Single-sample wind turbine generator transmission chain fault diagnosis method and system

Abstract

The invention discloses a single-sample wind turbine generator transmission chain fault diagnosis method and system, and belongs to the technical field of data processing. Comprising the following steps: performing empirical mode decomposition on an actual monitoring fault signal to obtain signal components of different modes; and calculating an energy value and a peak value of each modal component, and selecting a part of modal components with large energy and high peak values for signal reconstruction to obtain a new fault signal. Wavelet packet decomposition is carried out on the new fault signal, and signal reconstruction is carried out on a third-layer component of wavelet packet decomposition. And taking the variance of the reconstructed signal as a characteristic value of fault diagnosis. And the characteristic values are subjected to nonlinear equalization processing, so that the problem of mutual flooding of signals is solved. The concept of distinction degree is introduced to quantify the distinction between different fault signals. According to the fault diagnosis method, the distinction degree among the four faults on the transmission chain of the wind turbine generator is obviously increased before and after processing, and the method has high robustness. In actual production, the method is of great significance in reducing troubleshooting time and increasing production efficiency in engineering.

Description

Technical Field

[0001] The present invention belongs to the technical field of data processing, and in particular relates to a single-sample wind turbine transmission chain fault diagnosis method and system. Background Art

[0002] Due to the special working characteristics of wind power equipment, wind power equipment is generally located in remote areas, the operating environment is relatively harsh, and maintenance is difficult. The single unit capacity of wind turbines is large, the overall structure of the unit is relatively complex, and the coupling between the components is relatively tight. Therefore, as the number of wind turbines installed increases, the probability of unit failure will also increase. The wind turbine operating cost assessment shows that the operation and maintenance costs of onshore wind turbines are about 15% to 25% of the total cost, and offshore wind turbines are about 22% to 40% of the total cost. The working conditions of offshore wind turbines are worse than those on land, and the operation and maintenance operations are significantly affected by tides, typhoons, and thunderstorms. In addition, offshore maintenance operations have the characteristics of short effective time and poor accessibility. Therefore, it is very important for operating companies to improve the reliability of wind turbines, reduce the failure rate of units, and timely and accurately determine the reasons for wind turbine shutdowns and the types of failures for maintenance and cost savings.

[0003] Wind turbines are not a single structure, but mainly consist of turbines, gearboxes and generators. The frequent switching of the operating state of the transmission chain inside the wind turbine and the complexity of the coupling relationship between the transmission chain structures are the main causes of wind turbine failures. Gearbox failure is widely considered to be the main problem of condition monitoring in all subsystems of the wind turbine transmission chain, and bearing failure is the main factor affecting the operation of the wind turbine gearbox. About 40% of generator failures are related to bearings, 38% to stators and 10% to rotors. Therefore, only considering the fault diagnosis problem of a single part of the wind turbine unit cannot meet the needs of actual engineering. It is necessary to solve the fault classification problem of high similarity in different parts starting from the wind turbine transmission chain. In addition, in actual wind farms, wind turbine equipment has the characteristics of low fault tolerance. Therefore, it is not easy to collect a large amount of fault data, and how to solve the fault diagnosis problem with a small number of samples or even a single sample has become a research focus. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention proposes a single-sample wind turbine transmission chain fault diagnosis method and system, which realizes wind turbine transmission chain fault diagnosis with only a single sample based on empirical mode decomposition and signal equalization processing methods.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] A single-sample wind turbine transmission chain fault diagnosis method comprises the following steps:

[0007] S1: Get original fault data;

[0008] S2: Decompose the original fault data through empirical mode decomposition, and reconstruct it according to preset requirements to obtain new fault data;

[0009] S3: Perform wavelet packet decomposition on the new fault data to extract the variance as the characteristic value of fault diagnosis;

[0010] S4: Perform balanced processing on the characteristic values ​​to achieve fault classification.

