3 This is a structural diagram of a multi-scale convolution branch in one embodiment of this application; Figure 4 This is a structural diagram of a lightweight Transformer in one embodiment of this application; Figure 5 This is a structural diagram of a wind turbine blade fault identification device according to one embodiment of this application; Figure 6 This is a structural diagram of a terminal device in one embodiment of this application. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] To address the issue of poor accuracy in traditional wind turbine blade fault identification methods, this invention provides a method, device, terminal equipment, and medium for wind turbine blade fault identification. This method constructs a third signal containing both time and frequency domain information by performing frequency domain transformation and mode decomposition on the nacelle vibration signal. This overcomes the fragmentation of time and frequency domain features in traditional methods from the data source, thus improving the accuracy of wind turbine blade fault identification. Through parallel multi-scale convolutional and Transformer branches, deep feature extraction is performed on the third signal from both spatial and temporal dimensions, enabling comprehensive and coordinated capture of fault features. This significantly enhances the sensitivity of the blade fault identification model to weak fault signals, achieving high-precision and high-reliability identification of wind turbine blade faults. The blade fault identification model employs a parallel model architecture and a cross-fusion mechanism, allowing the model to adaptively focus on the features most relevant to the current fault and suppress noise interference. This design ensures stable diagnostic performance even when facing variable operating conditions not covered by the training data, with significantly better generalization ability than single-architecture models, further improving the accuracy of wind turbine blade fault identification.

[0027] Example 1 like Figure 1 As shown, the wind turbine blade fault identification method provided by the present invention specifically includes steps 11 to 14.

[0028] Step 11: Collect vibration signals from the wind turbine nacelle.

[0029] Specifically, an accelerometer (such as an ICP-type accelerometer with a sensitivity of 100 mV / g) is installed at a critical location in the wind turbine nacelle (such as near the main shaft bearing) to collect vibration signals at a sampling frequency of 10 kHz. The signal acquisition duration should cover at least several rotational cycles of the wind turbine (e.g., 30 seconds) to ensure the capture of periodic fault characteristics. For example, using a 2MW wind turbine as an example, longitudinal vibration signals of the nacelle are collected at the rated speed of 15 rpm. The signal includes the blade passing frequency (approximately 0.25 Hz) and its harmonic components.

[0030] Another feasible implementation is to collect vibration signals from the wind turbine nacelle using a SCADA (Supervisory Control and Data Acquisition) system. It should be understood that a SCADA system is an automation system used to monitor and control industrial processes. It is typically used to collect, monitor, and analyze the status data of equipment or systems in real time.

[0031] Step 12: Perform frequency domain transformation and mode decomposition on the vibration signal to obtain the first signal and the second signal.

[0032] In this embodiment of the invention, a Fast Fourier Transform (FFT) is used to perform frequency domain conversion on the vibration signal, transforming the vibration signal from a time-domain signal to a frequency-domain signal. Taking a SCADA system as an example, since the vibration signal acquired by the SCADA system has discrete characteristics, it is divided into a finite number of sampling points in time. For a discrete signal of finite length... The formula for calculating its discrete Fourier transform is as follows:

[0033] in, The complex rotation coefficients used to construct the butterfly operation in the Fast Fourier Transform are represented. Indicates frequency The first signal at a certain point characterizes the vibration signal at a certain frequency. Amplitude and phase information at the location, , This represents the total number of sampling points of the signal. Indicates the index of the time-domain sampling point. Indicates the first The amplitude of the vibration signal at each sampling point.

[0034] However, the Fast Fourier Transform (FFT) loses time-series information during signal conversion, providing only the spectral information of the vibration signal at a certain moment, and failing to provide information on how the signal changes over time. To address this issue, this invention introduces mode decomposition. Specifically, VMD (Variational Mode Decomposition) is used to decompose the vibration signal into multiple Intrinsic Mode Functions (IMFs), each corresponding to a specific frequency range of the signal. Unlike FFT, VMD decomposition can provide information on how the vibration signal changes over time because it preserves time-series information during the decomposition process.

[0035] In VMD decomposition, each intrinsic mode function (IMF) is an AM-FM (amplitude modulation-frequency modulation) signal, and its expression is:

[0036]

[0037] in, Indicates the first One eigenmode function express The signal amplitude of each modal component Indicates the first The time-varying phase of each modal component, Indicates the first The instantaneous frequency of each modal component, , This indicates the total number of sampling points for the signal.

