Data-driven offshore wind turbine fault diagnosis system and diagnosis method thereof

Abstract

The application relates to the field of intelligent detection of offshore faults, and particularly discloses a data-driven offshore wind turbine fault diagnosis system and a diagnosis method thereof. The system uses an artificial intelligence detection technology, uses a deep neural network model as a feature extractor, and uses a correlation model between the difference between vibration signals at the two ends of a generator and the rotating speed of the generator to perform offshore wind turbine fault diagnosis detection. In this way, the timeliness and effectiveness of offshore wind turbine fault maintenance can be ensured, and normal operation of the offshore wind turbine can be ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent detection of offshore wind turbine faults, and more particularly, to a data-driven offshore wind turbine fault diagnosis system and a diagnosis method thereof. BACKGROUND

[0002] In recent years, the energy industry transformation process in China has been accelerating, and China has become the largest market in the world in terms of wind power generation scale and growth. As a renewable energy source, wind energy has the characteristics of being green and clean, and is the research focus of countries around the world. Compared with onshore wind power, offshore wind resources are more abundant, have large power generation capacity, long power generation time, little impact on the living environment, do not occupy arable land, and can be developed on a large scale, and therefore, developing offshore wind power generation technology has become a new trend in the development of the wind power industry.

[0003] In the harsh environment of offshore wind farms, maintenance personnel cannot arrive at the fault point in time to carry out maintenance, and therefore, offshore wind power monitoring systems based on wireless communication are increasingly important. For example, in the daily monitoring of offshore wind turbines, the following problems are often encountered: how to confirm whether the wind turbine is operating normally? How can the abnormal point of a wind turbine be quickly located? How can real-time information and historical working information of each wind turbine be obtained?

[0004] Therefore, an optimized offshore wind turbine fault diagnosis system is expected to accurately detect and diagnose offshore wind turbine faults to ensure the normal operation of offshore wind turbines. SUMMARY

[0005] To solve the above technical problems, the present application is proposed. The embodiments of the present application provide a data-driven offshore wind turbine fault diagnosis system and a diagnosis method thereof, which uses a deep neural network model as a feature extractor by adopting artificial intelligence detection technology, and uses a correlation model between the difference of vibration signals at the head and tail ends of a generator and the rotational speed thereof to perform offshore wind turbine fault diagnosis and detection. In this way, the timeliness and effectiveness of offshore wind turbine fault maintenance can be ensured to ensure the normal operation of offshore wind turbines.

[0006] According to an aspect of the present application, a data-driven offshore wind turbine fault diagnosis system is provided, which comprises:

[0007] a vibration data acquisition module configured to acquire first and second vibration signals within a predetermined time period collected by first and second vibration sensors disposed at the head and tail ends of a generator;

[0008] a domain conversion module configured to convert the first and second vibration signals into first and second vibration gram angle and field images using the gram angle field principle.

[0009] a spatial attention encoding module, configured to obtain a first vibration feature matrix and a second vibration feature matrix by using a first convolutional neural network with a spatial attention mechanism respectively on the first and second vibration gram angles and field images;

[0010] a vibration transfer module, configured to calculate a vibration transfer matrix of the first vibration feature matrix relative to the second vibration feature matrix;

[0011] a feature value correction module, configured to correct feature values at respective positions in the vibration transfer matrix to obtain a corrected vibration transfer matrix;

[0012] a rotation speed data acquisition module, configured to acquire rotation speed values of the generator at a plurality of predetermined time points in a predetermined time period collected by a transfer measuring instrument;

[0013] a rotation speed data encoding module, configured to obtain a rotation speed feature vector by using a time series encoder containing a one-dimensional convolutional layer on the rotation speed values at the plurality of predetermined time points in the predetermined time period;

[0014] a feature fusion module, configured to fuse the rotation speed feature vector and the corrected vibration transfer matrix to obtain a classification feature vector; and

[0015] a fault diagnosis result generation module, configured to obtain a classification result by using a classifier on the classification feature vector, the classification result being used to indicate whether the offshore wind turbine has a fault.

[0016] In the above data-driven offshore wind turbine fault diagnosis system, the spatial attention encoding module is further configured to: in a forward transmission process of each layer of the first convolutional neural network, perform the following operations on input data respectively: perform convolution processing on the input data to generate a convolution feature map; perform pooling processing on the convolution feature map to generate a pooling feature map; perform nonlinear activation on the pooling feature map to generate an activation feature map; calculate a mean value of each position along a channel dimension of the activation feature map to generate a spatial feature matrix; calculate a Softmax function value of each position in the spatial feature matrix to obtain a spatial score matrix; and calculate a point-by-point multiplication of the spatial feature matrix and the spatial score matrix to obtain a feature matrix; wherein the feature matrix output by a last layer of the first convolutional neural network is the first vibration feature matrix and the second vibration feature matrix.

[0017] In the above data-driven offshore wind turbine fault diagnosis system, the vibration transfer module is further configured to calculate the vibration transfer matrix of the first vibration feature matrix relative to the second vibration feature matrix according to the following formula:

[0018] wherein the formula is:

[0019] S = T * F

[0020] wherein F represents the first vibration feature matrix, T represents the vibration transfer matrix, and S represents the second vibration feature matrix.

[0021] In the above data-driven offshore wind turbine fault diagnosis system, the feature value correction module comprises: a self-correlation coding unit configured to calculate the product between the vibration transfer matrix and the transpose of the vibration transfer matrix to obtain a self-correlation feature matrix; a feature value compression unit configured to take the square root of the feature values at each position in the self-correlation feature matrix to obtain a self-correlation compressed feature matrix, and divide the feature values at each position in the self-correlation compressed feature matrix by the product of the number of rows and the number of columns of the self-correlation compressed feature matrix to obtain a dimension-compressed self-correlation compressed feature matrix; a first convolution unit configured to perform convolution coding on the vibration transfer matrix to obtain a first convolution vibration transfer matrix; a fusion unit configured to calculate the positional sum of the first convolution vibration transfer matrix and the dimension-compressed self-correlation compressed feature matrix to obtain a fusion feature matrix; a second convolution unit configured to perform convolution coding on the fusion feature matrix to obtain a multi-scale feature matrix; a position information coding unit configured to convert the two-dimensional position coordinates of each position in the coordinate tensor of the vibration transfer matrix into one-dimensional numerical values to obtain a position feature matrix; and a re-fusion unit configured to calculate the positional dot product of the position feature matrix and the multi-scale feature matrix to obtain the corrected vibration transfer matrix.

[0022] In the above data-driven offshore wind turbine fault diagnosis system, the feature value compression unit is further configured to: take the square root of the feature values at each position in the self-correlation feature matrix to obtain the self-correlation compressed feature matrix according to the following formula, and divide the feature values at each position in the self-correlation compressed feature matrix by the product of the number of rows and the number of columns of the self-correlation compressed feature matrix to obtain the dimension-compressed self-correlation compressed feature matrix.

[0023] wherein the formula is:

[0024]

[0025] wherein M represents the vibration transfer matrix, M1 represents the dimension-compressed self-correlation compressed feature matrix, represents matrix multiplication, represents taking the square root of the value at each position of the matrix, and n is the product of the number of rows and the number of columns of the matrix.

[0026] In the data-driven offshore wind turbine fault diagnosis system, the position information encoding unit is further configured to convert two-dimensional position coordinates of each position in the coordinate tensor of the vibration transfer matrix into one-dimensional numerical values to obtain the position feature matrix according to the following formula:

[0027] The formula is:

[0028] M2 = Φ (P M )

[0029] wherein M represents the vibration transfer matrix, M2 represents the position feature matrix, for mapping two-dimensional position coordinates into one-dimensional numerical values, P M represents the (x, y) coordinate tensor of the matrix M.