[0011] Preferably, in S2, the method of decomposing the original fault data by empirical mode decomposition and reconstructing the original fault data according to preset requirements to obtain new fault data includes:

[0012] Perform empirical mode decomposition on the original fault data to obtain signal components of different modes;

[0013] Calculate the energy value and peak value of each modal signal component;

[0014] Select the modal signal components whose energy values ​​and peak values ​​meet the preset requirements to perform signal reconstruction to obtain new fault data;

[0015] Among them, the expression for calculating the energy value of each modal signal component is:

[0016] ;

[0017] In the formula, Indicates time, represents the time domain signal value, is the energy value of the time domain signal;

[0018] The expression for calculating the peak value of each modal signal component is:

[0019] ;

[0020] In the formula, The time variable is The time domain signal value, is the signal standard deviation, is the signal mean, Represents the entire time domain value of the signal, Indicates the signal kurtosis value.

[0021] Preferably, in S3, the method of performing wavelet packet decomposition on the new fault data to extract the variance as the characteristic value of the fault diagnosis includes:

[0022] Perform 3-layer wavelet packet decomposition on the new fault data;

[0023] Reconstruct the signal by using the third layer component of wavelet packet decomposition;

[0024] The variance of the reconstructed signal is used as the characteristic value of fault diagnosis;

[0025] Among them, the variance calculation formula of the reconstructed signal is:

[0026] ,

[0027] In the formula, the data is For the overall A sample of is the number of sample individuals, is the sample mean.

[0028] Preferably, in S4, the method of performing equalization processing on the characteristic values ​​to achieve fault classification includes:

[0029] ,

[0030] in, represents the fault signal variance, represents the regulatory factor, is the base of the logarithm, Indicates the signal value after equalization.

[0031] The present invention also provides a single-sample wind turbine transmission chain fault diagnosis system, comprising: an acquisition module, a reconstruction module, an extraction module and a classification module;

[0032] The acquisition module is used to acquire original fault data;

[0033] The reconstruction module is used to decompose the original fault data through empirical mode decomposition, and reconstruct it according to preset requirements to obtain new fault data;

[0034] The extraction module is used to perform wavelet packet decomposition on new fault data to extract variance as a characteristic value for fault diagnosis;

[0035] The classification module is used to perform balanced processing on the characteristic values ​​to achieve fault classification.

[0036] Preferably, the reconstruction module comprises: a first decomposition unit, a first calculation unit and a first reconstruction unit;

[0037] The first decomposition unit is used to perform empirical mode decomposition on the original fault data to obtain signal components of different modes;

[0038] The first calculation unit is used to calculate the energy value and peak value of each modal signal component;

[0039] The first reconstruction unit is used to select a modal signal component whose energy value and peak value meet preset requirements to perform signal reconstruction to obtain new fault data;

[0040] Among them, the expression for calculating the energy value of each modal signal component is:

[0041] ;

[0042] In the formula, Indicates time, represents the time domain signal value, is the energy value of the time domain signal;

[0043] The expression for calculating the peak value of each modal signal component is:

[0044] ;

[0045] In the formula, The time variable is The time domain signal value, is the signal standard deviation, is the signal mean, Represents the entire time domain value of the signal, Indicates the signal kurtosis value.

[0046] Preferably, the extraction module comprises: a second decomposition unit, a second reconstruction unit and a second calculation unit;

[0047] The second decomposition unit is used to perform 3-layer wavelet packet decomposition on the new fault data;

[0048] The second reconstruction unit is used to reconstruct the third layer component of the wavelet packet decomposition;

[0049] The second calculation unit is used to use the variance of the reconstructed signal as a characteristic value for fault diagnosis;

[0050] Among them, the variance calculation formula of the reconstructed signal is:

[0051] ,

[0052] In the formula, the data is For the overall A sample of is the number of sample individuals, is the sample mean.

