Wind turbine gearbox fault diagnosis method, device and computer equipment

Abstract

This invention relates to the field of wind power generation technology, and discloses a method, device, and computer equipment for fault diagnosis of wind turbine gearboxes. The method includes: collecting multi-source data from the wind turbine gearbox; extracting time-domain features, frequency-domain features, temperature features, and operating condition features from the multi-source data; performing weighted feature fusion on the time-domain features, frequency-domain features, temperature features, and operating condition features to obtain real-time fused features; acquiring historical health data of the wind turbine gearbox; establishing a health benchmark for the wind turbine gearbox based on the historical health data; and constructing a health model based on the real-time fused features and the health benchmark; and using the health model to diagnose faults in the wind turbine gearbox. This invention combines multi-source data from the wind turbine gearbox with dynamic feature fusion and deep learning modeling to achieve gearbox health status assessment and fault type identification.

Description

Technical Field

[0001] This invention relates to the field of wind power generation technology, specifically to a method, device, and computer equipment for diagnosing gearbox faults in wind turbine generators. Background Technology

[0002] Existing wind turbine gearbox fault diagnosis technologies mainly include the following two categories:

[0003] 1. Single sensor analysis method:

[0004] ① Vibration signal analysis: Gear wear and bearing damage are detected by time-frequency domain characteristics (such as root mean square, kurtosis, and spectral energy), but it is sensitive to environmental noise and has difficulty distinguishing fault modes under complex working conditions.

[0005] ② Temperature monitoring method: It judges lubrication failure or friction abnormality based on temperature change trend, but the response is lagging and cannot capture instantaneous impact failure.

[0006] ③SCADA (Supervisory Control and Data Acquisition) parameter threshold method: It uses parameters such as power and speed to set fixed threshold alarms, but it lacks multi-parameter coupling analysis and has a high false alarm rate.

[0007] 2. Simple data fusion method:

[0008] Static weights are used to fuse features such as vibration and temperature, but the impact of dynamic changes in operating conditions on the sensitivity of features is not considered, resulting in insufficient robustness of the fusion index.

[0009] Therefore, there is an urgent need for a fault diagnosis method for wind turbine gearboxes based on multi-source data fusion. Summary of the Invention

[0010] In view of this, the present invention provides a method, apparatus and computer equipment for diagnosing gearbox faults in wind turbines, in order to solve the problem of low accuracy in diagnosing gearbox faults in wind turbines due to the lack of multi-source data fusion in the prior art.

[0011] In a first aspect, the present invention provides a method for diagnosing faults in a wind turbine gearbox, the method comprising:

[0012] Collect multi-source data from wind turbine gearboxes;

[0013] The time-domain features, frequency-domain features, temperature features, and operating condition features of the multi-source data are extracted respectively.

[0014] Weighted feature fusion is performed on time domain features, frequency domain features, temperature features, and operating condition features to obtain real-time fused features;

[0015] Historical health data of wind turbine gearboxes are obtained, a health benchmark for wind turbine gearboxes is established based on the historical health data, and a health model is constructed based on real-time fused features and the health benchmark.

[0016] A health model is used to diagnose faults in the gearbox of a wind turbine.

[0017] This invention provides a method for fault diagnosis of wind turbine gearboxes. By collecting multi-source data from wind turbine gearboxes, it avoids the limitations of single data sources and provides a rich and comprehensive data foundation for fault diagnosis. Features are extracted from four dimensions: time domain, frequency domain, temperature, and operating conditions, constructing a complete fault-sensitive feature system. Dynamic weighted feature fusion is achieved, and weights are automatically allocated according to the concentration of feature distribution to highlight the contribution of fault-sensitive features. A health benchmark is established using historical health data, and a health model is constructed based on real-time fused features and the health benchmark. This method can accurately assess the health status of the gearbox, providing a scientific basis for fault diagnosis and solving the problem of low accuracy in wind turbine gearbox fault diagnosis caused by the lack of multi-source data fusion in existing technologies.

[0018] In one alternative implementation, the multi-source data includes vibration signals, temperature signals, and operating parameters;

[0019] Collect multi-source data from wind turbine gearboxes, including:

[0020] Three-axis vibration sensors are installed on the high-speed shaft, low-speed shaft and planetary carrier of the wind turbine gearbox to collect vibration signals of the wind turbine gearbox.