[0030] In the data-driven offshore wind turbine fault diagnosis system, the rotation speed data encoding module is further configured to arrange rotation speed values at a plurality of predetermined time points in a predetermined time period into an input vector in time dimension, perform full connection encoding on the input vector using a full connection layer of the time sequence encoder according to the following formula to extract high-dimensional hidden features of feature values at each position in the input vector, wherein the formula is: wherein X is the input vector, Y is the output vector, W is the weight matrix, B is the bias vector, represents matrix multiplication; and perform one-dimensional convolution encoding on the input vector using a one-dimensional convolution layer of the time sequence encoder according to the following formula to extract high-dimensional hidden correlation features between feature values at each position in the input vector, wherein the formula is:

[0031]

[0032] wherein a is the width of the convolution kernel in the x direction, F is the convolution kernel parameter vector, G is the local vector matrix for convolution kernel function operation, w is the size of the convolution kernel, and X represents the input vector.

[0033] In the data-driven offshore wind turbine fault diagnosis system, the feature fusion module is further configured to fuse the rotation speed feature vector and the corrected vibration transfer matrix according to the following formula to obtain the classification feature vector:

[0034] The formula is:

[0035]

[0036] wherein V represents the rotation speed feature vector, M' represents the corrected vibration transfer matrix, V' represents the classification feature vector, represents matrix multiplication.

[0037] In the aforementioned data-driven offshore wind turbine fault diagnosis system, the fault diagnosis result generation module is further configured to: process the classification feature vector using the classifier according to the following formula to obtain the classification result, wherein the formula is: softmax{(W n B n ):…:(W1,B1)|X}, where W1 to W n The weight matrix is ​​B1 to B1. n X is the bias vector, and X is the classification feature vector.

[0038] According to another aspect of this application, a diagnostic method for a data-driven offshore wind turbine fault diagnosis system includes:

[0039] Acquire first and second vibration signals within a predetermined time period by first and second vibration sensors deployed at the head and tail ends of the generator.

[0040] The first vibration signal and the second vibration signal are converted into a first vibration Gram angle and field image and a second vibration Gram angle and field image using the Gram angle field principle.

[0041] The first and second vibration Gram angles and field images are respectively obtained by using a first convolutional neural network with a spatial attention mechanism to obtain a first vibration feature matrix and a second vibration feature matrix.

[0042] Calculate the vibration transfer matrix of the first vibration feature matrix relative to the second vibration feature matrix;

[0043] The eigenvalues ​​at each position in the vibration transfer matrix are corrected to obtain the corrected vibration transfer matrix.

[0044] The rotational speed values ​​of the generator at multiple predetermined time points within the predetermined time period are acquired by the transfer measuring instrument;

[0045] The rotational speed values ​​at multiple predetermined time points within the predetermined time period are passed through a temporal encoder containing a one-dimensional convolutional layer to obtain a rotational speed feature vector.

[0046] The rotational speed feature vector and the corrected vibration transfer matrix are fused to obtain a classification feature vector; and

[0047] The classification feature vector is passed through a classifier to obtain a classification result, which is used to indicate whether the offshore wind turbine has a fault.

[0048] In the diagnostic method of the above-mentioned data-driven offshore wind turbine fault diagnosis system, the first and second vibration Gram angles and field images are respectively obtained by using a first convolutional neural network with a spatial attention mechanism to obtain a first vibration feature matrix and a second vibration feature matrix. This includes: each layer of the first convolutional neural network, during the forward propagation of the layer, performing the following on the input data: convolution processing to generate a convolutional feature map; pooling processing to generate a pooled feature map; nonlinear activation to generate an activation feature map; calculating the mean value along the channel dimension of each position of the activation feature map to generate a spatial feature matrix; calculating the Softmax-like function value at each position in the spatial feature matrix to obtain a spatial score matrix; and calculating the positional dot product of the spatial feature matrix and the spatial score map to obtain a feature matrix; wherein the feature matrix output by the last layer of the first convolutional neural network is the first vibration feature matrix and the second vibration feature matrix.

[0049] In the above-mentioned diagnostic method for a data-driven offshore wind turbine fault diagnosis system, calculating the vibration transfer matrix of the first vibration feature matrix relative to the second vibration feature matrix includes: calculating the vibration transfer matrix of the first vibration feature matrix relative to the second vibration feature matrix using the following formula;

[0050] The formula is as follows:

[0051] S = T * F

[0052] Where F represents the first vibration feature matrix, T represents the vibration transfer matrix, and S represents the second vibration feature matrix.

[0053] In the diagnostic method of the above-mentioned data-driven offshore wind turbine fault diagnosis system, the correction of the eigenvalues ​​at each position in the vibration transfer matrix to obtain the corrected vibration transfer matrix includes: calculating the product between the vibration transfer matrix and the transpose of the vibration transfer matrix to obtain an autocorrelation feature matrix; taking the square root of the eigenvalues ​​at each position in the autocorrelation feature matrix to obtain an autocorrelation compressed feature matrix, and dividing the eigenvalues ​​at each position in the autocorrelation compressed feature matrix by the product of the number of rows and columns of the autocorrelation compressed feature matrix to obtain a dimension-compressed autocorrelation compressed feature matrix; performing convolutional encoding on the vibration transfer matrix to obtain a first convolutional vibration transfer matrix; calculating the positional sum of the first convolutional vibration transfer matrix and the dimension-compressed autocorrelation compressed feature matrix to obtain a fused feature matrix; performing convolutional encoding on the fused feature matrix to obtain a multi-scale feature matrix; converting the two-dimensional position coordinates of each position in the coordinate tensor of the vibration transfer matrix into one-dimensional values ​​to obtain a position feature matrix; and calculating the positional dot product of the position feature matrix and the multi-scale feature matrix to obtain the corrected vibration transfer matrix.

[0054] In the diagnostic method of the above-mentioned data-driven offshore wind turbine fault diagnosis system, the eigenvalues ​​at each position in the autocorrelation feature matrix are squared to obtain the autocorrelation compressed feature matrix, and the eigenvalues ​​at each position in the autocorrelation compressed feature matrix are divided by the product of the number of rows and columns of the autocorrelation compressed feature matrix to obtain the dimension-compressed autocorrelation compressed feature matrix. This includes: taking the square root of the eigenvalues ​​at each position in the autocorrelation feature matrix using the following formula to obtain the autocorrelation compressed feature matrix, and dividing the eigenvalues ​​at each position in the autocorrelation compressed feature matrix by the product of the number of rows and columns of the autocorrelation compressed feature matrix to obtain the dimension-compressed autocorrelation compressed feature matrix.

[0055] The formula is as follows:

[0056]

[0057] Where M represents the vibration transfer matrix, and M1 represents the dimension-compressed self-correlated compressed feature matrix. Represents matrix multiplication. This represents the square root of the value at each position in the matrix, where n is the number of rows multiplied by the number of columns in the matrix.

[0058] In the above-mentioned diagnostic method of the data-driven offshore wind turbine fault diagnosis system, the two-dimensional position coordinates of each position in the coordinate tensor of the vibration transfer matrix are converted into one-dimensional values ​​to obtain the position feature matrix, including: converting the two-dimensional position coordinates of each position in the coordinate tensor of the vibration transfer matrix into one-dimensional values ​​to obtain the position feature matrix using the following formula.

[0059] The formula is as follows:

[0060] M2=Φ(P M )

[0061] Where M represents the vibration transfer matrix, and M2 represents the position feature matrix. Used to map two-dimensional position coordinates to one-dimensional numerical values, P M Let (x,y) represent the (x,y) coordinate tensor of matrix M.

[0062] In the diagnostic method of the above-mentioned data-driven offshore wind turbine fault diagnosis system, the rotational speed values ​​at multiple predetermined time points within the predetermined time period are processed by a time-series encoder containing a one-dimensional convolutional layer to obtain a rotational speed feature vector. This includes: arranging the rotational speed values ​​at multiple predetermined time points within the predetermined time period into a one-dimensional input vector according to the time dimension; and using the fully connected layer of the time-series encoder to perform fully connected encoding on the input vector using the following formula to extract high-dimensional latent features of the feature values ​​at each position in the input vector, wherein the formula is: Where X is the input vector, Y is the output vector, W is the weight matrix, and B is the bias vector. This represents matrix multiplication; the one-dimensional convolutional layer of the temporal encoder is used to perform one-dimensional convolutional encoding on the input vector using the following formula to extract high-dimensional implicit correlation features between feature values ​​at various positions in the input vector, wherein the formula is:

[0063]

[0064] Where a is the width of the convolution kernel in the x-direction, F is the convolution kernel parameter vector, G is the local vector matrix operated on by the convolution kernel function, w is the size of the convolution kernel, and X represents the input vector.

