Wind power equipment fault identification method and system based on time sequence vibration data analysis

Abstract

The invention relates to the technical field of new energy wind power equipment fault detection, and discloses a wind power equipment fault identification method and system based on time sequence vibration data analysis, and the method comprises the steps: obtaining a multi-axis vibration signal during the operation of equipment; performing frequency domain feature extraction to obtain frequency domain feature data of each axis; sampling P * P feature points from the feature data of each axis; filling the corresponding amplitudes into a P * P two-dimensional matrix according to the frequency sequence; combining the two-dimensional matrixes of the axes into a multi-channel spectrum image according to a preset channel mapping rule; and inputting the image into a classification model to complete state recognition. According to the scheme, input size standardization and cross-axis frequency index alignment are achieved, the model can learn multi-axis correlation features conveniently, and fault recognition accuracy and engineering availability are improved. The system comprises an acquisition module, a feature extraction module, a sampling module, a matrix generation module, a channel combination module and a recognition module, sensors can be arranged on a gearbox, a main shaft or a generator body, and three-axis signals are mapped into RGB.

Description

Technical Field

[0001] This invention relates to the field of fault detection technology for new energy wind power equipment, specifically to a method and system for wind power equipment fault identification based on time-series vibration data analysis. Background Technology

[0002] With the widespread deployment of large wind turbine generators in wind farms and offshore wind power applications, multi-channel sensors deployed around key components such as gearboxes, main shafts, and generator bodies have become the primary means of acquiring equipment health information. In engineering practice, frequency domain analysis is often used to extract vibration features, combined with machine learning and deep learning models to distinguish between normal and fault conditions. In recent years, with the evolution of computing and algorithms, transforming time-series vibration data into image or tensor representations that are easy for models to learn, and achieving input dimension standardization and cross-axis information fusion, has become an important direction for promoting intelligent operation and maintenance, conducting online monitoring, and providing early warnings. Under complex operating conditions and noise backgrounds, this type of representation helps enhance feature separability, improve model generalization and diagnostic robustness, and provide a reliable data foundation for wind power equipment condition assessment and operation and maintenance decisions.

[0003] Current technologies still suffer from several problems affecting diagnostic effectiveness and deployment efficiency, primarily including: First, the fusion methods for multi-axis vibration information are relatively crude, often relying on simple splicing or independent discrimination, making it difficult to achieve consistent alignment of cross-axis frequency indices within the same representation space, thus limiting the model's ability to capture linked fault features; Second, commonly used time-frequency representations have inconsistent input sizes and high feature dimensions, introducing additional overhead and instability to model training and online inference; Third, when single-axis or one-dimensional spectral vectors are directly used as input, spatial structure information is insufficient, making it difficult for convolutional models to fully exploit cross-frequency band and cross-axis correlation patterns. These problems limit the accuracy, robustness, and generalization ability of fault identification, thus necessitating a method to improve the accuracy and engineering usability of wind power equipment fault identification. Summary of the Invention

[0004] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a method and system for wind power equipment fault identification based on time-series vibration data analysis, thus solving the aforementioned problems.

[0005] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: a method for fault identification of wind power equipment based on time-series vibration data analysis, the method comprising: S1: Acquire multi-axis vibration signals of wind power equipment during operation; S2: Extract frequency domain features from the multi-axis vibration signal to obtain frequency domain feature data corresponding to each axis; S3: Sample from the frequency domain feature data of each axis There are 1 feature points; among them, This is the default value; S4: Fill the corresponding amplitudes of the feature points into a two-dimensional matrix according to the frequency sequence; S5: Combine the two-dimensional matrices corresponding to each axis according to the preset channel mapping rules to generate a multi-channel spectrum image; S6: Based on the multi-channel spectrum image, input the image into the classification model to identify and classify the operating status of the wind power equipment.

[0006] Furthermore, the multi-axis vibration signal is collected by a sensor installed at at least one monitoring position in the gearbox, main shaft, or generator body, and includes triaxial time-series vibration signals along the X-axis, Y-axis, and Z-axis directions.