[0053] Preferably, in the classification module, the process of performing balanced processing on the characteristic values ​​to achieve fault classification includes:

[0054] ,

[0055] in, represents the fault signal variance, represents the regulatory factor, is the base of the logarithm, Indicates the signal value after equalization.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] The present invention filters fault data from two dimensions, peak value and energy, and improves the probability of fault data features. The present invention performs three-layer wavelet packet decomposition to extract variance as a eigenvalue. From the results of discrimination calculation, this feature is effective. The proposed nonlinear signal equalization method allows the fault signal to be evenly spread out in two-dimensional coordinates. The experimental results intuitively show that there is a clear stratification between different faults, and the threshold of discrimination is improved, indicating that the present invention has high robustness. In actual production, when a fan stops working, maintenance personnel can use the present invention to determine the fault location more quickly, which is of great significance for reducing troubleshooting time and increasing production efficiency in engineering. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0059] Figure 1 A method flow chart of a single-sample wind turbine transmission chain fault diagnosis method according to an embodiment of the present invention;

[0060] Figure 2 It is a schematic diagram of the main monitoring bearing position of the fan in the example of the present invention;

[0061] Figure 3 It is a time domain diagram of fault signal data in an example of the present invention;

[0062] Figure 4 It is a schematic diagram of the energy values ​​of each mode decomposed by EMD in an example of the present invention;

[0063] Figure 5 It is a schematic diagram of each modal peak value of EMD decomposition in an example of the present invention;

[0064] Figure 6 A schematic diagram of the variance values ​​of each frequency band of the fault signal in an example of the present invention;

[0065] Figure 7 A schematic diagram of the variance of each frequency band after equalization in an example of the present invention;

[0066] Figure 8 This is a schematic diagram of the variance of the 5-8 frequency band after equalization in an example of the present invention;

[0067] Fig. 9Schematic diagram of improving the AlxeNet network training accuracy in an example of the present invention. DETAILED DESCRIPTION

[0068] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0069] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0070] Embodiment 1

[0071] like Figure 1 , which shows a method flow chart of a single-sample wind turbine transmission chain fault diagnosis method provided by an embodiment of the present invention, the method comprising the steps of:

[0072] S1: original fault data acquisition and cleaning;

[0073] S2: Decompose the original fault data through empirical mode decomposition and reconstruct it according to preset requirements to obtain new fault data;

[0074] S3: Perform wavelet packet decomposition on the new fault data to extract the variance as the characteristic value of fault diagnosis;

[0075] S4: Perform balanced processing on the characteristic values ​​to achieve fault classification.

[0076] As an optimization scheme of the embodiment, in step S1, the fan fault data comes from the SCADA data monitoring system. By installing multiple vibration sensors on the entire transmission chain of the fan, the operation of the fan can be monitored. The setting of the fan status monitoring point directly affects the correctness of the monitoring data and the accuracy of fault diagnosis. Generally, the installation position of the vibration sensor should have the characteristics of easy installation and replacement, appropriate number, and strong anti-interference ability. According to these requirements, the vibration sensor is generally installed near the fixed bearing. The position of the bearing required to be monitored in the entire transmission chain of the fan is as follows: Figure 2 shown.

[0077] Due to the particularity of the actual operation of the wind turbine and the influence of various factors such as wind direction and wind speed, the wind turbine in the actual wind farm will not work continuously. Therefore, the data collected from the wind farm monitoring terminal will contain a large amount of invalid data. It is necessary to obtain the downtime of the faulty wind turbine in the wind farm from the wind farm monitoring terminal. It is assumed that there is a fault in the internal structure of the shut down fan, and there may be fault information in the sensor vibration data before the shutdown. Then the maintenance personnel actually check and troubleshoot to find the specific fault location and cause of the wind turbine shutdown. The present invention collected a total of 4 causes of wind turbine shutdown from each wind farm. These 4 faults involve different positions of the wind turbine gearbox and generator. Since the vibration sensors installed at different measuring points have different sensitivities to faults at different positions, when selecting fault data, the appropriate data channel should be selected according to the measuring point to improve the accuracy and matching of the fault data. For the convenience of subsequent expression, each fault is numbered in the order of 1-4. The data channels corresponding to specific fault types are shown in Table 1:

[0078] Table 1 Data channels corresponding to each fault

[0079] Abbreviation Fault type Data Channel Type1 Pitting corrosion on the inner ring of the gearbox high-speed shaft bearing CH6 Type2 Gearbox intermediate stage bearing broken teeth CH5 Type3 Mid-term failure of the gearbox high-speed shaft bearing CH6 Type4 Mid-term wear of generator rear bearing CH8

[0080] It can be seen from Table 1 that the faults are distributed in different locations of the wind turbine. The diversity of faults increases the difficulty of wind turbine troubleshooting, so it is particularly necessary to classify these types of faults. According to the actual fault type and the corresponding data channel, the time domain fault vibration signal before the fault is obtained. By comparing and analyzing the fault vibration signal before the shutdown, the fault data with the most obvious fault characteristics are selected. The 4 groups of fault signal time domain data are as follows Figure 3 As shown, the horizontal axis is the time sampling point and the vertical axis is the signal amplitude.