[0065] In the diagnostic method of the above-mentioned data-driven offshore wind turbine fault diagnosis system, the fusion of the rotational speed feature vector and the corrected vibration transfer matrix to obtain the classification feature vector includes: fusing the rotational speed feature vector and the corrected vibration transfer matrix to obtain the classification feature vector using the following formula;

[0066] The formula is as follows:

[0067]

[0068] Where V represents the rotational speed feature vector, M' represents the corrected vibration transfer matrix, and V' represents the classification feature vector. This represents matrix multiplication.

[0069] In the diagnostic method of the above-mentioned data-driven offshore wind turbine fault diagnosis system, the classification feature vector is processed by a classifier to obtain a classification result, which is used to indicate whether the offshore wind turbine has a fault. This includes: processing the classification feature vector using the classifier with the following formula to obtain the classification result, wherein the formula is: softmax{(W n B n ):…:(W1,B1)|X}, where W1 to W n The weight matrix is ​​B1 to B1. n X is the bias vector, and X is the classification feature vector.

[0070] Compared with existing technologies, the data-driven offshore wind turbine fault diagnosis system and method provided in this application employ artificial intelligence detection technology, using a deep neural network model as a feature extractor. It utilizes a correlation model between the difference in vibration signals at the generator's beginning and end and its rotational speed to perform fault diagnosis and detection of the offshore wind turbine. This ensures the timeliness and effectiveness of fault maintenance for the offshore wind turbine, thereby guaranteeing its normal operation. Attached Figure Description

[0071] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0072] Figure 1 This is an application scenario diagram of a data-driven offshore wind turbine fault diagnosis system according to an embodiment of this application.

[0073] Figure 2 This is a block diagram of a data-driven offshore wind turbine fault diagnosis system according to an embodiment of this application.

[0074] Figure 3 This is a block diagram of the eigenvalue correction module in a data-driven offshore wind turbine fault diagnosis system according to an embodiment of this application.

[0075] Figure 4 This is a flowchart of a diagnostic method for a data-driven offshore wind turbine fault diagnosis system according to an embodiment of this application.

[0076] Figure 5 This is a schematic diagram of the architecture of a diagnostic method for a data-driven offshore wind turbine fault diagnosis system according to an embodiment of this application. Detailed Implementation

[0077] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0078] Scenario overview

[0079] As mentioned earlier, in recent years, my country's energy industry transformation has accelerated, and my country has become the world's largest and fastest-growing market for wind power. Wind energy, as a renewable energy source, is green and clean, making it a key research focus for countries worldwide. Compared to onshore wind power, offshore wind resources are more abundant, offering numerous advantages such as higher power generation capacity, longer operating hours, less impact on the environment, no occupation of arable land, and the potential for large-scale development. Therefore, developing offshore wind power technology has become a new trend in the wind power industry.

[0080] In the harsh environment of offshore wind farms, where maintenance personnel cannot reach the fault location in a timely manner for repairs, wireless communication-based offshore wind power monitoring systems become increasingly important. For example, in the daily monitoring of offshore wind turbines, the following problems are frequently encountered: How to confirm whether the turbine is operating normally? How to quickly locate abnormal conditions in a particular turbine? How to obtain real-time and historical operating information for each turbine?

[0081] Therefore, there is a need for an optimized offshore wind turbine fault diagnosis system to accurately detect and diagnose offshore wind turbine faults in order to ensure the normal operation of offshore wind turbines.

[0082] Currently, deep learning and neural networks have been widely applied in fields such as computer vision, natural language processing, and speech signal processing. Furthermore, deep learning and neural networks have demonstrated near-human or even surpassed human-level performance in areas such as image classification, object detection, semantic segmentation, and text translation.

[0083] The development of deep learning and neural networks has provided new solutions and approaches for fault diagnosis of offshore wind turbines.

[0084] Accordingly, the inventors of this application have discovered that during the operation of offshore wind turbines in actual scenarios, under normal conditions, there is a certain nonlinear correlation between the difference in vibration signals at the beginning and end of the generator and its rotational speed. Therefore, the inventors of this application intend to use this correlation information to perform fault diagnosis and detection of the offshore wind turbine in order to ensure the normal operation of the offshore wind turbine.

[0085] Specifically, in the technical solution of this application, firstly, first and second vibration signals within a predetermined time period are collected by first and second vibration sensors deployed at the beginning and end of the generator. Then, the first and second vibration signals are converted into first and second vibration Gram angular field images and second vibration Gram angular field images using the Gram principle. It should be understood that the Gram angular field (GAF) is based on the Gram principle, which can transfer time series from the classical Cartesian coordinate system to the polar coordinate system for representation. GAF can well preserve the dependence and correlation of the original vibration time series signal, possessing similar time series characteristics to the original vibration signal. Furthermore, since GAF can yield Gram angular sum field (GASF) and Gram angular difference field (GADF) depending on the trigonometric functions used for encoding, and GADF conversion is irreversible, the technical solution of this application selects the inversely convertible GASF conversion method for encoding the vibration signal. Specifically, in the embodiment, the encoding steps from the vibration signal to the GASF image are as follows: For a time series with C dimensions = {Q1, Q2, ..., Q...} C}, where each dimension contains n sampling points Q i ={q i1 q i2 , ..., q in First, the data for each dimension is normalized. Then, all values ​​in the data are integrated into the range [-1, 1]. After integration, the normalized values ​​are replaced with trigonometric function values ​​(Cos values), and polar coordinates are used instead of Cartesian coordinates, thus preserving the absolute temporal relationship of the sequence.

[0086] Considering that, in actual situations, the vibration signals detected by the vibration sensors located at the front and rear ends of the engine have specific implicit correlation information in space, a first convolutional neural network model with a spatial attention mechanism is used to perform feature mining on the first and second vibration gram angle and field images to extract high-dimensional local implicit feature distribution information with spatial location correlation features in the first and second vibration gram angle and field images.

[0087] It should be understood that, under normal conditions, the difference in vibration signals between the beginning and end of the generator exhibits a certain pattern of nonlinear correlation with its rotational speed. Therefore, in the technical solution of this application, this nonlinear correlation can be represented based on a transfer matrix; that is, specifically, the vibration transfer matrix of the first vibration characteristic matrix relative to the second vibration characteristic matrix is ​​calculated.

[0088] However, due to the spatial attention mechanism of the first convolutional neural network, the first vibration feature matrix and the second vibration feature matrix can enhance the extraction of local semantics, but at the same time weaken the correlation between local semantics, resulting in poor global semantic expression. The vibration transfer matrix also suffers from the same problem. Therefore, in the technical solution of this application, it is preferable to optimize the local semantic correlation expression ability and even the global expression ability of the vibration transfer matrix, that is:

[0089]

[0090] Both Cov1() and Cov2() are single convolutional layers. Used to map two-dimensional position coordinates to one-dimensional numerical values, P M Let ⊙ represent the (x,y) coordinate tensor of matrix M. These represent matrix dot multiplication and dot addition, respectively. Represents matrix multiplication. This represents the square root of the value at each position in the matrix, where n is the number of rows multiplied by the number of columns in the matrix.

[0091] Here, the optimization employs a location-proposal local reasoning transformation to comprehensively fuse the captured local semantics and further derived global semantics. Specifically, it uses location information as a proposal to reason about global scene semantics through a local perceptive field. Furthermore, by adjusting the parameters of the convolutional layers Cov1() and Cov2(), a balance can be achieved between local-global information transfer and global information reasoning prediction, thereby improving the classification accuracy of the classification feature vector by enhancing the class prediction capability of the vibrational transfer matrix.