[0007] Furthermore, S2 specifically refers to: S21: The multi-axis vibration signal is processed by time-domain segmentation using a sliding window, dividing the continuous time series into multiple adjacent and at least partially overlapping time segments, and using each time segment as a time-series feature unit, with the window length and step size being preset parameters. S22: Perform a fast Fourier transform on the vibration signal of each axis corresponding to each of the time-series feature units, calculate its amplitude spectrum, and use the amplitude spectrum as the frequency domain feature data corresponding to each axis.

[0008] Furthermore, when there are insufficient fault samples for training the classification model, a Gaussian distribution perturbation is injected into the temporal feature unit in the time domain by channel to generate simulated fault samples. The mean and standard deviation of the perturbation are determined by the scaling factor of the mean and standard deviation of the corresponding temporal feature unit, respectively. The simulated fault samples are then combined with existing samples for training the classification model.

[0009] Furthermore, step S4 specifically involves: sorting the feature points in ascending order of frequency, and then filling the amplitude values ​​of the sorted feature points sequentially using row-major order. A two-dimensional matrix is ​​generated, and two-dimensional matrices generated on different axes correspond to the same frequency index at the same matrix position.

[0010] Furthermore, S5 specifically includes: S51: Normalize the element values ​​of the two-dimensional matrix corresponding to each axis to obtain values ​​that are... A two-dimensional matrix; S52: Map the normalized two-dimensional matrix to... The range of pixel values; S53: Combine the two-dimensional matrices corresponding to each axis according to the one-to-one channel mapping rule to generate the multi-channel spectrum image, and the number of channels is consistent with the number of axes, and each channel corresponds to the same frequency index at the same pixel position.

[0011] Furthermore, when the multi-axis vibration signal includes a three-axis time series signal along the X-axis, Y-axis, and Z-axis, the two-dimensional matrices corresponding to the X-axis, Y-axis, and Z-axis are respectively assigned to the red, green, and blue channels of the image.

[0012] Furthermore, S6 specifically involves: inputting the multi-channel spectral image into a classification model based on a deep neural network to output a determination of the operating status of the wind power equipment. The classification model is preferably a convolutional neural network, which includes at least one convolutional layer, a pooling layer, and a fully connected layer. The output layer uses the Softmax function to calculate the class probability.

[0013] This invention also provides a wind power equipment fault identification system based on time-series vibration data analysis, including: The signal acquisition module is used to acquire multi-axis vibration signals of wind power equipment during operation; The frequency domain feature extraction module is used to extract frequency domain features from the multi-axis vibration signal to obtain frequency domain feature data corresponding to each axis. The feature point sampling module is used to sample from the frequency domain feature data of each axis. There are 1 feature points; among them, This is the default value; A two-dimensional matrix generation module is used to fill the corresponding amplitudes of the feature points into a two-dimensional matrix according to the frequency sequence. The channel mapping combination module is used to combine the two-dimensional matrices corresponding to each axis according to the preset channel mapping rules to generate a multi-channel spectral image. The status recognition module is used to input the multi-channel spectrum image into a classification model to identify and classify the operating status of wind power equipment.

[0014] (III) Beneficial Effects Compared with the prior art, the present invention provides a method and system for wind power equipment fault identification based on time-series vibration data analysis, which has the following beneficial effects: 1. A method and system for fault identification of wind power equipment based on time-series vibration data analysis, which obtains fault information by upsampling the frequency domain characteristic data of each axis. The feature points are filled with the corresponding amplitudes in order of frequency sequence. The two-dimensional matrix solves the problems of inconsistent input size and the difficulty of fully utilizing one-dimensional spectral vectors in convolutional models, thereby achieving standardized and spatially structured expression of input dimensions, which facilitates stable model learning and engineering deployment.