[0081] As an optimization scheme of the embodiment, in step S2, the EMD decomposition method decomposes the original fault data, divides the original signal into n inherent modes and residuals, and selects the amount that is conducive to fault classification when reconstructing the signal, and removes the modes with few low-frequency fault signals. The fault data is filtered from the two dimensions of peak value and energy, and the energy values ​​of the signal components under various modes of EMD decomposition and the peak index of each modal component of EMD decomposition are calculated, and the components with higher energy and larger peak values ​​are selected for reconstruction to obtain a new fault signal.

[0082] The empirical mode decomposition of the fan fault is carried out, taking the pitting corrosion of the inner ring of the high-speed gear bearing of the gearbox as an example.

[0083] According to the EMD decomposition results, it can be seen It contains high-frequency parts and has periodic impacts. Other high-frequency signals are mainly concentrated in the first few natural modes, and the low-frequency parts behind them are basically straight. Therefore, when reconstructing the signal, it is necessary to select the amount that is conducive to fault classification and remove the low-frequency modes with few fault signals.

[0084] According to the above results, the energy values ​​of the signal components under various modes of EMD decomposition are calculated. The calculation formula is as follows:

[0085] ,

[0086] In the formula, Indicates time, represents the time domain signal value, is the energy value of the time domain signal;

[0087] The calculation results are as follows Figure 4 As shown in the figure, the components with higher energy values ​​are concentrated in the first few components. The components with lower energy values ​​in the latter parts can be discarded, and only the components with higher energy are selected for reconstruction.

[0088] Peak Index Indicates signal The steepness of the peak of the probability density function reflects the magnitude of the impact component in the signal waveform and is defined as:

[0089] ,

[0090] In the formula, The time variable is The time domain signal value, is the signal standard deviation, is the signal mean, Represents the entire time domain value of the signal, Represents the signal kurtosis value. By calculation, the peak index of each modal component of EMD decomposition can be obtained. The results are as follows Figure 5 As shown in Figure 2, the peak values ​​of each modal component fluctuate greatly, and the fault characteristic signals are mostly hidden in the peaks with larger peak values, so the components with larger peak values ​​can be selected for reconstruction.

[0091] The four types of fault data of the wind turbine are decomposed by EMD respectively, and the three natural mode components with the largest modal component energy value and peak value are selected for reconstruction according to the above analysis results to obtain new fault signals respectively.

[0092] As an optimization scheme of the embodiment, in step S3, Daubechies wavelet is selected as the wavelet basis in wavelet packet analysis. The number of wavelet packet decomposition layers is directly related to the frequency band division of the signal. When the number of wavelet packet decomposition layers is too low, the range of the signal after frequency band division is large, and the information of each frequency band cannot be decomposed. Some features will be covered up, resulting in the extracted features not being obvious or even not being extracted. However, when the number of wavelet packet decomposition layers is too high, although the frequency bands of the signal division are more detailed, it will cause the later feature dimension to be too large. In addition, it will also separate the rich information contained in the unique frequency band, resulting in the extracted features not being obvious. Therefore, the selection of the number of wavelet packet decomposition layers should be determined according to the characteristics of the actual signal. According to the characteristics of the bearing vibration signal, the optimal number of wavelet packet decomposition layers should not exceed 5. Perform 3-layer wavelet packet decomposition on the fault signal as required.

[0093] After the fault data is decomposed by wavelet packets, the variance is calculated segment by segment to obtain the eigenvalue. The variance calculation formula is:

[0094] ,

[0095] In the formula, the data is For the overall A sample of is the number of sample individuals, is the sample mean.