[0092] Furthermore, to detect and diagnose faults in the offshore wind turbine, it is necessary to collect the generator's rotational speed values ​​at multiple predetermined time points within the predetermined time period using a transfer measurement instrument. Considering the dynamic variation of the rotational speed values ​​over time, a timing encoder is used to encode the rotational speed values ​​at multiple predetermined time points within the predetermined time period in order to fully extract this variation. Accordingly, in a specific example, the timing encoder consists of alternating fully connected layers and one-dimensional convolutional layers. It extracts the temporal correlation of the generator's rotational speed values ​​through one-dimensional convolutional encoding and extracts the high-dimensional implicit features of the generator's rotational speed values ​​through fully connected encoding.

[0093] In this way, the rotational speed feature vector and the corrected vibration transfer matrix are fused by matrix multiplication to obtain a classification feature vector. Then, the classification feature vector is processed by a classifier to obtain a classification result indicating whether there is a fault in the offshore wind turbine.

[0094] Based on this, this application proposes a data-driven offshore wind turbine fault diagnosis system, comprising: a vibration data acquisition module for acquiring a first vibration signal and a second vibration signal within a predetermined time period collected by first and second vibration sensors deployed at the front and rear ends of the generator; a domain transformation module for converting the first vibration signal and the second vibration signal into a first vibration Gram angle and field image and a second vibration Gram angle and field image using the Gram angle field principle; a spatial attention encoding module for obtaining a first vibration feature matrix and a second vibration feature matrix by using a first convolutional neural network with a spatial attention mechanism to encode the first and second vibration Gram angle and field images respectively; and a vibration transfer module for calculating the first vibration feature matrix relative to the second vibration Gram angle and field image. The system comprises: a vibration transfer matrix of a dynamic feature matrix; an eigenvalue correction module for correcting the eigenvalues ​​at each position in the vibration transfer matrix to obtain a corrected vibration transfer matrix; a speed data acquisition module for acquiring the speed values ​​of the generator at multiple predetermined time points within a predetermined time period, acquired by a transfer measuring instrument; a speed data encoding module for passing the speed values ​​at multiple predetermined time points within the predetermined time period through a time-series encoder containing a one-dimensional convolutional layer to obtain a speed feature vector; a feature fusion module for fusing the speed feature vector and the corrected vibration transfer matrix to obtain a classification feature vector; and a fault diagnosis result generation module for passing the classification feature vector through a classifier to obtain a classification result, the classification result being used to indicate whether the offshore wind turbine has a fault.

[0095] Figure 1 The illustration shows an application scenario of a data-driven offshore wind turbine fault diagnosis system according to an embodiment of this application. For example... Figure 1 As shown, in this application scenario, firstly, vibration is detected by first and second vibration sensors deployed at the beginning and end of the generator (e.g., as shown in the diagram). Figure 1 The first and second vibration signals (as shown in T1 and T2) are acquired within a predetermined time period, and transferred via a transfer measuring instrument (e.g., such as...). Figure 1 The N shown in the diagram collects the rotational speed values ​​of the generator at multiple predetermined time points within the predetermined time period. Then, the acquired first and second vibration signals within the predetermined time period, along with the rotational speed values ​​of the generator at multiple predetermined time points within the predetermined time period, are input to a server deployed with a data-driven offshore wind turbine fault diagnosis algorithm (e.g., such as...). Figure 1The cloud server P shown in the figure is capable of processing the first vibration signal and the second vibration signal within the predetermined time period, as well as the rotational speed values ​​of the generator at multiple predetermined time points within the predetermined time period, using a data-driven offshore wind turbine fault diagnosis algorithm, to generate a classification result indicating whether the offshore wind turbine has a fault.

[0096] After introducing the basic principles of this application, various non-limiting embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0097] Exemplary system

[0098] Figure 2 The diagram illustrates a block diagram of a data-driven offshore wind turbine fault diagnosis system according to an embodiment of this application. Figure 2 As shown, the data-driven offshore wind turbine fault diagnosis system 200 according to an embodiment of this application includes: a vibration data acquisition module 210, used to acquire a first vibration signal and a second vibration signal within a predetermined time period collected by first and second vibration sensors deployed at the front and rear ends of the generator; a domain transformation module 220, used to convert the first vibration signal and the second vibration signal into a first vibration Gram angle and field image and a second vibration Gram angle and field image using the Gram angle field principle; a spatial attention encoding module 230, used to obtain a first vibration feature matrix and a second vibration feature matrix by using a first convolutional neural network with a spatial attention mechanism to encode the first and second vibration Gram angle and field images respectively; and a vibration transfer module 240, used to calculate the first vibration feature matrix relative to the second vibration Gram angle and field image. The system comprises: a vibration transfer matrix of a feature matrix; an eigenvalue correction module 250 for correcting the eigenvalues ​​at each position in the vibration transfer matrix to obtain a corrected vibration transfer matrix; a speed data acquisition module 260 for acquiring the speed values ​​of the generator at multiple predetermined time points within a predetermined time period, acquired by a transfer measuring instrument; a speed data encoding module 270 for passing the speed values ​​at multiple predetermined time points within the predetermined time period through a time-series encoder containing a one-dimensional convolutional layer to obtain a speed feature vector; a feature fusion module 280 for fusing the speed feature vector and the corrected vibration transfer matrix to obtain a classification feature vector; and a fault diagnosis result generation module 290 for passing the classification feature vector through a classifier to obtain a classification result, the classification result being used to indicate whether the offshore wind turbine has a fault.

[0099] Specifically, in this embodiment, the vibration data acquisition module 210 and the domain conversion module 220 are used to acquire a first vibration signal and a second vibration signal within a predetermined time period collected by first and second vibration sensors deployed at the head and tail ends of the generator, and to convert the first vibration signal and the second vibration signal into a first vibration Gram angle and field image and a second vibration Gram angle and field image using the Gram angle field principle. As mentioned above, since there is a certain nonlinear correlation between the difference in vibration signals at the head and tail ends of the generator and its rotational speed under normal conditions during the operation of offshore wind turbines in actual scenarios, the technical solution of this application aims to use this correlation information to perform fault diagnosis and detection of the offshore wind turbine to ensure its normal operation.

[0100] Specifically, in the technical solution of this application, firstly, first and second vibration signals within a predetermined time period are collected by first and second vibration sensors deployed at the beginning and end of the generator. Then, the first and second vibration signals are converted into a first vibration Gramian angular field image and a second vibration Gramian angular field image using the Gramian angular field principle. It should be understood that the Gramian angular field (GAF) is based on the Gram principle, which can transfer time series from the classical Cartesian coordinate system to the polar coordinate system for representation. GAF can well preserve the dependence and correlation of the original vibration time series signal, possessing similar time series characteristics to the original vibration signal. Furthermore, since GAF can yield Gramian angular sum field (GASF) and Gramian angular difference field (GADF) depending on the trigonometric functions used for encoding, and GADF conversion is irreversible, the technical solution of this application selects the inversely convertible GASF conversion method for encoding the vibration signal.

[0101] More specifically, in the embodiments of this application, the encoding steps from the vibration signal to the GASF image are as follows: For a time series with C dimensions = {Q1, Q2, ..., Q...} C}, where each dimension contains n sampling points Q i ={q i1 q i2 , ..., q in First, the data for each dimension is normalized. Then, all values ​​in the data are integrated into the range [-1, 1]. After integration, the normalized values ​​are replaced with trigonometric function values ​​(Cos values), and polar coordinates are used instead of Cartesian coordinates, thus preserving the absolute temporal relationship of the sequence.