[0015] 2. A method and system for wind power equipment fault identification based on time-series vibration data analysis. By combining two-dimensional matrices generated from different axes with the same frequency index at the same matrix position and according to the one-to-one channel mapping rule into a multi-channel spectrum image, the problem of difficult alignment and coarse fusion of cross-axis frequency indexes in traditional multi-axis information fusion is solved. This achieves pixel-level cross-axis synchronous expression and improves the distinguishability and recognition accuracy of linkage faults and cross-axis correlation features. Attached Figure Description

[0016] Figure 1 A flowchart illustrating the wind power equipment fault identification method based on time-series vibration data analysis provided by this invention; Figure 2 A schematic diagram of the overall framework of the wind power equipment fault identification method based on time-series vibration data analysis provided by the present invention; Figure 3 A schematic diagram of the sensor installation location for the wind power equipment fault identification method based on time-series vibration data analysis provided by the present invention; Figure 4 A schematic diagram of a sliding window for the wind power equipment fault identification method based on time-series vibration data analysis provided by the present invention; Figure 5 A schematic diagram comparing normal and fault samples under a single time window for the wind power equipment fault identification method based on time-series vibration data analysis provided by the present invention. Figure 6 An RGB image of normal data for a time window of the wind power equipment fault identification method based on time-series vibration data analysis provided by the present invention; Figure 7 An RGB image of fault data for a time window in the wind power equipment fault identification method based on time-series vibration data analysis provided by the present invention; Figure 8 This is a schematic diagram of the wind power equipment fault identification system based on time-series vibration data analysis provided by the present invention. Detailed Implementation

[0017] 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.

[0018] To enable those skilled in the art to better understand the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0019] Please see Figure 1-2 , Figure 1 A flowchart illustrating the wind power equipment fault identification method based on time-series vibration data analysis provided by this invention; Figure 2 A schematic diagram of the overall framework of the wind power equipment fault identification method based on time-series vibration data analysis provided by the present invention; the wind power equipment fault identification method based on time-series vibration data analysis includes: S1: Acquire multi-axis vibration signals of wind power equipment during operation; S2: Extract frequency domain features from the multi-axis vibration signal to obtain frequency domain feature data corresponding to each axis; S3: Sample from the frequency domain feature data of each axis There are 1 feature points; among them, This is the default value; S4: Fill the corresponding amplitudes of the feature points into a two-dimensional matrix according to the frequency sequence; S5: Combine the two-dimensional matrices corresponding to each axis according to the preset channel mapping rules to generate a multi-channel spectrum image; S6: Based on the multi-channel spectrum image, input the image into the classification model to identify and classify the operating status of the wind power equipment.

[0020] In this embodiment, the method sequentially executes steps S1 to S6: First, multi-axis vibration signals are acquired during equipment operation to serve as the data source for subsequent processing, ensuring the real-time and completeness of subsequent feature construction; then, frequency domain features are extracted from the signals of each axis to obtain frequency domain feature data indexed by frequency and measured by amplitude, which highlights the energy distribution of different frequency components, facilitating subsequent characterization of key patterns with a lower dimension; based on this, samples are taken from the frequency domain feature data of each axis. There are feature points, among which The preset values ​​ensure that different samples have a uniform feature scale before entering the imaging stage, reducing instability caused by inconsistent sizes in subsequent processing. Then, the corresponding amplitudes of the aforementioned feature points are sequentially filled into a two-dimensional matrix according to the frequency sequence, giving the originally one-dimensional frequency domain data a two-dimensional representation with spatial adjacency, thus providing a learnable spatial structure for image-based discrimination methods. Next, according to preset channel mapping rules, the two-dimensional matrices corresponding to each axis are combined to generate a multi-channel spectral image, allowing multi-axis frequency domain information to be jointly presented in the same coordinate system and maintaining a consistent relationship at pixel positions, facilitating the model's simultaneous perception of cross-axis correlation features. Finally, this multi-channel spectral image is input into the classification model to determine the operating status of the output device, forming a processing link from signal acquisition to status recognition. This achieves input dimension standardization, expression form structuring, and cross-axis information synergistic utilization, improving the accuracy of fault identification and the stability of engineering deployment.

[0021] For further details, please refer to Figure 3 , Figure 3 This is a schematic diagram of the sensor installation position for the wind power equipment fault identification method based on time-series vibration data analysis provided by the present invention; the multi-axis vibration signal is collected by a sensor installed at at least one monitoring position in the gearbox, main shaft or generator body, and includes three-axis time-series vibration signals along the X-axis, Y-axis and Z-axis directions.