[0038] It should be noted that, in this embodiment of the invention, the variational problem is to decompose the input signal (vibration signal) into... Each IMF is analyzed using Hilbert transform, and the signal is demodulated using Gaussian smoothing (i.e., the square root of the L2 norm gradient) to obtain the bandwidth of each mode function. Under constraints, the sum of the components equals the input signal; therefore, the variational problem is described as follows:

[0039] in, Represents the first derivative in the time domain. Represents the Dirac function, Represents the imaginary unit. This indicates the original cabin vibration signal. , , Indicates the first One modal component, Indicates the first k The center frequency of each mode.

[0040] It is worth mentioning that FFT extracts frequency domain features (such as harmonic components), while VMD extracts time domain modal features (such as transient impulses). The two complement each other, overcoming the limitations of single-domain analysis and laying the foundation for subsequent fusion.

[0041] Step 13: Stack the first signal and the second signal to obtain a third signal with time-frequency domain characteristics.

[0042] Specifically, the expression for the third signal is: .

[0043] in, Indicates the third signal. Indicates the first signal. Indicates the second signal. Indicates signal stacking. Indicates a stacking operation. Indicates frequency The first signal at a certain point characterizes the vibration signal at a certain frequency. Amplitude and phase information at the location, Indicates the index of the time-domain sampling point. Indicates the first The amplitude of the vibration signal at each sampling point.

[0044] It is worth mentioning that the stacking operation integrates time-domain and frequency-domain information to generate an input signal rich in fault characteristics, which solves the problem of feature fragmentation and helps to improve the sensitivity to weak faults.

[0045] Step 14: Input the third signal into the blade fault identification model, and the blade fault identification model outputs the fault identification result of the wind turbine nacelle.

[0046] In embodiments of the present invention, such as Figure 2 As shown, the blade fault identification model includes a feature extraction module 201, a feature fusion module 202, and a fault identification module 203 connected in sequence.

[0047] The feature extraction module 201 includes parallel multi-scale convolutional branches 201A and 201B. Multi-scale convolutional branch 201A extracts multi-dimensional spatial features caused by wind turbine blade faults in the third signal. Transformer branch 201B extracts temporal features of the third signal. The feature fusion module 202 cross-fused the multi-dimensional spatial and temporal features to obtain fused features. The fault identification module 203 identifies the fault identification result corresponding to the fused features through a fully connected layer.

[0048] Furthermore, Figure 3 The structure of the multi-scale convolution branch is shown. The multi-scale convolution branch includes a first convolution branch, a second convolution branch, and a third convolution branch in parallel. The convolution kernel sizes of the different convolution branches are different. The first convolution branch is used to capture the high-frequency transient impact characteristics (such as pulses caused by blade surface damage) in the third signal caused by the uneven mass of the wind turbine blades. The second convolution branch is used to capture the periodic harmonic characteristics (such as vibrations caused by the mass distribution gradient) caused by the mass distribution gradient of the wind turbine blades in the third signal. The third convolution branch is used to capture the low-frequency oscillation mode of the whole machine and its time-varying trend (such as the whole machine vibration caused by mass eccentricity) caused by the mass eccentricity of the wind turbine blades in the third signal.

[0049] In one feasible implementation, the kernel size in the first convolution branch is The stride is 1, and the padding is 0; the kernel size in the second convolutional branch is... The stride is 1, and the padding is 1; the kernel size in the third convolutional branch is... The step size is 1, and the padding size is 3.

[0050] In one feasible implementation, the process of extracting the multi-dimensional spatial features caused by wind turbine blade failure from the third signal specifically includes steps 14.1 to 14.2.

[0051] Step 14.1: Perform convolution operations on the third signal through the first convolution branch, the second convolution branch, and the third convolution branch respectively to obtain the first convolution result, the second convolution result, and the third convolution result.

[0052] Specifically, the expression for convolution operation on the third signal is as follows: , Indicates the first The output signal of each neuron , Indicates the kernel size. Indicates the first The input signal of each neuron, Indicates the first The input signal of the first neuron and the second neuron The weights of each convolutional kernel, , This indicates the bias of the convolution kernel. express Activation function.

[0053] Step 14.2: Concatenate the results of the first, second, and third convolutions to obtain the multi-dimensional spatial features; the expression for the multi-dimensional spatial features is as follows: .

[0054] In traditional Transformer models, positional encoding is an indispensable part, used to provide the positional information of elements in a sequence. In many tasks, especially classification or sequence embedding tasks, it is often only necessary to encode the input sequence, without generating a new sequence. In the wind turbine blade fault identification task, omitting positional encoding not only reduces the number of model parameters but also lowers computational complexity, making the model more concise and efficient. Therefore, in this embodiment of the invention, the Transformer branch is a lightweight Transformer; the lightweight Transformer removes both positional encoding and the decoder. Omitting positional encoding not only reduces the number of parameters in the model but also lowers computational complexity, resulting in a simpler and more efficient model. The lightweight Transformer mainly consists of three parts: an input layer, an encoding layer, and an output layer, with the specific structure as follows: Figure 4 As shown.