[0102] Specifically, in this embodiment, the spatial attention encoding module 230 is used to obtain a first vibration feature matrix and a second vibration feature matrix by using a first convolutional neural network with a spatial attention mechanism to process the first and second vibration Gram angles and field images, respectively. It should be understood that, in practical situations, the vibration signals detected by vibration sensors located at the engine's front and rear ends have specific implicit spatial correlation information. Therefore, in the technical solution of this application, a first convolutional neural network model with a spatial attention mechanism is used to perform feature mining on the first and second vibration Gram angles and field images to extract high-dimensional local implicit feature distribution information with spatial location correlation features from the first and second vibration Gram angles and field images.

[0103] More specifically, in this embodiment, the spatial attention encoding module is further configured to: perform the following operations on the input data during the forward propagation of each layer of the first convolutional neural network: convolution processing on the input data to generate a convolutional feature map; pooling processing on the convolutional feature map to generate a pooled feature map; nonlinear activation on the pooled feature map to generate an activation feature map; calculating the mean along the channel dimension of each position of the activation feature map to generate a spatial feature matrix; calculating the Softmax-like function value at each position in the spatial feature matrix to obtain a spatial score matrix; and calculating the positional dot product of the spatial feature matrix and the spatial score map to obtain a feature matrix; wherein the feature matrix output by the last layer of the first convolutional neural network is the first vibration feature matrix and the second vibration feature matrix.

[0104] Specifically, in this embodiment, the vibration transfer module 240 is used to calculate the vibration transfer matrix of the first vibration feature matrix relative to the second vibration feature matrix. It should be understood that, under normal conditions, the difference in vibration signals between the beginning and end of the generator exhibits a certain pattern of nonlinear correlation with its rotational speed. Therefore, in the technical solution of this application, this nonlinear correlation can be represented based on the transfer matrix, thereby enabling fault detection of the offshore wind turbine. That is, specifically, the vibration transfer matrix of the first vibration feature matrix relative to the second vibration feature matrix is ​​calculated.

[0105] More specifically, in this embodiment of the application, the vibration transfer module is further configured to: calculate the vibration transfer matrix of the first vibration feature matrix relative to the second vibration feature matrix using the following formula;

[0106] The formula is as follows:

[0107] S = T * F

[0108] Where F represents the first vibration feature matrix, T represents the vibration transfer matrix, and S represents the second vibration feature matrix.

[0109] Specifically, in this embodiment, the feature value correction module 250 is used to correct the feature values ​​at each position in the vibration transfer matrix to obtain a corrected vibration transfer matrix. It should be understood that, due to the spatial attention mechanism of the first convolutional neural network, the first vibration feature matrix and the second vibration feature matrix can enhance the extraction of local semantics, but at the same time weaken the correlation between local semantics, resulting in poor global semantic expression. The vibration transfer matrix also suffers from the same problem. Therefore, in the technical solution of this application, it is preferable to optimize the local semantic correlation expression ability and even the global expression ability of the vibration transfer matrix. Accordingly, in a specific example, the formula for correcting the feature values ​​at each position in the vibration transfer matrix to obtain the corrected vibration transfer matrix is:

[0110]

[0111] Where M represents the vibration transfer matrix, and Cov1() and Cov2() are both single convolutional layers. Used to map two-dimensional position coordinates to one-dimensional numerical values, P M Let ⊙ represent the (x,y) coordinate tensor of matrix M. These represent matrix dot multiplication and dot addition, respectively. Represents matrix multiplication. Let n represent the square root of the value at each position in the matrix, where n is the number of rows multiplied by the number of columns in the matrix. It should be understood that the optimization described here employs a positional proposal-based local reasoning transformation to comprehensively integrate the captured local semantics and further derive global semantics. Specifically, it uses positional information as a proposal to reason about global scene semantics through a local perceptive field. Furthermore, by adjusting the parameters of the convolutional layers Cov1() and Cov2(), a balance can be further achieved between local-global information transfer and global information reasoning prediction, thereby improving the classification accuracy of the classification feature vector by enhancing the class prediction capability of the vibrational transfer matrix.

[0112] More specifically, in this embodiment, the eigenvalue correction module includes: first, calculating the product between the vibration transfer matrix and its transpose to obtain an autocorrelation feature matrix. Next, taking the square root of the eigenvalues ​​at each position in the autocorrelation feature matrix to obtain an autocorrelation compressed feature matrix, and dividing the eigenvalues ​​at each position in the autocorrelation compressed feature matrix by the product of the number of rows and columns of the autocorrelation compressed feature matrix to obtain a dimension-compressed autocorrelation compressed feature matrix. Accordingly, in a specific example, the eigenvalues ​​at each position in the autocorrelation feature matrix are taken square root using the following formula to obtain the autocorrelation compressed feature matrix, and the eigenvalues ​​at each position in the autocorrelation compressed feature matrix are divided by the product of the number of rows and columns of the autocorrelation compressed feature matrix to obtain the dimension-compressed autocorrelation compressed feature matrix;

[0113] The formula is as follows:

[0114]

[0115] Where M represents the vibration transfer matrix, and M1 represents the dimension-compressed self-correlated compressed feature matrix. Represents matrix multiplication. This represents the square root of the value at each position in the matrix, where n is the number of rows multiplied by the number of columns in the matrix.

[0116] Then, the vibration transfer matrix is ​​convolutionally encoded to obtain a first convolutional vibration transfer matrix. Next, the positional sum of the first convolutional vibration transfer matrix and the dimensionally compressed self-correlated compressed feature matrix is ​​calculated to obtain a fused feature matrix. Then, the fused feature matrix is ​​convolutionally encoded to obtain a multi-scale feature matrix. Next, the two-dimensional position coordinates of each position in the coordinate tensor of the vibration transfer matrix are converted into one-dimensional values ​​to obtain a position feature matrix. Accordingly, in a specific example, the position information encoding unit is further configured to: convert the two-dimensional position coordinates of each position in the coordinate tensor of the vibration transfer matrix into one-dimensional values ​​using the following formula to obtain the position feature matrix;

[0117] The formula is as follows:

[0118] M2=Φ(P M )

[0119] Where M represents the vibration transfer matrix, and M2 represents the position feature matrix. Used to map two-dimensional position coordinates to one-dimensional numerical values, P M Let (x,y) represent the (x,y) coordinate tensor of matrix M.

[0120] Finally, the corrected vibration transfer matrix is ​​obtained by multiplying the position feature matrix and the multi-scale feature matrix by position.

[0121] Figure 3 The diagram illustrates a block diagram of an eigenvalue correction module in a data-driven offshore wind turbine fault diagnosis system according to an embodiment of this application. Figure 3 As shown, the eigenvalue correction module 250 includes: a self-correlation encoding unit 251, used to calculate the product between the vibration transfer matrix and its transpose to obtain a self-correlation feature matrix; an eigenvalue compression unit 252, used to square-root the eigenvalues ​​at each position in the self-correlation feature matrix to obtain a self-correlation compressed feature matrix, and to divide the eigenvalues ​​at each position in the self-correlation compressed feature matrix by the product of the number of rows and columns of the self-correlation compressed feature matrix to obtain a dimension-compressed self-correlation compressed feature matrix; and a first convolution unit 253, used to perform convolution encoding on the vibration transfer matrix to obtain a first The system comprises: a convolutional vibration transfer matrix; a fusion unit 254, configured to calculate the positional sum of the first convolutional vibration transfer matrix and the dimensionally compressed self-correlated compressed feature matrix to obtain a fused feature matrix; a second convolutional unit 255, configured to perform convolutional encoding on the fused feature matrix to obtain a multi-scale feature matrix; a positional information encoding unit 256, configured to convert the two-dimensional positional coordinates of each position in the coordinate tensor of the vibration transfer matrix into one-dimensional numerical values ​​to obtain a positional feature matrix; and a re-fusion unit 257, configured to calculate the positional dot product of the positional feature matrix and the multi-scale feature matrix to obtain the corrected vibration transfer matrix.