[0021] Temperature sensors are installed in the bearing housing and near-meshing gearbox wall of the wind turbine gearbox to collect temperature signals from the wind turbine gearbox.

[0022] The SCADA monitoring system is used to collect the operating parameters of the wind turbine, including speed, power and torque.

[0023] A torque sensor is installed at the low-speed shaft coupling of the wind turbine gearbox, and an adaptive sampling frequency adjustment strategy is determined based on the rotational speed and torque.

[0024] This invention provides a fault diagnosis method for wind turbine gearboxes. By arranging various types of sensors on the high-speed shaft, low-speed shaft, and planetary carrier of the gearbox, it simultaneously collects multi-source data such as vibration and temperature. Combined with SCADA operating parameters, it comprehensively covers the gearbox's operating status information. Different types of data complement each other, avoiding the limitations of a single data source and providing a rich and comprehensive data foundation for fault diagnosis. The proposed sensor arrangement scheme, combined with miniature sensors, addresses the space constraints of the planetary carrier. A dynamic sampling frequency adjustment strategy based on speed and torque is designed to balance data accuracy and storage cost.

[0025] In an optional implementation, before extracting the time-domain features, frequency-domain features, temperature features, and operating condition features of the multi-source data respectively, the method further includes:

[0026] The CEEMDAN algorithm is used to reduce noise in the vibration signal. Simultaneously, a preset formula is used to standardize the vibration signal, temperature signal, and operating parameters.

[0027] When using the CEEMDAN algorithm to denoise the vibration signal, a preset speed synchronization noise template is used to suppress the interference components related to the vibration signal and gear meshing frequency, and a physical-statistical dual criterion is used to screen the intrinsic mode functions corresponding to the vibration signal.

[0028] The present invention provides a method for diagnosing gearbox faults in wind turbines, which introduces a speed synchronization noise template to suppress interference components related to gear meshing frequency in a targeted manner, and uses a physical statistical dual criterion (correlation coefficient + kurtosis) to screen effective IMFs and improve the ability to retain impact signals.

[0029] In one optional implementation, the time-domain features, frequency-domain features, temperature features, and operating condition features of the multi-source data are extracted, including:

[0030] Extract the time-domain and frequency-domain features of the vibration signal; the time-domain features include the root mean square value and kurtosis, and the frequency-domain features include the spectral energy integral and the sideband energy ratio.

[0031] Extract the temperature features of the temperature signal, including the rate of temperature rise and the temperature distribution uniformity index;

[0032] Extract operating condition characteristics from the operating parameters, including load-speed coupling factor and power fluctuation entropy.

[0033] This invention provides a fault diagnosis method for wind turbine gearboxes, extracting features from four dimensions: time domain, frequency domain, temperature, and operating conditions, thus constructing a complete fault-sensitive feature system. Time-domain features (root mean square value, kurtosis) are sensitive to uniform wear and localized impacts in the gearbox; frequency-domain features (spectral energy integral, sideband energy ratio) can accurately capture faults and modulation phenomena at specific frequencies; temperature features (temperature rise rate, temperature distribution uniformity index) effectively reflect lubrication status and cooling system faults; and operating condition features (load-speed coupling factor, power fluctuation entropy) characterize mechanical transmission efficiency and power stability. This multi-dimensional feature extraction method can deeply mine the fault information contained in the data, improving the sensitivity and accuracy of fault diagnosis.

[0034] In one optional implementation, weighted feature fusion is performed on time-domain features, frequency-domain features, temperature features, and operating condition features to obtain real-time fused features, including:

[0035] The time-domain features, frequency-domain features, temperature features, and operating condition features are normalized respectively.

[0036] Calculate the feature information entropy corresponding to the normalized time domain features, frequency domain features, temperature features and operating condition features, and calculate the weights of each feature based on the feature information entropy of each feature.

[0037] The time-domain features, frequency-domain features, temperature features, and operating condition features, along with their corresponding weights, are linearly fused to obtain real-time fused features.

[0038] This invention provides a fault diagnosis method for wind turbine gearboxes. Based on information entropy theory, it achieves dynamic weighted feature fusion, automatically allocating weights according to the concentration of feature distribution, highlighting the contribution of fault-sensitive features. The weights are adjusted in real time as the data changes, adapting to changes in feature importance under different operating conditions and fault scenarios. Compared to fixed-weight fusion methods, this method more accurately reflects the actual operating status of the gearbox. Simultaneously, by normalizing and weighted linear fusion to generate a comprehensive fault sensitivity index, the influence of dimensional differences is reduced, enhancing the scientific rigor and practicality of feature fusion.

