A two-stage early fault detection method, system, device and storage medium for a wind turbine generator

Abstract

This invention relates to the field of early fault diagnosis technology for wind turbines, and discloses a two-level diagnostic method, system, equipment, and storage medium for early faults in wind turbines. The two-level diagnostic method for early faults in wind turbines includes the following steps: multi-source data acquisition, vibration signal processing, feature extraction, comprehensive weight fusion and evaluation value calculation, fault threshold determination and diagnostic model establishment, and two-level fault diagnosis and judgment. In early fault diagnosis of wind turbines, it can avoid weak features being masked by noise, accurately perceive changes in signal disorder of early faults, and achieve accurate identification of early faults. By adopting a two-level fault diagnosis and judgment mechanism, an early fault is only determined when the logarithmic evaluation value is greater than the fault threshold and the root mean square value exceeds the corresponding fault threshold for a consecutive preset number of sampling points. This effectively filters instantaneous data fluctuations and effectively avoids false alarms under complex operating conditions, resulting in accurate and reliable diagnostic results.

Description

Technical Field

[0001] This invention belongs to the field of early fault diagnosis technology for wind turbines. Specifically, it relates to a two-level diagnosis method, system, equipment, and storage medium for early faults in wind turbines. Background Technology

[0002] As a clean and renewable energy source, the efficient development and utilization of wind energy has become a focus for various countries and energy companies. As its core component, the wind turbine, with its main shaft, gearbox, generator, and other key components, operates in a complex environment for a long time and inevitably experiences failures. Early failures are often subtle and difficult to identify, but once a serious failure occurs, it can lead to prolonged downtime, resulting in huge economic losses and, in severe cases, safety accidents. Therefore, it is necessary to diagnose and identify the early subtle failure characteristics of key components to guide scientific maintenance.

[0003] For early faults such as pitting, peeling, and corrosion of key components of wind turbines, there are already corresponding early fault diagnosis technologies in the existing technology. For example, the wind turbine gearbox fault diagnosis method, device and computer equipment disclosed in Chinese Patent No. CN120354058B, which realizes gearbox health status assessment and fault type identification by multi-source data fusion combined with GRU health model. However, existing diagnostic methods still have some shortcomings. For example, triggering an alarm solely based on a single health score falling below a fault threshold does not consider false alarms caused by data fluctuations, making it prone to false alarms due to sudden changes in operating conditions or signal interference, thus affecting the reliability of fault diagnosis. Furthermore, the model's robustness is insufficient when using only the entropy weight method to calculate objective weights. When operating conditions change abruptly or data distribution is abnormal, the subtle characteristics of early faults may be underestimated due to data fluctuations, resulting in low diagnostic accuracy under complex operating conditions. In addition, it is limited to features such as root mean square (RMS) and kurtosis, ignoring the entropy value, which captures subtle dynamic linear changes in signals, making it impossible to accurately perceive changes in signal disorder in early faults and hindering accurate identification of early faults. Summary of the Invention

[0004] To address or partially address the problems existing in related technologies, this application provides a two-level diagnostic method, system, device, and storage medium for early faults in wind turbines. This two-level diagnostic method for early faults, which involves multi-source data acquisition, vibration signal processing, feature extraction, comprehensive weight fusion and evaluation value calculation, fault threshold determination and diagnostic model establishment, and two-level fault diagnosis, can prevent weak features from being masked by noise, achieves high diagnostic accuracy under complex operating conditions, accurately detects changes in the signal disorder of early faults, realizes accurate identification of early faults, effectively avoids false alarms under complex operating conditions, and provides accurate and reliable diagnostic results.

[0005] The first aspect of this application provides a two-level diagnostic method for early faults in wind turbine generators, including the following steps: S1: Multi-source data acquisition: Collects multi-source data from the wind turbine, including operating parameters and vibration data of key components. The operating parameters and vibration data are synchronized and aligned using timestamps. S2: Vibration signal processing: The vibration data is sequentially subjected to VMD algorithm noise reduction processing and mathematical morphology combined morphological top-hat transformation signal enhancement processing. The combined morphological top-hat transformation is used to enhance positive and negative transient impacts. S3: Feature Extraction: Extract the time-domain and frequency-domain features of the processed vibration data. The time-domain features include root mean square value, kurtosis, and sample entropy. The frequency-domain features include whether a fault has occurred and the type of fault. Extract changes in operating parameters. S4: Comprehensive weight fusion and evaluation value calculation: The time domain features are normalized, the objective weight is calculated using the entropy weight method, and the comprehensive weight is obtained by combining the subjective weight of experts. Based on the comprehensive weight, the time domain features, frequency domain features, and actual working conditions are weighted and fused to calculate the final evaluation value and logarithmic evaluation value. S4.1: Normalization of time-domain features: positive indices are applied to root mean square value and kurtosis, while negative indices are applied to sample entropy; Among them, positive indicators: ; Negative indicators: ; S4.2: Calculate the entropy value: First calculate the proportion, then calculate the information entropy; Of which, the proportion is: ; Information entropy: ; S4.3: Weighting: First calculate the difference coefficient, then calculate the objective weight, and then use the subjective weight given by the experts, combined with the multiplication integration method to obtain the comprehensive weight; Among them, the coefficient of difference: ; Objective weighting: ; Subjective weighting: ; Overall weighting: ; S4.4: Calculate the final evaluation value S: ; S4.5: Calculate the logarithmic evaluation value: Perform a constant logarithmic transformation on the final evaluation value S to obtain the logarithmic evaluation value: log 10 (S); S5: Fault threshold determination and diagnostic model establishment: Based on historical data of wind turbine units, determine the fault threshold of logarithmic evaluation value and the fault threshold of root mean square value, and establish an early fault diagnosis model. S6: Two-level fault diagnosis and judgment: The early fault diagnosis model is used to perform two-level diagnosis. First, the first-level judgment is made based on the logarithmic evaluation value. If the logarithmic evaluation value is greater than the fault threshold, the second-level judgment is initiated. If the root mean square value exceeds the corresponding fault threshold for a consecutive preset number of sampling points, it is judged as an early fault.