[0022] In this embodiment, a vibration sensor group is installed at at least one monitoring location in the gearbox, main shaft, or generator body. Preferably, each monitoring location is equipped with a triaxial vibration sensor or an equivalent three-channel acquisition unit to acquire time-series vibration signals along the X, Y, and Z axes. During installation, the three-axis directions are aligned with the orthogonal direction of the equipment coordinate system or mounting base to ensure consistent axial definitions between different locations and different samples, facilitating subsequent matrix generation and channel combination by axis. The signals at each monitoring location are kept synchronized or aligned on the time reference to ensure comparability during subsequent processing. The purpose of triaxial acquisition is to simultaneously record structural responses in different directions, thereby covering directional features that may be generated by gear meshing, shaft imbalance, bearing defects, and electromagnetic excitation, and improving the probability of capturing early weak anomalies. Selecting at least one monitoring location allows for flexible deployment based on site space and wiring conditions, while selecting multiple locations can improve coverage and diagnostic robustness for different components. Through the above deployment and acquisition method, the obtained multi-axis vibration signals have consistency in coordinate definition and time reference, satisfying the requirements for subsequent frequency domain feature extraction. The input requirements for generating multi-channel spectral images by combining feature point sampling, two-dimensional matrix filling, and channel mapping are beneficial for accurately corresponding cross-axis information in the same image coordinates.

[0023] For further details, please refer to Figure 4 , Figure 4 This is a schematic diagram of a sliding window for the wind power equipment fault identification method based on time-series vibration data analysis provided by the present invention; S2 specifically refers to: S21: The multi-axis vibration signal is processed by time-domain segmentation using a sliding window, dividing the continuous time series into multiple adjacent and at least partially overlapping time segments, and using each time segment as a time-series feature unit, with the window length and step size being preset parameters. S22: Perform a fast Fourier transform on the vibration signal of each axis corresponding to each of the time-series feature units, calculate its amplitude spectrum, and use the amplitude spectrum as the frequency domain feature data corresponding to each axis.

[0024] In this embodiment, the multi-axis vibration signal is first segmented in the time domain using a sliding window. The continuous time series is divided into multiple adjacent and at least partially overlapping time segments according to a preset window length and step size, and each time segment is used as a temporal feature unit. This ensures the continuity of information between adjacent segments, reduces feature loss caused by segment boundaries, and allows subsequent processing to be carried out in fixed-scale minimum processing units, facilitating batch parallel processing and matching the time scale of equipment operating condition changes. A smaller step size results in greater overlap between continuous windows, thereby increasing the density of training samples and enhancing the preservation of temporal continuity. Conversely, a larger step size can reduce overlap and computational burden, but may ignore subtle transition features. Each segmented subsequence is treated as an independent input sample, thereby improving the model's ability to learn sequence patterns and distinguish between normal and fault operating states. This strategy ensures that the basic temporal features of the original signal are preserved throughout the feature learning process. The characteristic of the time-domain vibration signal is the temporal variation of amplitude. Early faults in wind turbines are often difficult to identify in the time domain, but can be effectively identified through frequency domain analysis. To this end, a fast Fourier transform is performed on the vibration signal of each axis within each time-series feature unit to obtain the amplitude spectrum indexed by frequency and valued by magnitude, which is then used as the frequency domain feature data corresponding to each axis. This frequency domain representation can make the time-domain energy distribution explicit as the difference in strength of different frequency components, which is beneficial to highlighting the periodic and harmonic characteristics related to rotating parts, thereby improving the discriminability of subsequent feature point selection and matrix construction.

[0025] For further details, please refer to Figure 5 , Figure 5 This diagram illustrates the comparison of normal and fault samples within a single time window for the wind power equipment fault identification method based on time-series vibration data analysis provided by the present invention. When there are insufficient fault samples for training the classification model, Gaussian-distributed perturbations are injected into the time-series feature units in the time domain to generate simulated fault samples. The mean and standard deviation of the perturbations are determined according to the scaling factors of the mean and standard deviation of the corresponding time-series feature units, respectively. The simulated fault samples are then combined with existing samples for training the classification model.