[0055] The feature encoder consists of multiple encoder layers, each containing a multi-head self-attention mechanism and a feed-forward neural network. The feed-forward neural network performs a non-linear transformation on the input, enhancing the model's expressive power. Each encoder layer also contains a residual connection and a normalization layer.

[0056] The multi-head self-attention mechanism divides the input sequence into multiple subsequences and calculates attention weights for each subsequence. These attention weights are then combined to gain a more comprehensive understanding of the input sequence. The self-attention mechanism calculates the similarity between the query matrix Q and the key matrix K by scaling the dot product, and then normalizes the similarity using the SoftMax function. Finally, the normalized similarity is multiplied by the value matrix V.

[0057]

[0058]

[0059]

[0060] in, Represents the characteristic matrix, Represents the weight matrix. Indicates the first The weight matrix of each attention head.

[0061] A feed-forward neural network (FFNN) is a fundamental component of a neural network, used in Transformer models to enhance the model's non-linear expressive capabilities. A FFNN typically consists of two or more fully connected layers, with each layer undergoing a non-linear transformation via an activation function (such as ReLU), as shown below:

[0062] in, The normalized output signal matrix of the attention layer. This represents the weight matrix of the first fully connected layer. This represents the bias term of the first fully connected layer. This represents the weight matrix of the second fully connected layer. This represents the bias term of the second fully connected layer.

[0063] Residual connections are a key component of the Transformer model. They allow the model to learn a direct mapping to the input sequence and help mitigate the vanishing or exploding gradient problems during training. Furthermore, to accelerate training and reduce overfitting, the results after residual connections are typically normalized, as shown in the following expression:

[0064] in, The input signal matrix of the coding layer, The input signal matrix of the feedforward neural network, This is the output matrix of the coding layer. This is a normalization operation.

[0065] By using feedforward neural networks to perform nonlinear transformations, residual connections, and layer normalization on the input signal, the vanishing or exploding gradient problem during training can be mitigated, improving the model's stability. Finally, these sub-layers extract temporal features from the sequence data, and these temporal features, along with other information, are input into the output layer to generate the final output sequence.

[0066] The following describes the process of cross-fusion of multi-dimensional spatial features and temporal features to obtain fused features, specifically including steps I to III.

[0067] Step I: Use multi-dimensional spatial features as the query sequence This is used to define the length of the output sequence; , Indicates query weight; Step II: Use time-domain features as key-value pair parameters ; , Indicates key weight. Representing time-domain characteristics, , Indicates the value weight; Step III, through calculation formula

[0068] Obtain fusion features ;in, Representing feature dimension, Key features The transpose of .

[0069] In this embodiment of the invention, feature input is fused. The fully connected layer is used for classification. Specifically, the fully connected layer consists of two hidden layers (128 and 64 neurons respectively) and an output layer (the number of neurons equals the number of fault categories, such as 4 categories: normal, imbalanced, crack, and icing). The Softmax function is used to output the fault probability, and the highest probability is taken as the recognition result.

[0070] This invention comprehensively captures fault features through time-frequency domain signal stacking, multi-scale convolution and Transformer parallel extraction, and cross-attention fusion, significantly improving the accuracy and robustness of identifying weak faults.

[0071] In a comparative embodiment of the present invention, the wind turbine blade fault identification method provided by the present invention was compared with the traditional method (fault identification method based on finite element model). The test results show that the fault identification accuracy of the wind turbine blade fault identification method provided by the present invention reaches 98.5%, which is more than 15% higher than that of the traditional method.

[0072] As can be seen, by performing frequency domain transformation and mode decomposition on the nacelle vibration signal, a third signal containing both time and frequency domain information is constructed. This overcomes the fragmentation of time and frequency domain features in traditional methods from the data source, which is beneficial to improving the accuracy of wind turbine blade fault identification. Through parallel multi-scale convolutional branches and Transformer branches, deep feature extraction is performed on the third signal from both spatial and temporal dimensions, enabling comprehensive and coordinated capture of fault features. This greatly enhances the sensitivity of the blade fault identification model to weak fault signals, thus achieving high-precision and high-reliability identification of wind turbine blade faults. The blade fault identification model employs a parallel model architecture and cross-fusion mechanism, allowing the model to adaptively focus on the features most relevant to the current fault and suppress noise interference. This design enables the model to maintain stable diagnostic performance even when facing variable operating conditions not covered by the training data, and its generalization ability is significantly better than that of a single-architecture model, which is beneficial to improving the accuracy of wind turbine blade fault identification.