[0006] In one optional scheme, in step S1, the operating parameters are the wind turbine status parameters collected by the SCADA system, including the main shaft speed, generator speed, generator real-time power, wind speed, pitch angle, 30-second average gear oil temperature, 30-second average nacelle temperature, and 30-second average generator speed, with a collection interval of 1 hour.

[0007] In one alternative, in step S1, the key components include the first-stage internal gear ring of the gearbox, the second-stage internal gear ring of the gearbox, the high-speed shaft of the gearbox, the front bearing of the generator, and the rear bearing of the generator. The vibration data includes radial vibration data of each key component collected by vibration sensors.

[0008] In one alternative approach, in step S2, the VMD algorithm is first used to denoise the vibration data, as shown in the following formula: ; In the formula, , ; The noise-reduced vibration data is then subjected to a mathematical morphological combination morphology top-hat transform signal enhancement process, as shown in the following formula: ; ; ; ; ; In the formula, f is the input vibration signal, γ(f) is the result of morphological opening operation, φ(f) is the result of morphological closing operation, ε(f) is the erosion operation, δ(f) is the dilation operation, and B is the structuring element.

[0009] In one alternative, in step S3, the frequency domain features are used to generate an amplitude-frequency diagram through fast Fourier transform, and the fault type is determined based on the peak value of the characteristic frequency. The fault types include peeling, wear, corrosion, and significant impact. Temporal features include root mean square value, kurtosis, and sample entropy, calculated using the following formulas: Root mean square value: ; kurtosis: ; Sample entropy: ; In the formula, ; ; ; Where m is the embedding dimension, m=2; r is the similarity tolerance, r=0.2σ, σ is the signal standard deviation; N is the signal length; and Θ is the Heaviside function.

[0010] In one alternative approach, in step S6, the preset number of sampling points is 5. When the logarithmic evaluation value is greater than the fault threshold and the root mean square value exceeds the corresponding fault threshold for five consecutive sampling points, it is determined to be an early fault.

[0011] In one alternative approach, the two-level diagnostic method for early faults in wind turbines also includes a model optimization step: verifying the model accuracy based on historical diagnostic data and dynamically updating the thresholds and overall weights for each characteristic fault.

[0012] The second aspect of this application provides a two-level diagnostic system for early faults in wind turbine generators, which performs the aforementioned two-level diagnostic method for early faults in wind turbine generators. The system includes: The data acquisition module is used to collect multi-source data from the wind turbine, including operating parameters and vibration data of key components. The operating parameters and vibration data are synchronized and aligned using timestamps. The data acquisition module includes vibration sensors and a SCADA system. The vibration sensors are deployed at the positions of the first-stage internal gear ring, the second-stage internal gear ring, the high-speed shaft of the gearbox, the front bearing of the generator, and the rear bearing of the generator to collect radial vibration data of each key component. The data processing module is used to sequentially perform VMD algorithm noise reduction processing and mathematical morphology combined morphology top-hat transformation signal enhancement processing on the vibration data. The combined morphology top-hat transformation is used to enhance positive and negative transient impacts. The feature extraction module is used to extract the time-domain and frequency-domain features of the processed vibration data. The time-domain features include root mean square value, kurtosis, and sample entropy, while the frequency-domain features include whether a fault has occurred and the type of fault. It also extracts changes in operating parameters. The feature fusion module is used to normalize the time-domain features, calculate the objective weights using the entropy weight method, and combine them with the subjective weights of experts to obtain the comprehensive weights. Based on the comprehensive weights, the time-domain features, frequency-domain features, and actual working conditions are weighted and fused to obtain the final evaluation value and the logarithmic evaluation value. The model building module is used to determine the fault thresholds of logarithmic evaluation values ​​and root mean square values ​​based on historical data of wind turbines, and to establish an early fault diagnosis model. The two-level diagnostic module is used to perform two-level diagnosis using an early fault diagnosis model. First, a first-level judgment is made based on the logarithmic evaluation value. If the logarithmic evaluation value is greater than the fault threshold, the second-level judgment is initiated. If the root mean square value exceeds the corresponding fault threshold for a consecutive preset number of sampling points, it is determined to be an early fault.

[0013] In one alternative, the two-level diagnostic system for early faults of wind turbines also includes a model optimization module, which is used to verify the accuracy of the model based on historical diagnostic data and dynamically update the thresholds and comprehensive weights of each characteristic fault.

[0014] A third aspect of this application provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the two-level diagnostic method for early faults of wind turbines as described above.

[0015] The fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned two-level diagnostic method for early faults in wind turbine generators.

[0016] The beneficial effects of this application are: The wind turbine early fault diagnosis method of this application sequentially applies VMD algorithm noise reduction and combined morphological top-hat transformation signal enhancement processing to vibration data. This removes noise while enhancing fault characteristics, effectively increasing the pulse characteristic amplitude of early faults, improving the signal-to-noise ratio, and enhancing the robustness of the model. This ensures that the subsequently extracted features reflect the true state of the equipment, while highlighting the weak features of early faults, thus preventing these features from being masked by noise. This comprehensive approach captures fault pulses and effectively improves the diagnostic accuracy of early faults under complex operating conditions. The method uses the root mean square (RMS) value to represent the weak impact on the wind turbine, and extracts the sample entropy of the time-domain signal to represent the weak dynamic linear changes of the signal. The combination of RMS, kurtosis, and entropy values ​​allows for more accurate capture of fault pulses, enabling precise perception of signal disorder changes in early faults and accurate identification of early faults. Finally, a comprehensive weight is obtained by fusing the objective weights of RMS, kurtosis, and sample entropy with the subjective weights of expert experience, resulting in the final evaluation value S and the logarithmic evaluation value log. 10(S) leverages both data objectivity and expert experience to comprehensively reflect the operating status of wind turbine units, fully considering false alarms caused by data fluctuations, and effectively improving the reliability of fault diagnosis. By adopting a two-level fault diagnosis and judgment mechanism, an early fault is only judged when the logarithmic evaluation value is greater than the fault threshold and the root mean square value exceeds the corresponding fault threshold for a consecutive preset number of sampling points. This effectively filters instantaneous data fluctuations and can effectively distinguish between "instantaneous exceeding of the fault threshold caused by data fluctuations" and "continuous exceeding of the fault threshold caused by real faults." This avoids the high false alarm rate problem of existing technologies that trigger alarms based solely on a single health score falling below the fault threshold, effectively preventing false alarms under complex operating conditions, and ensuring accurate and reliable diagnostic results.