[0122] Specifically, in this embodiment, the speed data acquisition module 260 and the speed data encoding module 270 are used to acquire the speed values ​​of the generator at multiple predetermined time points within a predetermined time period, collected by the transfer measuring instrument, and to obtain a speed feature vector by passing the speed values ​​at multiple predetermined time points within the predetermined time period through a time-series encoder containing a one-dimensional convolutional layer. That is, in the technical solution of this application, furthermore, in order to detect and diagnose the faults of the offshore wind turbine, it is also necessary to acquire the speed values ​​of the generator at multiple predetermined time points within a predetermined time period through the transfer measuring instrument. Furthermore, considering that the speed values ​​have a dynamic change pattern in the time dimension, a time-series encoder is used to encode the speed values ​​at multiple predetermined time points within the predetermined time period in order to fully extract this pattern. Accordingly, in a specific example, the time-series encoder consists of alternating fully connected layers and one-dimensional convolutional layers, which extracts the correlation of the generator's speed values ​​in the time dimension through one-dimensional convolutional encoding and extracts the high-dimensional implicit features of the generator's speed values ​​through fully connected encoding.

[0123] More specifically, in this embodiment, the rotational speed data encoding module is further configured to: arrange the rotational speed values ​​at multiple predetermined time points within the predetermined time period into a one-dimensional input vector according to the time dimension; and use the fully connected layer of the timing encoder to perform fully connected encoding on the input vector using the following formula to extract the high-dimensional latent features of the feature values ​​at each position in the input vector, wherein the formula is: Where X is the input vector, Y is the output vector, W is the weight matrix, and B is the bias vector. This represents matrix multiplication; the one-dimensional convolutional layer of the temporal encoder is used to perform one-dimensional convolutional encoding on the input vector using the following formula to extract high-dimensional implicit correlation features between feature values ​​at various positions in the input vector, wherein the formula is:

[0124]

[0125] Where a is the width of the convolution kernel in the x-direction, F is the convolution kernel parameter vector, G is the local vector matrix operated on by the convolution kernel function, w is the size of the convolution kernel, and X represents the input vector.

[0126] Specifically, in this embodiment, the feature fusion module 280 and the fault diagnosis result generation module 290 are used to fuse the rotational speed feature vector and the corrected vibration transfer matrix to obtain a classification feature vector, and then pass the classification feature vector through a classifier to obtain a classification result. The classification result is used to indicate whether the offshore wind turbine has a fault. That is, in the technical solution of this application, the rotational speed feature vector and the corrected vibration transfer matrix are further fused by matrix multiplication to obtain a classification feature vector. Then, the classification feature vector is further processed by a classifier to obtain a classification result indicating whether the offshore wind turbine has a fault. Accordingly, in a specific example, the classifier is used to process the classification feature vector with the following formula to obtain the classification result, wherein the formula is: softmax{(W n B n ):…:(W1,B1)|X}, where W1 to W n The weight matrix is ​​B1 to B1. n X is the bias vector, and X is the classification feature vector.

[0127] More specifically, in this embodiment of the application, the feature fusion module is further configured to: fuse the rotational speed feature vector and the corrected vibration transfer matrix using the following formula to obtain the classification feature vector;

[0128] The formula is as follows:

[0129]

[0130] Where V represents the rotational speed feature vector, M' represents the corrected vibration transfer matrix, and V' represents the classification feature vector. This represents matrix multiplication.

[0131] In summary, the data-driven offshore wind turbine fault diagnosis system 200 described in this application embodiment is explained. It employs artificial intelligence detection technology, using a deep neural network model as a feature extractor, and utilizes a correlation model between the difference in vibration signals at the generator's beginning and end and its rotational speed to perform fault diagnosis and detection of the offshore wind turbine. This ensures the timeliness and effectiveness of fault maintenance for the offshore wind turbine, thereby guaranteeing its normal operation.

[0132] As described above, the data-driven offshore wind turbine fault diagnosis system 200 according to the embodiments of this application can be implemented in various terminal devices, such as servers for data-driven offshore wind turbine fault diagnosis algorithms. In one example, the data-driven offshore wind turbine fault diagnosis system 200 according to the embodiments of this application can be integrated into the terminal device as a software module and / or hardware module. For example, the data-driven offshore wind turbine fault diagnosis system 200 can be a software module in the operating system of the terminal device, or it can be an application developed for the terminal device; of course, the data-driven offshore wind turbine fault diagnosis system 200 can also be one of many hardware modules of the terminal device.

[0133] Alternatively, in another example, the data-driven offshore wind turbine fault diagnosis system 200 and the terminal device can also be separate devices, and the data-driven offshore wind turbine fault diagnosis system 200 can be connected to the terminal device via wired and / or wireless networks, and transmit interactive information in accordance with an agreed data format.

[0134] Exemplary method

[0135] Figure 4 The diagram illustrates a flowchart of a data-driven fault diagnosis system for offshore wind turbines. For example... Figure 4As shown, the diagnostic method of the data-driven offshore wind turbine fault diagnosis system according to an embodiment of this application includes the following steps: S110, acquiring a first vibration signal and a second vibration signal within a predetermined time period collected by first and second vibration sensors deployed at the front and rear ends of the generator; S120, converting the first vibration signal and the second vibration signal into a first vibration Gram angle and field image and a second vibration Gram angle and field image using the Gram angle field principle; S130, obtaining a first vibration feature matrix and a second vibration feature matrix by using a first convolutional neural network with a spatial attention mechanism on the first and second vibration Gram angle and field images respectively; S140, calculating the phase of the first vibration feature matrix... For the vibration transfer matrix of the second vibration feature matrix; S150, the feature values ​​at each position in the vibration transfer matrix are corrected to obtain a corrected vibration transfer matrix; S160, the rotational speed values ​​of the generator at multiple predetermined time points within the predetermined time period are acquired by the transfer measuring instrument; S170, the rotational speed values ​​at multiple predetermined time points within the predetermined time period are passed through a time encoder containing a one-dimensional convolutional layer to obtain a rotational speed feature vector; S180, the rotational speed feature vector and the corrected vibration transfer matrix are fused to obtain a classification feature vector; and S190, the classification feature vector is passed through a classifier to obtain a classification result, the classification result being used to indicate whether the offshore wind turbine has a fault.

[0136] Figure 5 The illustration shows a schematic diagram of the architecture of a diagnostic method for a data-driven offshore wind turbine fault diagnosis system according to an embodiment of this application. Figure 5 As shown, in the network architecture of the diagnostic method of the data-driven offshore wind turbine fault diagnosis system, firstly, the obtained first vibration signal (e.g., as shown in the figure) is analyzed using the Gram angle field principle. Figure 5 The P1 shown in the diagram and the second vibration signal (e.g., as shown in the diagram) Figure 5 The P2 shown is converted into the first vibration Gram angle and field image (e.g., as shown in the figure). Figure 5 The Q1 shown is the second vibration Gram angle and field image (e.g., as shown in the figure). Figure 5 (Q2 as shown); then, the first and second vibration gram angles and field images are respectively processed using a first convolutional neural network with a spatial attention mechanism (e.g., as shown in the image). Figure 5 The CNN1 shown is used to obtain the first vibration feature matrix (e.g., as shown in the diagram). Figure 5 The MF1 shown is the second vibration characteristic matrix (e.g., as shown in the figure). Figure 5 (as shown in MF2); then, calculate the vibration transfer matrix of the first vibration feature matrix relative to the second vibration feature matrix (e.g., as shown in MF2); Figure 5(as shown in MT1); then, the eigenvalues ​​at each position in the vibration transfer matrix are corrected to obtain the corrected vibration transfer matrix (e.g., as shown in MT1); Figure 5 (as shown in MT2); then, the obtained rotational speed values ​​at multiple predetermined time points within the predetermined time period (e.g., such as MT2); Figure 5 The IN shown is transmitted through a temporal encoder containing a one-dimensional convolutional layer (e.g., as shown in the figure). Figure 5 E as shown in the figure is used to obtain the rotational speed feature vector (e.g., as shown in the figure). Figure 5 The V shown in the diagram is then used to fuse the rotational speed feature vector and the corrected vibration transfer matrix to obtain a classification feature vector (e.g., as shown in the diagram). Figure 5 The VF shown in the diagram); and finally, the classification feature vector is passed through a classifier (e.g., such as...). ​ The S shown is used to obtain a classification result, which is used to indicate whether there is a fault in the offshore wind turbine.