[0026] In this embodiment, to expand the training set and maintain the consistency of amplitude scale for each channel when there are insufficient fault samples, Gaussian perturbations are injected into each time-series feature unit in the time domain according to the channel to generate simulated fault samples. The mathematical expression of this can be written in the following form:

[0027] In the formula, The mean is Standard deviation is The Gaussian random variable. The mean and standard deviation are determined by the statistics of that time-series characteristic unit:

[0028] in, and These represent the mean and standard deviation of the corresponding channel within the time-series feature unit, respectively. and This is a preset scaling factor used to adjust the strength of the simulated fault. When and When the value is close to zero, the result is Approaching normal signal; and As the sample size increases, the simulated signal gradually exhibits more significant abnormal characteristics. The generated simulated fault samples are merged with existing samples for classification model training or calibration. This can alleviate overfitting and class imbalance problems caused by the scarcity of fault samples without changing the sampling and labeling system. Figure 5 For the present invention A comparison between fault samples and normal samples obtained from time simulation.

[0029] Furthermore, step S4 specifically involves: sorting the feature points in ascending order of frequency, and then filling the amplitude values ​​of the sorted feature points sequentially using row-major order. A two-dimensional matrix is ​​generated, and two-dimensional matrices generated on different axes correspond to the same frequency index at the same matrix position.

[0030] In this embodiment, the feature points of each axis obtained in step S3 are first stably sorted according to frequency from low to high, and then the corresponding amplitudes are sequentially filled with a size of row priority. A two-dimensional matrix, specifically mapped according to index: the... The nth feature point is filled into the matrix. Okay, number The matrix is ​​arranged so that the frequency increases monotonically in the horizontal direction and continues sequentially from the end of the top row to the beginning of the bottom row in the vertical direction. For multi-axis data, the above sorting and filling are performed with the same starting frequency and sampling step size, so that the two-dimensional matrices generated on different axes correspond to the same frequency index at the same row and column positions. Through this ordered filling and cross-axis index alignment, on the one hand, the one-dimensional frequency domain data is structured into a two-dimensional representation with local adjacency relationships, which facilitates subsequent image-based feature learning to capture adjacent frequency band patterns; on the other hand, it ensures the comparability of each axis at pixel positions, providing a consistent coordinate benchmark for channel combination to generate multi-channel spectral images, thereby reducing fusion ambiguity and improving recognition stability.

[0031] Furthermore, S5 specifically includes: S51: Normalize the element values ​​of the two-dimensional matrix corresponding to each axis to obtain values ​​that are... A two-dimensional matrix; S52: Map the normalized two-dimensional matrix to... The range of pixel values; S53: Combine the two-dimensional matrices corresponding to each axis according to the one-to-one channel mapping rule to generate the multi-channel spectrum image, and the number of channels is consistent with the number of axes, and each channel corresponds to the same frequency index at the same pixel position.

[0032] In this embodiment, the two-dimensional matrix corresponding to each axis is first normalized axis by axis, and the matrix elements are linearly compressed according to the maximum-minimum criterion. The interval is used to eliminate differences in dimensions and magnitudes across different axes and to maintain the relative strength of each frequency component; then the normalization result is transformed using a linear mapping to... The pixel value range is used to obtain a single-channel grayscale matrix that is easy to store and render. Based on this, the grayscale matrices of each axis are combined according to a one-to-one channel mapping rule, with the number of channels matching the number of axes. Following the row-column-frequency correspondence established in step S4, each channel corresponds to the same frequency index at the same pixel position, thus forming a matrix of size [missing information]. ( The above processing aims to achieve unified calibration and pixel-level alignment of cross-axis data, avoiding bias caused by the dominance of single-axis amplitude values, and facilitating subsequent classification models to jointly perceive multi-axis frequency domain information under the same spatial coordinates, thereby improving the stability and discrimination accuracy of fault identification.