[0017] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the two-level diagnostic method for early faults in wind turbines presented in this application. Figure 2 This is a schematic diagram of a sub-process of the two-level diagnosis method for early faults of wind turbines in this application; Figure 3 This is the two-level fault diagnosis and judgment logic diagram of the two-level early fault diagnosis method for wind turbines in this application; Figure 4 This is a structural block diagram of the two-level diagnostic system for early faults in wind turbines in this application; Figure 5 This is a fault characteristic trend diagram of the two-level diagnosis method for early faults of wind turbines in this application; Figure 6 It is a fault characteristic trend diagram of a diagnostic method that combines subjective weights and objective weights with single verification; Figure 7 It is a fault characteristic trend chart based on a diagnostic method that combines objective weights with dual verification. Detailed Implementation

[0020] The specific embodiments of this application will be further described in detail below with reference to the accompanying drawings and examples. The following examples are used to illustrate this application, but are not intended to limit the scope of this application. Similarly, the following examples are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection, an electrical connection, or a connection that allows communication between them; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0022] In this invention, the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to a specific feature, structure, material, or characteristic described in connection with that embodiment or example, which is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0023] Existing methods for early fault diagnosis of wind turbines still have some shortcomings. For example, triggering an alarm solely based on a single health score falling below a fault threshold does not consider false alarms caused by data fluctuations. This makes them prone to false alarms due to sudden changes in operating conditions and signal interference, affecting the reliability of fault diagnosis. Using only the entropy weight method to calculate objective weights results in insufficient robustness of the model. When operating conditions change abruptly or data distribution is abnormal, the subtle characteristics of early faults may be underestimated due to data fluctuations, leading to low diagnostic accuracy under complex operating conditions. Furthermore, methods are limited to features such as root mean square value and kurtosis, neglecting the entropy value, which captures subtle dynamic linear changes in signals. This makes it difficult to accurately perceive changes in signal disorder during early faults and achieve accurate identification of early faults.

[0024] To address the aforementioned problems, this application proposes improvements and innovations, including the following embodiments.

[0025] Example 1: Please see Figure 1 , Figure 2 , Figure 3 , Figure 4 , Figure 5 This embodiment provides a two-level diagnostic method for early faults in wind turbine generators, including the following steps: S1: Multi-source data acquisition: Collects multi-source data from the wind turbine, including operating parameters and vibration data of key components. The operating parameters and vibration data are synchronized and aligned using timestamps.

[0026] Specifically, vibration sensors are deployed at key component locations to collect radial vibration data for each key component, including the first-stage internal gear ring of the gearbox, the second-stage internal gear ring of the gearbox, the high-speed shaft of the gearbox, the front bearing of the generator, and the rear bearing of the generator. The operating parameters are the status parameters of the wind turbine collected by the wind turbine's built-in SCADA system, i.e., the data acquisition and monitoring control system, including the main shaft speed, generator speed, generator real-time power, wind speed, pitch angle, 30-second average gear oil temperature, 30-second average nacelle temperature, and 30-second average generator speed. The collection interval is 1 hour, and the operating parameters and vibration data are marked with the same timestamp to ensure data synchronization.

[0027] Thus, by collecting operating parameters through the SCADA system, the real-time nature and completeness of operating data can be ensured, providing reliable data for subsequent operating condition fusion. By deploying vibration sensors at key component locations to collect radial vibration data, the fault characteristics of key components can be captured in a targeted manner, which can effectively improve the pertinence and accuracy of fault diagnosis. The multi-source data collected in step S1 covers operating parameters and vibration signals, avoiding the limitations of a single data source, and can comprehensively and accurately reflect the operating status of the wind turbine.

[0028] S2: Vibration signal processing: The vibration data is sequentially subjected to VMD algorithm noise reduction processing and mathematical morphology combined morphological top-hat transformation signal enhancement processing. The combined morphological top-hat transformation is used to enhance positive and negative transient impacts.

[0029] Specifically, the vibration data is first denoised using the VMD algorithm in Python's VMD library. By iteratively updating the modal functions and center frequencies, noise and fault components are separated, as shown in the following formula: ; Among them, u k f(t) is the k-th intrinsic mode function (BLIMFs), f(t) is the original vibration signal, δ(t) is the Dirac function, ∗ is the convolution operation, and j is the imaginary unit.

[0030] In the formula, , ; Among them, for the update mode Each mode update considers the residual of the original signal minus all other modes, with accuracy guaranteed by Lagrange multipliers and adjusted by the bandwidth constraint term 1+2α(ω−ωk)2, where α is the bandwidth penalty parameter; the larger the α, the narrower the mode bandwidth. For updating the center frequency... That is, the centroid of the current power spectrum.

[0031] The goal of the VMD algorithm is to decompose the original signal f(t) into K eigenmode functions u. k (t) minimizes the sum of estimated bandwidths for each mode, and the sum of all modes equals the original signal. Using K and α, background noise, shaft rotation frequency, and fault characteristic frequencies can be clearly separated into different modes, allowing direct extraction of modes containing fault information for analysis with an extremely high signal-to-noise ratio.