[0073] Example 2 like Figure 5 As shown, the present invention also provides a wind turbine blade fault identification device, the device 500 comprising: The data acquisition module 501 is used to collect vibration signals from the wind turbine nacelle. The first data processing module 502 is used to perform frequency domain conversion and mode decomposition on the vibration signal to obtain the first signal and the second signal. The second data processing module 503 is used to stack the first signal and the second signal to obtain a third signal with time-frequency domain characteristics. The fault identification module 504 is used to input the third signal into the blade fault identification model, and the blade fault identification model outputs the fault identification result of the wind turbine nacelle. The blade fault identification model includes a feature extraction module, a feature fusion module, and a fault identification module connected in sequence. The feature extraction module includes parallel multi-scale convolutional branches and Transformer branches. The multi-scale convolutional branches are used to extract multi-dimensional spatial features caused by wind turbine blade faults in the third signal. The Transformer branches are used to extract the temporal features of the third signal. The feature fusion module is used to cross-fuse the multi-dimensional spatial features and temporal features to obtain fused features. The fault identification module is used to identify the fault identification result corresponding to the fused features through a fully connected layer.

[0074] It should be noted that the information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of this application. Their specific functions and technical effects can be found in the method embodiments section, and will not be repeated here. Those skilled in the art will understand that, for the sake of convenience and brevity, the division of the above-mentioned functional units and modules is only used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0075] like Figure 6 As shown, embodiments of the present invention provide a terminal device, such as... Figure 6 As shown, the terminal device D10 of this embodiment includes: at least one processor D100 ( Figure 6 The diagram shows only one processor, a memory D101, and a computer program D102 stored in the memory D101 and executable on the at least one processor D100, wherein the processor D100 executes the computer program D102 to implement the steps in any of the above method embodiments.

[0076] Specifically, when the processor D100 executes the computer program D102, it collects the vibration signal of the wind turbine nacelle; performs frequency domain conversion and mode decomposition on the vibration signal to obtain a first signal and a second signal; stacks the first signal and the second signal to obtain a third signal with time and frequency domain characteristics; inputs the third signal to the blade fault identification model, and outputs the fault identification result of the wind turbine nacelle from the blade fault identification model. This study constructs a third signal containing both time and frequency domain information by performing frequency domain transformation and mode decomposition on the nacelle vibration signal. This overcomes the fragmentation of time and frequency domain features in traditional methods from the data source, thus improving the accuracy of wind turbine blade fault identification. Through parallel multi-scale convolutional and Transformer branches, deep feature extraction is performed on the third signal from both spatial and temporal dimensions, enabling comprehensive and coordinated capture of fault features. This significantly enhances the sensitivity of the blade fault identification model to weak fault signals, achieving high-precision and high-reliability identification of wind turbine blade faults. The blade fault identification model employs a parallel model architecture and cross-fusion mechanism, allowing the model to adaptively focus on the features most relevant to the current fault and suppress noise interference. This design ensures stable diagnostic performance even when facing variable operating conditions not covered by the training data, with significantly better generalization ability than single-architecture models, further improving the accuracy of wind turbine blade fault identification.

[0077] The processor D100 can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0078] In some embodiments, the memory D101 may be an internal storage unit of the terminal device D10, such as a hard disk or memory of the terminal device D10. In other embodiments, the memory D101 may be an external storage device of the terminal device D10, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the terminal device D10. Furthermore, the memory D101 may include both internal and external storage units of the terminal device D10. The memory D101 is used to store the operating system, applications, bootloader, data, and other programs, such as the program code of the computer program. The memory D101 can also be used to temporarily store data that has been output or will be output.

[0079] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps described in the various method embodiments above.

[0080] This application provides a computer program product that, when run on a terminal device, enables the terminal device to implement the steps described in the various method embodiments above.

[0081] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of protection of this application is limited to these examples; within the framework of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of one or more embodiments of this application as described above, which are not provided in detail for the sake of brevity.

[0082] One or more embodiments in this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of this application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments in this application should be included within the protection scope of this application.