[0033] Furthermore, when the multi-axis vibration signal includes a three-axis time series signal along the X-axis, Y-axis, and Z-axis, the two-dimensional matrices corresponding to the X-axis, Y-axis, and Z-axis are respectively assigned to the red, green, and blue channels of the image.

[0034] In this embodiment, when the multi-axis vibration signal is a three-axis time series signal along the X, Y, and Z axes, the channel mapping adopts a fixed color binding rule: the two-dimensional matrix corresponding to the X-axis is assigned to the red channel (R) of the image, the two-dimensional matrix corresponding to the Y-axis is assigned to the green channel (G) of the image, and the two-dimensional matrix corresponding to the Z-axis is assigned to the blue channel (B) of the image, forming a matrix of size [missing information]. RGB image; specifically at the pixel level, let ,in It is a three-axis two-dimensional matrix. The output image is generated using the one-to-one correspondence between rows, columns, and frequency indices established in step S4. This ensures that the three channels correspond to the same frequency components at the same pixel position, thereby synchronously expressing the three-axis frequency domain information in the same coordinate system. This facilitates the subsequent classification model to perceive cross-axis differences and correlation patterns at once, and also makes it easier to store, visualize, and load using a general image interface.

[0035] To help those skilled in the art understand, please refer to Figure 6-7 , Figure 6 An RGB image of normal data for a time window. Figure 7 The RGB image of the fault data is a time window. The RGB image of the fault data is significantly different from the RGB image of the normal data, which helps the subsequent neural network to distinguish the fault data image, thereby identifying faults in wind power equipment, further demonstrating the effectiveness of the method of the present invention.

[0036] Furthermore, S6 specifically involves: inputting the multi-channel spectral image into a classification model based on a deep neural network to output a determination of the operating status of the wind power equipment. The classification model is preferably a convolutional neural network, which includes at least one convolutional layer, a pooling layer, and a fully connected layer. The output layer uses the Softmax function to calculate the class probability.

[0037] In this embodiment, the multi-channel spectral image obtained in step S5 is used as an input tensor and fed into the classification model. The image size is... The channel order corresponds one-to-one with each axis and maintains consistency with the frequency index at the pixel position. The classification model is based on a deep neural network, preferably a convolutional neural network. The network includes at least one convolutional layer for extracting local spectral texture and cross-axis correlation features, a pooling layer for compressing feature maps and improving robustness to local perturbations, and a fully connected layer for global discrimination in a high-dimensional feature space. The output layer uses the Softmax function to calculate the probability distribution of the device being in normal or faulty categories. The model can be trained using labeled training samples to determine network parameters. During inference, it provides a category probability for a single input image and outputs the operating status judgment accordingly. This achieves automatic learning and stable classification of multi-axis frequency domain information under a unified image representation, thereby improving the accuracy of fault identification and the usability of online applications.

[0038] Please see Figure 8 The present invention also provides a wind power equipment fault identification system based on time-series vibration data analysis, comprising: The signal acquisition module is used to acquire multi-axis vibration signals of wind power equipment during operation; The frequency domain feature extraction module is used to extract frequency domain features from the multi-axis vibration signal to obtain frequency domain feature data corresponding to each axis. The feature point sampling module is used to sample from the frequency domain feature data of each axis. There are 1 feature points; among them, This is the default value; A two-dimensional matrix generation module is used to fill the corresponding amplitudes of the feature points into a two-dimensional matrix according to the frequency sequence. The channel mapping combination module is used to combine the two-dimensional matrices corresponding to each axis according to the preset channel mapping rules to generate a multi-channel spectral image. The status recognition module is used to input the multi-channel spectrum image into a classification model to identify and classify the operating status of wind power equipment.

[0039] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0040] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for fault identification of wind power equipment based on time-series vibration data analysis, characterized in that, The method includes: S1: Acquire multi-axis vibration signals of wind power equipment during operation; S2: Extract frequency domain features from the multi-axis vibration signal to obtain frequency domain feature data corresponding to each axis; S3: Sample from the frequency domain feature data of each axis There are 1 feature points; among them, This is the default value; S4: Fill the corresponding amplitudes of the feature points into a two-dimensional matrix according to the frequency sequence; S5: Combine the two-dimensional matrices corresponding to each axis according to the preset channel mapping rules to generate a multi-channel spectrum image; S6: Based on the multi-channel spectrum image, input the image into the classification model to identify and classify the operating status of the wind power equipment.