[0032] The noise-reduced vibration data is then subjected to a mathematical morphological combination morphology top-hat transform signal enhancement process, as shown in the following formula: ; ; ; ; ; In the formula, f is the input vibration signal, γ(f) is the result of morphological opening operation, φ(f) is the result of morphological closing operation, ε(f) is the erosion operation, δ(f) is the dilation operation, and B is the structuring element.

[0033] By using a combined morphological top-hat transform operator based on mathematical morphology to enhance the signal, all minute details in the image can be enhanced simultaneously. Specifically, it can effectively enhance positive transient impacts, such as the impact signal of gear peeling, and negative transient impacts, such as the dent signal of bearing corrosion. This avoids situations where other components are not captured due to a single excessively large impact, thus enabling comprehensive capture of fault pulses. The combined morphological top-hat transform algorithm is a mature existing algorithm and will not be described in detail in this specification.

[0034] Thus, by sequentially performing VMD algorithm noise reduction processing and combined morphological top-hat transformation signal enhancement processing on the vibration data, noise is removed while fault characteristics are enhanced. This effectively increases the pulse characteristic amplitude of early faults, improves the signal-to-noise ratio, enhances the robustness of the model, ensures that the features extracted subsequently reflect the true state of the equipment, and highlights the weak features of early faults. This prevents weak features from being masked by noise, enables comprehensive capture of fault pulses, and effectively improves the diagnostic accuracy of early faults under complex working conditions.

[0035] S3: Feature Extraction: Extract the time-domain and frequency-domain features of the processed vibration data. The time-domain features include root mean square value, kurtosis, and sample entropy. The frequency-domain features include whether a fault has occurred and the type of fault. Extract changes in operating parameters.

[0036] Specifically, the features corresponding to the frequency domain signal are used to generate an amplitude-frequency diagram through Fast Fourier Transform (FFT), and the fault type is determined based on the peak value of the characteristic frequency. The fault types include peeling, wear, corrosion, and significant impact. The features corresponding to the time-domain signal include the root mean square value, kurtosis, and sample entropy, which are calculated using the following formulas: Root mean square value: ; kurtosis: ; Sample entropy: ; In the formula, ; ; ; Where m is the embedding dimension, m=2; r is the similarity tolerance, r=0.2σ, σ is the signal standard deviation; N is the signal length; and Θ is the Heaviside function.

[0037] In this way, the root mean square value is used to represent the weak impact on the wind turbine, and the sample entropy of the time domain signal is extracted to represent the weak dynamic linear change of the signal. The combination of root mean square value, kurtosis and entropy value can more accurately capture the fault pulse, thereby accurately sensing the signal disorder changes of early faults and realizing accurate identification of early faults.

[0038] S4: Comprehensive weight fusion and evaluation value calculation: The time domain features are normalized, the objective weight is calculated using the entropy weight method, and the comprehensive weight is obtained by combining the subjective weight of experts. Based on the comprehensive weight, the time domain features, frequency domain features, and actual working conditions are weighted and fused to calculate the final evaluation value and logarithmic evaluation value.

[0039] Considering that the impact of early-stage failures is often small, and that the root mean square (RMS) value can reliably capture failures such as pitting and spalling, the RMS value, kurtosis, and sample entropy are added to the adaptive weighted model with objective weights. At the same time, expert experience is introduced as a subjective weight and added to the adaptive weighted model. The subjective weights and objective weights are combined through a multiplicative ensemble method to obtain a comprehensive weight, thereby obtaining a more comprehensive final evaluation value S, which can better capture early-stage failures.

[0040] Specifically, the root mean square value, kurtosis, and sample entropy are normalized before weighting. After normalization, the entropy weight method is used to calculate the weights based on each feature. Then, subjective weights obtained from expert suggestions are combined to obtain a real-time comprehensive weight. Finally, the final evaluation value S and the logarithmic evaluation value log are calculated. 10 (S), the calculation process is as follows: S4.1: Normalization of time-domain features: The root mean square value and kurtosis are positive indicators, with larger values ​​indicating a higher likelihood of faults; the sample entropy is a negative indicator, with smaller values ​​indicating a higher likelihood of faults. Among them, positive indicators: ; Negative indicators: ; In the formula, min(x) j ), max(x j ) represent the maximum and minimum values ​​of the features in historical health data, respectively.

[0041] S4.2: Calculate the entropy value: First calculate the proportion, then calculate the information entropy. The smaller the entropy value, the stronger the feature's ability to distinguish faults. Of which, the proportion is: ; Information entropy: ; S4.3: Weighting: First calculate the difference coefficient, then calculate the objective weight, and then use the subjective weight given by the experts, combined with the multiplication integration method to obtain the comprehensive weight; Among them, the coefficient of difference: The larger the value, the greater the degree of variation of the j-th feature, and the more information it provides.

[0042] Objective weighting: ; Subjective weighting: ; Overall weighting: ; S4.4: Calculate the final evaluation value S: The evaluation value S ranges from [0,1], and the closer it is to 1, the higher the probability of failure.

[0043] S4.5: Calculate the logarithmic evaluation value: Perform a constant logarithmic transformation on the final evaluation value S to obtain the logarithmic evaluation value: log 10 (S); the logarithmic evaluation value ranges from (−∞, 0).

[0044] Considering that early faults are relatively minor, the final evaluation value S may be too small, even as small as 0.01, affecting the accuracy of fine-tuning. Therefore, after calculating the final evaluation value S, a constant logarithmic transformation is performed on the final evaluation value S. The change in the logarithmic evaluation value reflects the change in the final evaluation value S. This can amplify the subtle fluctuations in the final evaluation value S caused by early faults, thereby effectively improving the sensitivity to early faults.