[0096] The variance values ​​of different faults in each frequency band are as follows: Figure 6 As shown in the figure. It is observed that the variance eigenvalue of fault 4 is much larger than other characteristic signals, and other faults are clearly stratified compared with each other, indicating that the method is meaningful for the classification of fault 4. However, the eigenvalues ​​of the other three faults are basically submerged in the figure, and the characteristic data are not displayed at all. Therefore, it is impossible to clearly classify the four faults, so optimization processing is still needed.

[0097] As an optimization scheme for the embodiment, in step S4, the variances generated by the characteristic signal in different frequency bands are unevenly distributed, showing an uneven variance distribution and a large dynamic range. Smaller variances are easily submerged, which is not conducive to the classification of fan faults. In order to effectively extract the characteristics of different fault signals, the present invention proposes a nonlinear equalization method for the frequency band components obtained by wavelet packet decomposition of the bearing fault signal, so that the fault signal is evenly expanded in the two-dimensional coordinates, so that the fault characteristics have a higher degree of distinction. The purpose is to amplify the values ​​with small variances and reduce the values ​​with large variances. Different fault signals can be divided into specific ranges in two-dimensional space for comparison.

[0098] Assume that the fault signal variance is Expression, control factor express, is the base of the logarithm and the function result is The nonlinear equilibrium function is:

[0099] ,

[0100] After verification, this experiment sets the control factor is 2.

[0101] The variance distribution diagram after each fault is balanced is as follows: Figure 7 As shown. Figure 7 It can be seen that the four fault signals are spread out on the coordinate axis, and the signals will not drown each other. However, there is still overlap between the faults, especially in the low-frequency part, so we discard this part and take the high-frequency part as the research object. Select the variance value of the 5-8 frequency band, such as Figure 8 shown.

[0102] By observation Figure 8 The experimental results show that the fault characteristic values ​​have very good discrimination in the high frequency part, especially in the 6-7 frequency band. Therefore, this frequency band can be used as the standard for fault classification, which also shows that the experimental method proposed in this paper is effective for bearing fault diagnosis at four different positions of the fan.

[0103] By introducing the concept of discrimination, the distinction between different bearing status signals can be better described. In the original data, a and b are used to represent different bearing statuses, and the discrimination d is defined as:

[0104] ,

[0105] In the formula, Indicates a fault The minimum value of the signal variance, Indicates a fault The maximum value of the signal variance.

[0106] In order to distinguish and measure the processed signal, the discrimination D is established as:

[0107] ,

[0108] Discrimination is used to describe whether the data is balanced. When the discrimination increases to a certain extent, the data will be balanced. At this time, the data polarization phenomenon will be weakened, the data with larger amplitude will be relatively reduced, and the data with smaller amplitude will be relatively enlarged, and the data will reach a stable state. By calculating the discrimination of frequency band 6 before and after fault processing, the relationship between each fault is quantitatively analyzed. The calculation results of the discrimination between each fault in the original data are shown in Table 2.

[0109] Table 2 Discrimination between various faults in the original data

[0110] Type1 Type2 Type3 Type4 Type1 \ * * * Type2 2.8849e-04 \ * * Type3 0.0295 0.0275 \ * Type4 2.0930 2.0910 1.0252 \

[0111] The calculation results of the discrimination between the faults after equalization are shown in Table 3.

[0112] Table 3 Distinguishing degree between faults after equalization

[0113] Type1 Type2 Type3 Type4 Type1 \ * * 8 Type2 3.8561 \ * * Type3 19.2960 11.9759 \ * Type4 15.1328 7.8127 4.3539 \

[0114] The changes in the distinction between various faults in the original data and the data after equalization are shown in Table 4.

[0115] Table 4 Differences in fault discrimination before and after equalization

[0116] Type1 Type2 Type3 Type4 Type1 \ * * * Type2 3.8558 \ * * Type3 19.2665 11.9484 \ * Type4 13.0398 5.7217 3.3287 \

[0117] The calculation results show that the difference between the fault signals after nonlinear equalization is obviously positive compared with before the processing, which indicates that the fault distinction increases after equalization, which meets the expected experimental goals.