[0137] More specifically, in steps S110 and S120, a first vibration signal and a second vibration signal within a predetermined time period are acquired by first and second vibration sensors deployed at the beginning and end of the generator. The first and second vibration signals are then converted into a first vibration Gram angle and field image and a second vibration Gram angle and field image using the Gram angle field principle. It should be understood that, in actual offshore wind turbine operation, under normal conditions, there is a certain nonlinear correlation between the difference in vibration signals at the beginning and end of the generator and its rotational speed. Therefore, in the technical solution of this application, it is desirable to use this correlation information for fault diagnosis and detection of the offshore wind turbine to ensure its normal operation.

[0138] Specifically, in the technical solution of this application, firstly, first and second vibration signals within a predetermined time period are collected by first and second vibration sensors deployed at the beginning and end of the generator. Then, the first and second vibration signals are converted into a first vibration Gramian angular field image and a second vibration Gramian angular field image using the Gramian angular field principle. It should be understood that the Gramian angular field (GAF) is based on the Gram principle, which can transfer time series from the classical Cartesian coordinate system to the polar coordinate system for representation. GAF can well preserve the dependence and correlation of the original vibration time series signal, possessing similar time series characteristics to the original vibration signal. Furthermore, since GAF can yield Gramian angular sum field (GASF) and Gramian angular difference field (GADF) depending on the trigonometric functions used for encoding, and GADF conversion is irreversible, the technical solution of this application selects the inversely convertible GASF conversion method for encoding the vibration signal.

[0139] More specifically, in step S130, the first and second vibration Gram angle and field images are respectively processed using a first convolutional neural network with a spatial attention mechanism to obtain a first vibration feature matrix and a second vibration feature matrix. It should be understood that, in practical situations, the vibration signals detected by vibration sensors located at the engine's front and rear ends have specific implicit spatial correlation information. Therefore, in the technical solution of this application, a first convolutional neural network model with a spatial attention mechanism is used to perform feature mining on the first and second vibration Gram angle and field images to extract high-dimensional local implicit feature distribution information with spatial location correlation features from the first and second vibration Gram angle and field images.

[0140] More specifically, in step S140, the vibration transfer matrix of the first vibration feature matrix relative to the second vibration feature matrix is ​​calculated. It should be understood that, under normal conditions, the difference in vibration signals between the beginning and end of the generator exhibits a certain pattern of nonlinear correlation with its rotational speed. Therefore, in the technical solution of this application, this nonlinear correlation can be represented based on the transfer matrix, thereby enabling fault detection of the offshore wind turbine. That is, specifically, the vibration transfer matrix of the first vibration feature matrix relative to the second vibration feature matrix is ​​calculated.

[0141] More specifically, in step S150, the feature values ​​at each position in the vibration transfer matrix are corrected to obtain the corrected vibration transfer matrix. It should be understood that, due to the spatial attention mechanism of the first convolutional neural network, the first vibration feature matrix and the second vibration feature matrix can enhance the extraction of local semantics, but at the same time weaken the correlation between local semantics, resulting in poor global semantic expression. The vibration transfer matrix also suffers from the same problem. Therefore, in the technical solution of this application, it is preferable to optimize the local semantic correlation expression ability and even the global expression ability of the vibration transfer matrix. Accordingly, in a specific example, the formula for correcting the feature values ​​at each position in the vibration transfer matrix to obtain the corrected vibration transfer matrix is:

[0142]

[0143] Where M represents the vibration transfer matrix, and Cov1() and Cov2() are both single convolutional layers. Used to map two-dimensional position coordinates to one-dimensional numerical values, P M Let ⊙ represent the (x,y) coordinate tensor of matrix M. These represent matrix dot multiplication and dot addition, respectively. Represents matrix multiplication. Let n represent the square root of the value at each position in the matrix, where n is the number of rows multiplied by the number of columns in the matrix. It should be understood that the optimization described here employs a positional proposal-based local reasoning transformation to comprehensively integrate the captured local semantics and further derive global semantics. Specifically, it uses positional information as a proposal to reason about global scene semantics through a local perceptive field. Furthermore, by adjusting the parameters of the convolutional layers Cov1() and Cov2(), a balance can be further achieved between local-global information transfer and global information reasoning prediction, thereby improving the classification accuracy of the classification feature vector by enhancing the class prediction capability of the vibrational transfer matrix.

[0144] More specifically, in steps S160 and S170, the rotational speed values ​​of the generator at multiple predetermined time points within the predetermined time period are acquired by the transfer measuring instrument, and the rotational speed values ​​at multiple predetermined time points within the predetermined time period are processed by a time-series encoder containing a one-dimensional convolutional layer to obtain a rotational speed feature vector. That is, in the technical solution of this application, furthermore, in order to detect and diagnose the faults of the offshore wind turbine, it is also necessary to acquire the rotational speed values ​​of the generator at multiple predetermined time points within the predetermined time period using a transfer measuring instrument. Furthermore, considering that the rotational speed values ​​have a dynamic change pattern in the time dimension, a time-series encoder is used to encode the rotational speed values ​​at multiple predetermined time points within the predetermined time period in order to fully extract this pattern. Accordingly, in a specific example, the time-series encoder consists of alternating fully connected layers and one-dimensional convolutional layers, which extract the temporal correlation of the generator's rotational speed values ​​through one-dimensional convolutional encoding and extract the high-dimensional implicit features of the generator's rotational speed values ​​through fully connected encoding.

[0145] More specifically, in steps S180 and S190, the rotational speed feature vector and the corrected vibration transfer matrix are fused to obtain a classification feature vector, and the classification feature vector is then passed through a classifier to obtain a classification result. The classification result is used to indicate whether the offshore wind turbine has a fault. That is, in the technical solution of this application, the rotational speed feature vector and the corrected vibration transfer matrix are further fused by matrix multiplication to obtain a classification feature vector. Then, the classification feature vector is further processed by a classifier to obtain a classification result indicating whether the offshore wind turbine has a fault.

[0146] In summary, the diagnostic method of the data-driven offshore wind turbine fault diagnosis system described in the embodiments of this application is explained. It employs artificial intelligence detection technology, using a deep neural network model as a feature extractor, and utilizes a correlation model between the difference in vibration signals at the beginning and end of the generator and its rotational speed to perform fault diagnosis and detection of the offshore wind turbine. This ensures the timeliness and effectiveness of fault maintenance for the offshore wind turbine, thereby guaranteeing its normal operation.

[0147] The basic principles of this application have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this application are merely examples and not limitations, and should not be considered as essential features of each embodiment of this application. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the application to the necessity of employing the aforementioned specific details for implementation.

[0148] The block diagrams of devices, apparatuses, devices, and systems involved in this application are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.

[0149] It should also be noted that in the apparatus, equipment, and methods of this application, the components or steps can be disassembled and / or recombined. These disassemblies and / or recombinations should be considered as equivalent solutions of this application.