[0045] Thus, by eliminating dimensional differences through normalization, the influence of feature amplitude on weight allocation can be avoided, enhancing the scientific rigor and practicality of feature fusion. After normalization, weights are calculated based on the characteristics of each feature, resulting in real-time weighted feature fusion data of time-domain features, frequency-domain features, and actual operating conditions. A comprehensive weight based on a combination of subjective and objective weights is adopted. Weights are calculated based on the characteristics of each feature, and then subjective weights obtained from expert advice are combined to obtain a real-time comprehensive weight. That is, objective weights of each feature are calculated based on historical data, and subjective weights are assigned to each feature based on expert advice and actual conditions. Finally, the multiplicative integration method is used to combine the two to obtain the comprehensive weight. This approach utilizes data objectivity and is backed by expert experience, effectively preventing important operating parameters from being assigned too low a weight due to data limitations, thus failing to reflect their impact. The final evaluation value S integrates multi-dimensional information from the time domain, frequency domain, and operating conditions, effectively overcoming the limitations of the entropy weight method, further improving the robustness of the model, and comprehensively reflecting the operating status of the wind turbine. This effectively avoids false alarms and improves the reliability of fault diagnosis.

[0046] S5: Fault Threshold Determination and Diagnostic Model Establishment: Based on historical data of wind turbine units, determine the fault threshold of logarithmic evaluation value and the fault threshold of root mean square value, and establish an early fault diagnosis model.

[0047] Specifically, historical health and fault data of wind turbines are collected to determine the fault thresholds for the logarithmic evaluation value and the root mean square value for each feature based on the historical data, thus establishing an early fault diagnosis model. For example, 1440 sets of health data and 300 sets of fault data of wind turbines were collected over one year. Combined with early fault cases, the fault threshold for the logarithmic evaluation value was set to -1.8, and the fault threshold for the root mean square value was set to 0.39. Subsequently, the comprehensive weight calculation logic, fault thresholds, and two-level fault diagnosis judgment rules were embedded into the model to establish an early fault diagnosis model.

[0048] S6: Two-level fault diagnosis and judgment: Two-level diagnosis is performed using an early fault diagnosis model. First, based on the logarithmic evaluation value log... 10 (S) Make the first-level judgment, if the logarithmic evaluation value log 10If (S) is greater than the fault threshold, the second-level judgment is initiated. If the root mean square value exceeds the corresponding fault threshold for a preset number of consecutive sampling points, it is judged as an early fault.

[0049] Specifically, based on the data fluctuation patterns of wind turbine generators, instantaneous data fluctuations typically do not exceed 3 sampling points. Therefore, in this embodiment, the preset number of sampling points is 5 to balance sensitivity and reliability. When the logarithmic evaluation value log... 10 (S) When the value is greater than the fault threshold and the root mean square value exceeds the corresponding fault threshold for five consecutive sampling points, it is determined to be an early fault. For example... Figure 3 The two-level fault diagnosis and judgment logic diagram is shown. The first level judgment is to calculate the logarithmic evaluation value in real time. 10 (S) is compared with the fault threshold of -1.8. If log 10 If (S) ≤ -1.8, it is considered to be in a healthy state; if log 10 If (S) > -1.8, it indicates an abnormal risk, and the second-level judgment is initiated. The second-level judgment is to monitor the root mean square value in real time and compare it with the fault threshold of 0.39. If the root mean square value of 5 consecutive sampling points does not exceed 0.39, it is judged as a false alarm caused by data fluctuation and no alarm is triggered. If the root mean square value of 5 consecutive sampling points exceeds 0.39, it is judged as an early fault, and the fault information is output and an alarm is triggered at the same time.

[0050] Thus, the early fault diagnosis method for wind turbines in this application, by adopting a two-level fault diagnosis and judgment mechanism, only determines faults that occur when the logarithmic evaluation value log is simultaneously satisfied. 10 (S) When the number of consecutive preset number of sampling points with a value greater than the fault threshold and the root mean square value exceeds the corresponding fault threshold, it is determined to be an early fault. This can effectively filter instantaneous data fluctuations and effectively distinguish between "instantaneous exceeding of the fault threshold caused by data fluctuations" and "continuous exceeding of the fault threshold caused by real faults". This avoids the problem of high false alarm rate in existing technologies that trigger alarms by simply having a single health score below the fault threshold. It can effectively avoid false alarms in complex working conditions and the diagnostic results are accurate and reliable.

[0051] Please see Figure 5 , Figure 6 , Figure 7In this embodiment, a public dataset is used as an example. The bearing No. 3 in the public dataset is used for verification. It is known that bearing No. 3 has already shown early failure at point 1085 under a specific working environment. Therefore, the wind turbine early failure diagnosis method of this application is applied to verify the public dataset to determine whether the method can capture the early failure characteristics of this bearing earlier. Furthermore, on the same dataset and with the same failure threshold, it is compared with other corresponding methods. That is, the two-level early failure diagnosis method of this application with dual verification combining subjective and objective weights, the diagnosis method with single verification combining subjective and objective weights, and the diagnosis method with dual verification combining objective weights are verified in sequence. By comparing the three methods, it is verified which method can detect early failure characteristics earlier and reduce the possibility of false alarms.

[0052] Based on historical data, the logarithmic evaluation value is log 10 The fault threshold for (S) is set to -1.8, and the fault threshold for the root mean square value is set to 0.39. The two-level early fault diagnosis method of this application uses a logarithmic evaluation value (log). 10 (S) A fault is defined as a value greater than the fault threshold and a root mean square value exceeding the corresponding fault threshold for a consecutive preset number of sampling points. A specific fault characteristic trend chart is shown below. Figure 5 As shown; the diagnostic method that combines subjective and objective weights with single-validation only uses the logarithmic evaluation value log. 10 (S) is greater than the fault threshold as the standard for faults, and no double verification is performed. The specific fault characteristic trend chart is as follows: Figure 6 As shown; the diagnostic method combining objective weights and dual verification generates only the S-value and logarithmic evaluation value log using objective weights. 10 (S) Without incorporating subjective weights as suggested by experts, i.e., using only the entropy weight method, supplemented by double verification, the specific fault characteristic trend chart is as follows: Figure 7 As shown.