[0118] In order to further verify the experimental results proposed by the present invention, a comparative experiment is added to analyze the experimental results. Firstly, the wind turbine transmission chain fault signal is processed by improved fuzzy clustering.

[0119] Then, the fault diagnosis and analysis of the wind turbine transmission chain fault is carried out by improving the AlxeNet network. The fault signal in the time domain is converted into a two-dimensional image by the Markov transfer field method. 1500 data are selected from 1880 data sets as training sets and 380 data are selected as test sets. The model training accuracy is as follows Fig. 9 shown.

[0120] From the training results, we can see that the maximum training accuracy of the model is 94.92%, and the maximum verification accuracy is 72.70%. This shows that the multi-sample model is not effective in diagnosing faults in the wind turbine transmission chain. Therefore, the fault diagnosis method proposed in the present invention realizes the display of obvious distinction between faults, and has obvious advantages over other methods in processing wind turbine transmission chain fault signals.

[0121] Embodiment 2

[0122] The present invention also provides a single-sample wind turbine transmission chain fault diagnosis system, comprising: an acquisition module, a reconstruction module, an extraction module and a classification module;

[0123] The acquisition module is used to obtain original fault data;

[0124] The reconstruction module is used to decompose the original fault data through empirical mode decomposition, and reconstruct it according to preset requirements to obtain new fault data;

[0125] The extraction module is used to perform wavelet packet decomposition on new fault data to extract variance as the characteristic value of fault diagnosis;

[0126] The classification module is used to perform balanced processing on the characteristic values ​​to achieve fault classification.

[0127] In this embodiment, the reconstruction module includes: a first decomposition unit, a first calculation unit and a first reconstruction unit;

[0128] The first decomposition unit is used to perform empirical mode decomposition on the original fault data to obtain signal components of different modes;

[0129] The first calculation unit is used to calculate the energy value and peak value of each modal signal component;

[0130] The first reconstruction unit is used to select the modal signal component whose energy value and peak value meet the preset requirements to perform signal reconstruction to obtain new fault data;

[0131] Among them, the expression for calculating the energy value of each modal signal component is:

[0132] ;

[0133] In the formula, Indicates time, represents the time domain signal value, is the energy value of the time domain signal;

[0134] The expression for calculating the peak value of each modal signal component is:

[0135] ;

[0136] In the formula, The time variable is The time domain signal value, is the signal standard deviation, is the signal mean, Represents the entire time domain value of the signal, Indicates the signal kurtosis value.

[0137] In an embodiment, the extraction module comprises: a second decomposition unit, a second reconstruction unit, and a second calculation unit;

[0138] The second decomposition unit is used to perform 3-layer wavelet packet decomposition on the new fault data;

[0139] The second reconstruction unit is used for reconstructing the third layer component of the wavelet packet decomposition;

[0140] The second calculation unit is used to use the variance of the reconstructed signal as a characteristic value for fault diagnosis;

[0141] Among them, the variance calculation formula of the reconstructed signal is:

[0142] ,

[0143] In the formula, the data is For the overall A sample of is the number of sample individuals, is the sample mean.

[0144] In this embodiment, in the classification module, the process of performing equalization processing on the characteristic values ​​to achieve fault classification includes:

[0145] ,

[0146] in, represents the fault signal variance, represents the regulatory factor, is the base of the logarithm, Indicates the signal value after equalization.

[0147] The present invention uses only a single sample to diagnose the fault of the wind turbine transmission chain, which is in line with the low failure rate of the wind turbine. In actual production, when a wind turbine fails to shut down, the present invention can determine the fault location more quickly, which is of great significance for reducing troubleshooting time and increasing production efficiency in engineering.

[0148] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A single-sample wind turbine transmission chain fault diagnosis method, characterized in that: The following steps are involved: S1: Get original fault data; S2: Decompose the original fault data through empirical mode decomposition, and reconstruct it according to preset requirements to obtain new fault data; S3: Perform wavelet packet decomposition on the new fault data to extract the variance as the characteristic value of fault diagnosis; S4: Perform balanced processing on the characteristic values ​​to achieve fault classification.