Claims

1. A wind turbine blade fault identification method, characterized by, The method comprises the following steps: Collecting a vibration signal of a wind turbine nacelle; Performing frequency domain conversion and modal decomposition on the vibration signal respectively to obtain a first signal and a second signal; Stacking the first signal and the second signal to obtain a third signal with time-frequency domain characteristics; Inputting the third signal into a blade fault identification model to output a fault identification result of the wind turbine nacelle by the blade fault identification model; the blade fault identification model comprises a feature extraction module, a feature fusion module and a fault identification module connected in sequence; wherein The feature extraction module comprises a multi-scale convolution branch and a Transformer branch in parallel; the multi-scale convolution branch is used to extract multi-dimensional spatial features caused by wind turbine blade faults in the third signal; the Transformer branch is used to extract time domain features of the third signal; The feature fusion module is used to cross-fuse the multi-dimensional spatial features and the time domain features to obtain fusion features; The fault identification module is used to identify the fault identification result corresponding to the fusion features through a full connection layer.

2. The wind turbine blade fault identification method of claim 1, wherein, The expression of the third signal is: wherein, represents a third signal, represents a first signal, represents a second signal, represents a signal stack, represents a stacking operation, represents a first signal at a frequency , characterizing the amplitude, phase information of the vibration signal at a frequency , , represents the total number of sampling points of a signal, represents a time-domain sampling point index, represents the vibration signal amplitude of the th sampling point, , characterizing the complex rotation coefficient used in the construction of the butterfly operation in the fast Fourier transform, represents the th eigenmode function, represents the signal amplitude of the modal component, represents the time-varying phase of the th modal component.

3. The method of claim 2, wherein, The multi-scale convolution branch comprises a first convolution branch, a second convolution branch and a third convolution branch in parallel, and the convolution kernel sizes of different convolution branches are different from each other; the first convolution branch is used to capture high-frequency transient impact features in the third signal caused by uneven wind turbine blade mass; the second convolution branch is used to capture periodic harmonic features in the third signal caused by wind turbine blade mass distribution gradient; and the third convolution branch is used to capture whole machine low-frequency swing modal and time-varying trend caused by wind turbine blade mass eccentricity.

4. The wind turbine blade fault identification method of claim 3, wherein, The convolution kernel size in the first convolution branch is ; The convolution kernel size in the second convolution branch is ; The convolution kernel size in the third convolution branch is .

5. The wind turbine blade fault identification method of claim 4, wherein, The extraction of the multi-dimensional spatial features caused by the wind turbine blade faults in the third signal comprises: The third signal is subjected to convolution operation through the first convolution branch, the second convolution branch and the third convolution branch respectively to obtain first convolution result, second convolution result and third convolution result; the expression of the convolution operation on the third signal is , represents the output signal of the first neuron, , represents the convolution kernel size, represents the input signal of the first neuron, represents the input signal of the first neuron and the weight of the first convolution kernel, , represents the bias of the convolution kernel, represents the activation function;​​ The first convolution result, the second convolution result and the third convolution result are spliced to obtain the multi-dimensional space feature; and an expression of the multi-dimensional space feature is .

6. The wind turbine blade fault identification method of claim 5, wherein, The Transformer branch is a lightweight Transformer; the lightweight Transformer removes position encoding and a decoder.

7. The wind turbine blade fault identification method of claim 6, wherein, The cross-fusion of the multi-dimensional spatial features and the time domain features to obtain the fusion features comprises: using the multi-dimensional spatial features as a query sequence , for defining a sequence length of the output; , denotes a query weight; Temporal features are taken as key-value pair parameters ; , denotes a key weight, denotes a temporal feature, , denotes a value weight; by a calculation formula obtaining fused features ; wherein, denotes a feature dimension, denotes a transpose of a key feature .

8. A wind turbine blade fault identification apparatus, characterized by, The method comprises the following steps: A data acquisition module is configured to collect a vibration signal of a wind turbine nacelle; A first data processing module is configured to perform frequency domain conversion and modal decomposition on the vibration signal respectively to obtain a first signal and a second signal; A second data processing module is configured to stack the first signal and the second signal to obtain a third signal with time-frequency domain characteristics; A fault identification module is configured to input the third signal into a blade fault identification model to output a fault identification result of the wind turbine nacelle by the blade fault identification model; The blade fault identification model comprises a feature extraction module, a feature fusion module and a fault identification module connected in sequence; wherein The feature extraction module comprises a multi-scale convolution branch and a Transformer branch in parallel; the multi-scale convolution branch is used to extract multi-dimensional spatial features caused by wind turbine blade faults in the third signal; the Transformer branch is used to extract time domain features of the third signal; The feature fusion module is configured to cross-fuse the multi-dimensional space features and the time domain features to obtain fused features. The fault identification module is configured to identify a fault identification result corresponding to the fused features through a full connection layer.

9. A terminal device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the method of any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1-9. The computer program is executed by the processor to implement the method of any one of claims 1 to 7.

Images