[0053] Depend on Figure 5 , Figure 6 , Figure 7 The fault characteristic trend chart shows that... Figure 5 The two-level early fault diagnosis method in this application detected the first early fault at point 1228. Although the method previously included a logarithmic evaluation value (log),... 10 (S) In cases where the set fault threshold is exceeded, but because the root mean square value of all 5 sampling points exceeds the set fault threshold, the two-level fault diagnosis and judgment mechanism determines that the false alarm is caused by data fluctuation before point 1228 and does not trigger an alarm; in contrast Figure 6The diagnostic method using a combination of subjective and objective weights and single-validation shows that it identifies a fault at 349 points. This is because environmental factors or minor faults cause the logarithmic evaluation value to drop. 10 (S) exceeded the fault threshold, but this was just a simple fluctuation and did not indicate that a fault had actually occurred, so the method produced a false alarm; in contrast... Figure 7 The diagnostic method using objective weighting combined with dual verification shows that although it achieves the same fault point identification as the method in this application (detecting the first early fault at point 1228), the logarithmic evaluation value obtained by this method is significantly different. 10 (S) fluctuates greatly and is very unstable in terms of results. This indicates that the final evaluation value S lacks the introduction of subjective weights, resulting in low stability and large errors. Thus, it is very easy to falsely report under complex working conditions, and the diagnostic results are not as accurate and reliable as the two-level early fault diagnosis method of this application.

[0054] The wind turbine early fault diagnosis method of this application sequentially applies VMD algorithm noise reduction processing and combined morphological top-hat transformation signal enhancement processing to vibration data. This removes noise while enhancing fault features, effectively increasing the pulse feature amplitude of early faults, improving the signal-to-noise ratio, enhancing the robustness of the model, ensuring that the subsequently extracted features reflect the true state of the equipment, and highlighting the weak features of early faults. This prevents weak features from being masked by noise, comprehensively captures fault pulses, and effectively improves the diagnostic accuracy of early faults under complex operating conditions. The method uses the root mean square value to represent the weak impact on the wind turbine, and extracts the sample entropy of the time-domain signal to represent the weak dynamic linear changes of the signal. The combination of root mean square value, kurtosis, and entropy value can more accurately capture fault pulses, thereby accurately perceiving changes in the signal disorder of early faults and achieving accurate identification of early faults. By integrating the objective weights of root mean square value, kurtosis, and sample entropy with the subjective weights of expert experience, a comprehensive weight is obtained, thus yielding the final evaluation value S and the logarithmic evaluation value log. 10 (S) leverages both data objectivity and expert experience to comprehensively reflect the operating status of wind turbine units, fully considering false alarms caused by data fluctuations, and effectively improving the reliability of fault diagnosis; by adopting a two-level fault diagnosis and judgment mechanism, only when the logarithmic evaluation value log is met... 10(S) When the number of consecutive preset number of sampling points with a value greater than the fault threshold and the root mean square value exceeds the corresponding fault threshold, it is determined to be an early fault. This can effectively filter instantaneous data fluctuations and effectively distinguish between "instantaneous exceeding of the fault threshold caused by data fluctuations" and "continuous exceeding of the fault threshold caused by real faults". This avoids the problem of high false alarm rate in existing technologies that trigger alarms by simply having a single health score below the fault threshold. It can effectively avoid false alarms in complex working conditions and the diagnostic results are accurate and reliable.

[0055] In some implementations, the two-level early fault diagnosis method for wind turbines also includes a model optimization step: verifying the model's accuracy based on historical diagnostic data, and dynamically updating the fault thresholds and overall weights for each feature. Thus, after obtaining the aforementioned early fault diagnosis model, historical data is used for verification to determine the accuracy of the model's diagnostic results. The fault thresholds and weights for each feature are gradually updated and optimized based on historical data, thereby continuously improving the model's diagnostic accuracy and adapting to feature changes caused by wind turbine performance degradation. Specifically, the fault threshold update uses a "sliding window method," and the weight update re-executes the calculation process of steps S4.2-S4.4.

[0056] Specifically, the "sliding window method" divides continuous historical data into multiple overlapping or continuous subsets, i.e., windows, with a fixed window size. Each time, the window slides forward by a preset step size, recalculating the fault threshold and comprehensive weight using only the data within the current window. This effectively filters out interference from outdated data, ensuring that parameters are always updated based on recent valid data, thus adapting to the characteristic changes caused by wind turbine performance degradation. For example, a window can be set to contain 1440 sets of data within one year. Every time 100 new sets of data are accumulated, the window slides forward by 100 sets. That is, each time it is updated, the earliest 100 sets of data are removed, and the latest 100 sets of data are included, keeping the window always containing 1440 sets of recent data. The characteristic fault threshold and comprehensive weight are recalculated based on the updated window data. This process is repeated to ensure that the fault threshold always reflects the health status characteristics of the most recent year, avoiding false alarms or false negatives due to early outdated data causing the fault threshold to be too low or too high, thereby ensuring diagnostic accuracy.

[0057] Example 2: Please see Figures 1-5 Corresponding to the aforementioned embodiment of the two-level early fault diagnosis method for wind turbines, this embodiment also provides a two-level early fault diagnosis system for wind turbines that implements the aforementioned two-level early fault diagnosis method for wind turbines. The system includes the following components: The data acquisition module is used to collect multi-source data from the wind turbine, including operating parameters and vibration data of key components. The operating parameters and vibration data are synchronized and aligned using timestamps. The data acquisition module includes vibration sensors and a SCADA system. The vibration sensors are deployed at the positions of the first-stage internal gear ring, the second-stage internal gear ring, the high-speed shaft of the gearbox, the front bearing of the generator, and the rear bearing of the generator to collect radial vibration data of each key component.

[0058] The data processing module is used to sequentially perform VMD algorithm noise reduction processing and mathematical morphology combined morphology top-hat transformation signal enhancement processing on the vibration data. The combined morphology top-hat transformation is used to enhance positive and negative transient impacts.