2. The single-sample wind turbine transmission chain fault diagnosis method according to claim 1, characterized in that: In S2, the method of decomposing the original fault data by empirical mode decomposition and reconstructing the original fault data according to preset requirements to obtain new fault data includes: Perform empirical mode decomposition on the original fault data to obtain signal components of different modes; Calculate the energy value and peak value of each modal signal component; Select the modal signal components whose energy values ​​and peak values ​​meet the preset requirements to perform signal reconstruction to obtain new fault data; Among them, the expression for calculating the energy value of each modal signal component is: ； In the formula, Indicates time, represents the time domain signal value, is the energy value of the time domain signal; The expression for calculating the peak value of each modal signal component is: ； In the formula, The time variable is The time domain signal value of is the signal standard deviation, is the signal mean, Represents the entire time domain value of the signal, Indicates the signal kurtosis value.

3. The single-sample wind turbine transmission chain fault diagnosis method according to claim 1, characterized in that: In S3, the method of performing wavelet packet decomposition on the new fault data and extracting the variance as the characteristic value of the fault diagnosis includes: Perform 3-layer wavelet packet decomposition on the new fault data; Reconstruct the signal by using the third layer component of wavelet packet decomposition; The variance of the reconstructed signal is used as the characteristic value of fault diagnosis; Among them, the variance calculation formula of the reconstructed signal is: ， In the formula, the data is For the overall A sample of is the number of sample individuals, is the sample mean.

4. The single-sample wind turbine transmission chain fault diagnosis method according to claim 1, characterized in that: In S4, the method of performing equalization processing on the characteristic values ​​to achieve fault classification includes: ， in, represents the fault signal variance, represents the regulatory factor, is the base of the logarithm, Indicates the signal value after equalization.

5. A single-sample wind turbine transmission chain fault diagnosis system, characterized in that: include: Acquisition module, reconstruction module, extraction module and classification module; The acquisition module is used to acquire original fault data; The reconstruction module is used to decompose the original fault data through empirical mode decomposition, and reconstruct it according to preset requirements to obtain new fault data; The extraction module is used to perform wavelet packet decomposition on new fault data to extract variance as a characteristic value for fault diagnosis; The classification module is used to perform balanced processing on the characteristic values ​​to achieve fault classification.

6. The single-sample wind turbine transmission chain fault diagnosis system according to claim 5, characterized in that: The reconstruction module includes: a first decomposition unit, a first calculation unit and a first reconstruction unit; The first decomposition unit is used to perform empirical mode decomposition on the original fault data to obtain signal components of different modes; The first calculation unit is used to calculate the energy value and peak value of each modal signal component; The first reconstruction unit is used to select a modal signal component whose energy value and peak value meet preset requirements to perform signal reconstruction to obtain new fault data; Among them, the expression for calculating the energy value of each modal signal component is: ； In the formula, Indicates time, represents the time domain signal value, is the energy value of the time domain signal; The expression for calculating the peak value of each modal signal component is: ； In the formula, The time variable is The time domain signal value of is the signal standard deviation, is the signal mean, Represents the entire time domain value of the signal, Indicates the signal kurtosis value.

7. The single-sample wind turbine transmission chain fault diagnosis system according to claim 5, characterized in that: The extraction module includes: a second decomposition unit, a second reconstruction unit and a second calculation unit; The second decomposition unit is used to perform 3-layer wavelet packet decomposition on the new fault data; The second reconstruction unit is used to reconstruct the third layer component of the wavelet packet decomposition; The second calculation unit is used to use the variance of the reconstructed signal as a characteristic value for fault diagnosis; Among them, the variance calculation formula of the reconstructed signal is: ， In the formula, the data is For the overall A sample of is the number of sample individuals, is the sample mean.

8. The single-sample wind turbine transmission chain fault diagnosis system according to claim 5, characterized in that: In the classification module, the process of performing balanced processing on the characteristic values ​​to achieve fault classification includes: ， in, represents the fault signal variance, represents the regulatory factor, is the base of the logarithm, Indicates the signal value after equalization.

Images