2. The wind power equipment fault identification method based on time-series vibration data analysis according to claim 1, characterized in that: The multi-axis vibration signal is collected by a sensor installed at at least one monitoring position in the gearbox, main shaft, or generator body, and includes triaxial time-series vibration signals along the X-axis, Y-axis, and Z-axis directions.

3. The wind power equipment fault identification method based on time-series vibration data analysis according to claim 1, characterized in that, Specifically, S2 is: S21: The multi-axis vibration signal is processed by time-domain segmentation using a sliding window, dividing the continuous time series into multiple adjacent and at least partially overlapping time segments, and using each time segment as a time-series feature unit, with the window length and step size being preset parameters. S22: Perform a fast Fourier transform on the vibration signal of each axis corresponding to each of the time-series feature units, calculate its amplitude spectrum, and use the amplitude spectrum as the frequency domain feature data corresponding to each axis.

4. The wind power equipment fault identification method based on time-series vibration data analysis according to claim 3, characterized in that: When there are insufficient fault samples for training the classification model, a Gaussian distribution perturbation is injected into the temporal feature unit in the time domain by channel to generate simulated fault samples. The mean and standard deviation of the perturbation are determined by the scaling factor of the mean and standard deviation of the corresponding temporal feature unit, respectively. The simulated fault samples are then combined with existing samples for training the classification model.

5. The wind power equipment fault identification method based on time-series vibration data analysis according to claim 1, characterized in that, S4 specifically involves: sorting the feature points in ascending order of frequency, and then filling the amplitude values ​​of the sorted feature points sequentially using row-major order. A two-dimensional matrix is ​​generated, and two-dimensional matrices generated on different axes correspond to the same frequency index at the same matrix position.

6. The wind power equipment fault identification method based on time-series vibration data analysis according to claim 1, characterized in that, Specifically, S5 is: S51: Normalize the element values ​​of the two-dimensional matrix corresponding to each axis to obtain values ​​that are... A two-dimensional matrix; S52: Map the normalized two-dimensional matrix to... The range of pixel values; S53: Combine the two-dimensional matrices corresponding to each axis according to the one-to-one channel mapping rule to generate the multi-channel spectrum image, and the number of channels is consistent with the number of axes, and each channel corresponds to the same frequency index at the same pixel position.

7. The wind power equipment fault identification method based on time-series vibration data analysis according to claim 6, characterized in that: When the multi-axis vibration signal includes a three-axis time series signal along the X-axis, Y-axis, and Z-axis, the two-dimensional matrices corresponding to the X-axis, Y-axis, and Z-axis are respectively assigned to the red, green, and blue channels of the image.

8. The wind power equipment fault identification method based on time-series vibration data analysis according to claim 1, characterized in that, S6 specifically involves inputting the multi-channel spectral image into a classification model based on a deep neural network to output a determination of the operating status of the wind power equipment. The classification model is preferably a convolutional neural network, which includes at least one convolutional layer, a pooling layer, and a fully connected layer. The output layer uses the Softmax function to calculate the class probability.

9. A wind power equipment fault identification system based on time-series vibration data analysis, characterized in that, include: The signal acquisition module is used to acquire multi-axis vibration signals of wind power equipment during operation; The frequency domain feature extraction module is used to extract frequency domain features from the multi-axis vibration signal to obtain frequency domain feature data corresponding to each axis. The feature point sampling module is used to sample from the frequency domain feature data of each axis. There are 1 feature points; among them, This is the default value; A two-dimensional matrix generation module is used to fill the corresponding amplitudes of the feature points into a two-dimensional matrix according to the frequency sequence. The channel mapping combination module is used to combine the two-dimensional matrices corresponding to each axis according to the preset channel mapping rules to generate a multi-channel spectral image. The status recognition module is used to input the multi-channel spectrum image into a classification model to identify and classify the operating status of wind power equipment.

Images