[0059] The feature extraction module is used to extract the time-domain and frequency-domain features of the processed vibration data. The time-domain features include root mean square value, kurtosis, and sample entropy, while the frequency-domain features include whether a fault has occurred and the type of fault. It also extracts changes in operating parameters.

[0060] The feature fusion module is used to normalize time-domain features, calculate objective weights using the entropy weight method, and combine them with expert subjective weights to obtain a comprehensive weight. Based on the comprehensive weight, the time-domain features, frequency-domain features, and actual operating conditions are weighted and fused to obtain the final evaluation value S and the logarithmic evaluation value log. 10 (S). That is: the feature fusion module performs normalization, entropy calculation, comprehensive weight fusion and final evaluation value S calculation, and outputs the S value and the weight of each feature.

[0061] The model building module is used to determine the fault thresholds for logarithmic evaluation values ​​and root mean square values ​​based on historical data of wind turbines, and to establish an early fault diagnosis model. Specifically, the model building module stores historical data, fault thresholds, and comprehensive weights to construct the diagnostic model.

[0062] The two-level diagnostic module is used to perform two-level diagnostics using an early fault diagnosis model. First, a first-level judgment is made based on the final evaluation value S. If the logarithmic evaluation value log... 10 If (S) is greater than the fault threshold, the second-level judgment is initiated. If the root mean square value exceeds the corresponding fault threshold for a preset number of consecutive sampling points, it is judged as an early fault. That is, the two-level diagnostic module executes two-level diagnostic judgment logic, outputs diagnostic results, and if a fault is judged, it sends alarm information to the operation and maintenance platform through audible and visual alarms and remote communication.

[0063] In some implementations, the two-level early fault diagnosis system for wind turbines also includes a model optimization module. This module verifies the model's accuracy based on historical diagnostic data and dynamically updates the fault thresholds and overall weights for each characteristic fault. Specifically, the optimization module updates the fault thresholds and weights periodically, or after accumulating 100 sets of data, and generates an optimization report.

[0064] Example 3: Corresponding to the aforementioned two-level diagnostic method and system for early faults of wind turbines, this embodiment also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the aforementioned two-level diagnostic method for early faults of wind turbines.

[0065] Those skilled in the art will understand that the computer device can be a desktop computer, laptop, handheld computer, cloud server, or other computing device, and the computer device can interact with the user through a keyboard, mouse, remote control, touchpad, or voice control device.

[0066] The memory includes at least one type of readable storage medium, which can be a magnetic medium, such as a floppy disk, hard disk, or magnetic tape; or an optical medium, such as a compact disc read-only memory (CD-ROM) or a digital versatile disc (DVD); or a semiconductor medium, such as a solid-state disk (SSD), random access memory (RAM), read-only memory (ROM), a smart media card (SMC), a secure digital card (SD), a flash card, or a register. In some embodiments, the memory can be an internal storage unit of the computer device, such as the hard disk or RAM of the computer device; in other embodiments, the memory can be an external storage device of the computer device, such as a plug-in hard disk, a smart media card (SMC), a secure digital card (SD), or a flash card equipped on the computer device. Of course, the memory can also include both the internal storage unit and the external storage device of the computer device. In this embodiment, the memory is used to store the operating system and various application software installed on the computer device, such as the program code of the two-level diagnosis method for early faults of wind turbines described in various embodiments of this application.

[0067] The processor may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip. This processor is typically used to control the overall operation of the computer device. In this embodiment, the processor is used to run program code stored in the memory or process data, for example, to run the program code of the two-level early fault diagnosis method for wind turbines described in various embodiments of this application.

[0068] Example 4: Corresponding to the aforementioned two-level diagnostic method, system, and computer equipment for early faults of wind turbines, this embodiment also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned two-level diagnostic method for early faults of wind turbines.

[0069] Those skilled in the art will understand that the computer-readable storage medium can be any available medium accessible to a computer as described above, or a data storage device such as a server integrating one or more available media, and the computer program can be stored in the computer-readable storage medium. Through the above description of the embodiments, those skilled in the art will clearly understand that the methods in the foregoing embodiments can be implemented entirely or partially by software, hardware, firmware, or any combination thereof. Based on this, the technical solution of this application, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a terminal device to execute the two-level early fault diagnosis method for wind turbines described in the embodiments of this application.

[0070] Finally, it should be noted that although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention, all of which should be included within the protection scope of this application.

Claims

1. A two-stage early fault diagnosis method for a wind turbine generator unit, characterized in that, At least the following steps are included: S1: Multi-source data acquisition: Acquire multi-source data of the wind turbine, including operating parameters and vibration data of key components. The operating parameters and vibration data are synchronized and aligned using timestamps. S2: Vibration signal processing: The vibration data is sequentially subjected to VMD algorithm noise reduction processing and mathematical morphology combined morphological top-hat transformation signal enhancement processing. The combined morphological top-hat transformation is used to enhance positive and negative transient impacts. S3: Feature Extraction: Extract the time-domain and frequency-domain features of the processed vibration data. The time-domain features include root mean square value, kurtosis, and sample entropy. The frequency-domain features include whether a fault has occurred and the type of fault. Extract changes in operating parameters. S4: Comprehensive weight fusion and evaluation value calculation: The time-domain features are normalized, the objective weight is calculated using the entropy weight method, and the comprehensive weight is obtained by combining the expert subjective weight. Based on the comprehensive weight, the time-domain features, frequency-domain features, and actual working conditions are weighted and fused to calculate the final evaluation value and logarithmic evaluation value. S4.1: Normalization of time-domain features: positive indices are applied to root mean square value and kurtosis, while negative indices are applied to sample entropy; Wherein, the positive indicators: ; Negative indicators: ; S4.2: Calculate the entropy value: First calculate the proportion, then calculate the information entropy; wherein the ratio of: ; Information entropy: ; S4.3: Weighting: First calculate the difference coefficient, then calculate the objective weight, and then use the subjective weight given by the experts, combined with the multiplication integration method to obtain the comprehensive weight; Among them, the coefficient of difference: ; Objective weighting: ; Subjective weighting: ; Overall weighting: ; S4.4: Calculate the final evaluation value S: ; S4.5: Calculate the log evaluation value: Perform a constant log transformation on the final evaluation value S to obtain the log evaluation value: log 10 (S); S5: Fault threshold determination and diagnostic model establishment: Based on historical data of wind turbine units, determine the fault threshold of the logarithmic evaluation value and the fault threshold of the root mean square value, and establish an early fault diagnosis model; S6: Two-level fault diagnosis and judgment: The early fault diagnosis model is used to perform two-level diagnosis. First, a first-level judgment is made based on the logarithmic evaluation value. If the logarithmic evaluation value is greater than the fault threshold, the second-level judgment is initiated. If the root mean square value exceeds the corresponding fault threshold for a consecutive preset number of sampling points, it is judged as an early fault.