[0150] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of this application. Therefore, this application is not intended to be limited to the aspects shown herein, but rather to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0151] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this application to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A data-driven based offshore wind turbine fault diagnosis system, characterized in that, The method comprises the following steps: a vibration data acquisition module is configured to acquire first vibration signals and second vibration signals collected by first and second vibration sensors arranged at the head end and tail end of the generator within a predetermined time period; a domain transformation module is configured to transform the first vibration signals and the second vibration signals into first vibration gram angle and field images and second vibration gram angle and field images using the gram angle field principle; a spatial attention encoding module is configured to obtain first and second vibration feature matrices by respectively using a first convolutional neural network with a spatial attention mechanism to process the first and second vibration gram angle and field images; a vibration transfer module is configured to calculate a vibration transfer matrix of the first vibration feature matrix relative to the second vibration feature matrix; an eigenvalue correction module is configured to correct the eigenvalues of each position in the vibration transfer matrix to obtain a corrected vibration transfer matrix; a rotational speed data acquisition module is configured to acquire rotational speed values of the generator at a plurality of predetermined time points within the predetermined time period collected by a transfer measuring instrument; a rotational speed data encoding module is configured to obtain a rotational speed feature vector by processing the rotational speed values at the plurality of predetermined time points within the predetermined time period through a time series encoder comprising a one-dimensional convolution layer; a feature fusion module is configured to fuse the rotational speed feature vector and the corrected vibration transfer matrix to obtain a classification feature vector; and a fault diagnosis result generation module is configured to obtain a classification result by processing the classification feature vector through a classifier, wherein the classification result is used to indicate whether the offshore wind turbine has a fault. The eigenvalue correction module comprises: a self-association encoding unit configured to calculate the product between the vibration transfer matrix and the transpose matrix of the vibration transfer matrix to obtain a self-association feature matrix; an eigenvalue compression unit configured to take the square root of the eigenvalues of each position in the self-association feature matrix to obtain a self-association compressed feature matrix, and divide the eigenvalues of each position in the self-association compressed feature matrix by the product of the number of rows and the number of columns of the self-association compressed feature matrix to obtain a dimension-compressed self-association compressed feature matrix; a first convolution unit configured to perform convolutional encoding on the vibration transfer matrix to obtain a first convolutional vibration transfer matrix; a fusion unit configured to calculate the positional sum of the first convolutional vibration transfer matrix and the dimension-compressed self-association compressed feature matrix to obtain a fusion feature matrix; a second convolution unit configured to perform convolutional encoding on the fusion feature matrix to obtain a multi-scale feature matrix; a position information encoding unit configured to convert the two-dimensional position coordinates of each position in the coordinate tensor of the vibration transfer matrix into one-dimensional numerical values to obtain a position feature matrix; and a re-fusion unit configured to calculate the positional dot product of the position feature matrix and the multi-scale feature matrix to obtain the corrected vibration transfer matrix. The spatial attention encoding module is further configured to perform the following operations on the input data in the forward propagation process of each layer of the first convolutional neural network:

2. The data-driven based offshore wind turbine fault diagnosis system of claim 1, wherein, performing convolutional processing on the input data to generate a convolutional feature map; performing pooling processing on the convolutional feature map to generate a pooled feature map; ​ performing nonlinear activation on the pooled feature map to generate an activated feature map; calculating the mean value of each position of the activated feature map along the channel dimension to generate a spatial feature matrix; calculating the Softmax function value of each position in the spatial feature matrix to obtain a spatial score matrix; and calculating the point-by-point multiplication of the spatial feature matrix and the spatial score map to obtain a feature matrix; wherein the feature matrix output by the last layer of the first convolutional neural network is the first vibration feature matrix and the second vibration feature matrix.

3. The data-driven based offshore wind turbine fault diagnosis system of claim 2, wherein, The vibration transfer module is further used to calculate the vibration transfer matrix of the first vibration feature matrix relative to the second vibration feature matrix according to the following formula: wherein the formula is: S = T * F wherein F represents the first vibration feature matrix, T represents the vibration transfer matrix, and S represents the second vibration feature matrix.

4. The data-driven based offshore wind turbine fault diagnosis system of claim 3, wherein, The feature value compression unit is further used to square the feature values of each position in the autocorrelation feature matrix according to the following formula to obtain the autocorrelation compressed feature matrix, and divide the feature values of each position in the autocorrelation compressed feature matrix by the product of the number of rows and the number of columns of the autocorrelation compressed feature matrix to obtain the dimension-compressed autocorrelation compressed feature matrix. wherein the formula is: where M represents the vibration transfer matrix, Mi represents the dimensionally compressed auto-correlation compressed feature matrix, represents matrix multiplication, represents taking the square root of the value at each location of the matrix, and n is the number of rows times the number of columns of the matrix.

5. The data-driven based offshore wind turbine fault diagnosis system of claim 4, wherein, The position information encoding unit is further used to convert the two-dimensional position coordinates of each position in the coordinate tensor of the vibration transfer matrix into one-dimensional numerical values according to the following formula to obtain the position feature matrix. wherein the formula is: M2 = Φ(P M ) where M represents the vibration transfer matrix, M2represents the position characteristic matrix, for mapping two-dimensional position coordinates to one-dimensional values, P M represents the (x, y) coordinate tensor of matrix M.

6. The data-driven based offshore wind turbine fault diagnosis system of claim 5, wherein, The rotation speed data encoding module is further used to: arrange the rotation speed values at a plurality of predetermined time points in a predetermined time period into a one-dimensional input vector according to the time dimension; use the full connection layer of the time sequence encoder to full connection encode the input vector with the following formula to extract the high-dimensional hidden features of the feature values of each position in the input vector, wherein the formula is: wherein X is the input vector, Y is the output vector, W is the weight matrix, B is the bias vector, represents matrix multiplication; use the one-dimensional convolution layer of the time sequence encoder to perform one-dimensional convolution encoding on the input vector according to the following formula to extract high-dimensional implicit correlation features between the feature values of each position in the input vector, wherein the formula is: wherein a is the width of the convolution kernel in the x direction, F is the convolution kernel parameter vector, G is the local vector matrix operated with the convolution kernel function, w is the size of the convolution kernel, and X represents the input vector.

7. The data-driven based offshore wind turbine fault diagnosis system of claim 6, wherein, The feature fusion module is further used to fuse the rotation speed feature vector and the corrected vibration transfer matrix according to the following formula to obtain the classification feature vector. wherein the formula is: where V represents the rotational speed feature vector, M ' represents the corrected vibration transfer matrix, V ' represents the classification feature vector, represents matrix multiplication.

8. The data-driven based offshore wind turbine fault diagnosis system of claim 7, wherein, The fault diagnosis result generation module is further configured to use the classifier to process the classification feature vector according to a formula to obtain the classification result, where the formula is: softmax{(W n ,B n ):…:(W1,B1)|X} where W1 to W n are weight matrices, B1 to B n are bias vectors, and X is the classification feature vector.

9. A diagnosis method of a data-driven offshore wind turbine fault diagnosis system, using the data-driven offshore wind turbine fault diagnosis system according to any one of claims 1-8, characterized in that, It comprises: obtaining first vibration signals and second vibration signals in a predetermined time period collected by first and second vibration sensors deployed at the head end and tail end of the generator; using the principle of Gram angle field to convert the first vibration signals and the second vibration signals into first vibration Gram angle and field images and second vibration Gram angle and field images; using a first convolutional neural network with a spatial attention mechanism to obtain a first vibration feature matrix and a second vibration feature matrix from the first and second vibration Gram angle and field images, respectively; calculating the vibration transfer matrix of the first vibration feature matrix relative to the second vibration feature matrix; correcting the feature values of each position in the vibration transfer matrix to obtain a corrected vibration transfer matrix; obtaining a plurality of rotational speed values of the generator at a plurality of predetermined time points in a predetermined time period collected by a transfer measurement instrument; obtaining a plurality of rotational speed values of the generator at a plurality of predetermined time points in a predetermined time period collected by a transfer measurement instrument; obtaining a plurality of rotational speed values of the generator at a plurality of predetermined time points in a predetermined time period collected by a transfer measurement instrument; obtaining a plurality of rotational speed values of the generator at a plurality of predetermined time points in a predetermined time period collected by a transfer measurement instrument; obtaining a plurality of rotational speed values of the generator at a plurality of predetermined time points in a predetermined time period collected by a transfer measurement instrument; obtaining a plurality of rotational speed values of the generator at a plurality of predetermined time points in a predetermined time period collected by a transfer measurement instrument; obtaining a plurality of rotational speed values of the generator at a plurality of predetermined time points in a predetermined time period collected by a transfer measurement instrument; obtaining a plurality of rotational speed values of the generator at a plurality of predetermined time points in a predetermined time period collected by a transfer measurement instrument; obtaining a plurality of rotational speed values of the generator at a plurality of predetermined time points in a predetermined time period collected by a transfer measurement instrument

Images