2. The two-level diagnostic method for early faults of wind turbine units according to claim 1, characterized in that: In step S1, the operating parameters are the wind turbine status parameters collected by the SCADA system, including the main shaft speed, generator speed, generator real-time power, wind speed, pitch angle, gear oil temperature 30s average value, nacelle temperature 30s average value and generator speed 30s average value, with a collection interval of 1 hour. In step S1, the key components include the first-stage internal gear ring of the gearbox, the second-stage internal gear ring of the gearbox, the high-speed shaft of the gearbox, the front bearing of the generator, and the rear bearing of the generator. The vibration data includes radial vibration data of each key component collected by vibration sensors.

3. The two-level diagnostic method for early faults of wind turbine units according to claim 1, characterized in that: In step S2, the vibration data is first denoised using the VMD algorithm, as shown in the following formula: ； In the formula, , ; The noise-reduced vibration data is then subjected to a mathematical morphology-based combined morphological cap transform signal enhancement process, as shown in the following formula: ； ； ； ； ； In the formula, f is the input vibration signal, γ(f) is the result of morphological opening operation, φ(f) is the result of morphological closing operation, ε(f) is the erosion operation, δ(f) is the dilation operation, and B is the structuring element.

4. The two-level diagnostic method for early faults of wind turbine units according to claim 1, characterized in that: In step S3, the frequency domain features are used to generate an amplitude-frequency diagram through fast Fourier transform, and the fault type is determined based on the peak value of the characteristic frequency. The fault type includes at least peeling, wear, corrosion, and significant impact. The time-domain features include root mean square value, kurtosis, and sample entropy, calculated using the following formulas: Root mean square value: ; kurtosis: ; Sample entropy: ; In the formula, ; ； ； Where m is the embedding dimension, m=2; r is the similarity tolerance, r=0.2σ, σ is the signal standard deviation; N is the signal length; and Θ is the Heaviside function.

5. The two-level diagnostic method for early faults of wind turbine units according to claim 1, characterized in that: In step S6, the preset number of sampling points is 5. When the logarithmic evaluation value is greater than the fault threshold and the root mean square value exceeds the corresponding fault threshold for five consecutive sampling points, it is determined to be an early fault.

6. The two-level diagnostic method for early faults of wind turbine units according to claim 1, characterized in that: It also includes model optimization steps: verifying the model accuracy based on historical diagnostic data, and dynamically updating the fault thresholds and overall weights for each feature.

7. A two-level diagnostic system for early faults in wind turbine generators, characterized in that, The system implementing the two-level early fault diagnosis method for wind turbines as described in any one of claims 1-6 includes: The data acquisition module is used to collect multi-source data from the wind turbine, including operating parameters and vibration data of key components. The operating parameters and vibration data are synchronized and aligned using timestamps. The data acquisition module includes vibration sensors and a SCADA system. The vibration sensors are deployed at the positions of the first-stage internal gear ring, the second-stage internal gear ring, the high-speed shaft of the gearbox, the front bearing of the generator, and the rear bearing of the generator to collect radial vibration data of each key component. The data processing module is used to sequentially perform VMD algorithm noise reduction processing and mathematical morphology combined morphology top-hat transformation signal enhancement processing on the vibration data. The combined morphology top-hat transformation is used to enhance positive and negative transient impacts. The feature extraction module is used to extract the time-domain and frequency-domain features of the processed vibration data. The time-domain features include root mean square value, kurtosis, and sample entropy. The frequency-domain features include whether a fault has occurred and the type of fault. The module also extracts changes in operating parameters. The feature fusion module is used to normalize the time-domain features, calculate the objective weights using the entropy weight method, and obtain the comprehensive weights by combining the subjective weights of experts. Based on the comprehensive weights, the time-domain features, frequency-domain features, and actual working conditions are weighted and fused to obtain the final evaluation value and the logarithmic evaluation value. The model building module is used to determine the fault threshold of the logarithmic evaluation value and the fault threshold of the root mean square value based on the historical data of the wind turbine, and to establish an early fault diagnosis model. The two-level diagnostic module is used to perform two-level diagnosis using the early fault diagnosis model. First, a first-level judgment is made based on the logarithmic evaluation value. If the logarithmic evaluation value is greater than the fault threshold, a second-level judgment is initiated. If the root mean square value exceeds the corresponding fault threshold for a consecutive preset number of sampling points, it is determined to be an early fault.

8. The two-level diagnostic system for early faults of wind turbine generators according to claim 7, characterized in that: It also includes a model optimization module, which is used to verify the model accuracy based on historical diagnostic data and dynamically update the fault thresholds and overall weights for each feature.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, it implements the two-level diagnosis method for early faults of wind turbine units as described in any one of claims 1-6.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by the processor, it implements the two-level diagnosis method for early faults of wind turbine units as described in any one of claims 1-6.

Images