[0039] In one optional implementation, a health benchmark for the wind turbine gearbox is established based on historical health data, and a health model is constructed based on real-time fused features and the health benchmark, including:

[0040] Historical health data was used to train a pre-defined gated loop unit model to obtain the health benchmark of the wind turbine gearbox.

[0041] A health score is calculated based on real-time fusion features, a health baseline, and a preset health status tolerance threshold, and the trend of the health score change within a preset time period is also calculated.

[0042] The health score and its changing trend, along with a pre-defined expert rule base, are used as the fault type decision rules. A health model is then constructed based on these fault type decision rules.

[0043] This invention provides a method for fault diagnosis of wind turbine gearboxes. It utilizes historical health data to train a GRU model to establish a health benchmark. The GRU model can effectively process long-sequence data and capture the temporal variation patterns of gearbox operating states. Strict training constraints are set, including an appropriate loss function, sufficient training data, and overfitting control strategies, to ensure the model's reliability and generalization ability. By calculating real-time health scores and comparing real-time fused features with the health benchmark, combined with dynamically adjusted thresholds, the health status of the gearbox can be accurately assessed, providing a scientific basis for fault diagnosis.

[0044] In one optional implementation, a health model is used to diagnose faults in the wind turbine gearbox, including:

[0045] Acquire real-time multi-source data of wind turbine gearboxes and input the real-time multi-source data of wind turbine gearboxes into the health model to obtain the predicted health score;

[0046] An alarm is triggered when the predicted health score is lower than the preset alarm threshold, indicating a potential fault risk.

[0047] The real-time feature weights are dynamically updated based on the weight learning rate, historical fusion features, and real-time fusion features, and the real-time feature weight vector is calculated.

[0048] A standard library of fault types for wind turbine gearbox faults is constructed, and the cosine similarity between the real-time feature weight vector and the standard weight pattern of each fault type in the standard library is calculated. The fault type with the highest cosine similarity is taken as the fault diagnosis result.

[0049] This invention provides a fault diagnosis method for wind turbine gearboxes. It collects data in real time and calculates a health score, enabling timely detection of abnormal changes in gearbox operating status and achieving early warning of faults. Combining threshold alarms and cosine similarity-based fault mode matching, and utilizing an expert rule base for fault type decision-making, it considers both the absolute value of the health score and its changing trend. Furthermore, by comparing with a standard weighted model, it improves the accuracy and reliability of fault diagnosis.

[0050] In one optional implementation, the wind turbine gearbox fault diagnosis method further includes:

[0051] The fault type decision rules and health model are updated using the newly diagnosed fault types.

[0052] The present invention provides a method for diagnosing gearbox faults in wind turbines. The online self-optimization mechanism of the model can update the model parameters and decision rules using newly diagnosed fault data, forming a diagnostic closed loop, continuously improving diagnostic accuracy, and adapting to changes in equipment performance and new fault types.

[0053] Secondly, the present invention provides a wind turbine gearbox fault diagnosis device, the device comprising:

[0054] The data acquisition module is used to collect multi-source data from the wind turbine gearbox;

[0055] The feature extraction module is used to extract time-domain features, frequency-domain features, temperature features, and operating condition features from multi-source data, respectively.

[0056] The feature fusion module is used to perform weighted feature fusion on time domain features, frequency domain features, temperature features and operating condition features to obtain real-time fused features;

[0057] A health model is constructed to obtain historical health data of wind turbine gearboxes, establish a health benchmark for wind turbine gearboxes based on historical health data, and construct a health model based on real-time fused features and the health benchmark.

[0058] The fault diagnosis module is used to diagnose faults in the gearbox of wind turbines using a health model.

[0059] Thirdly, the present invention provides a computer device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the wind turbine gearbox fault diagnosis method described in the first aspect or any corresponding embodiment.

[0060] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute the wind turbine gearbox fault diagnosis method described in the first aspect or any corresponding embodiment thereof.

[0061] Fifthly, the present invention provides a computer program product, including computer instructions, which are used to cause a computer to execute the wind turbine gearbox fault diagnosis method described in the first aspect or any corresponding embodiment thereof. Attached Figure Description

[0062] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0063] Figure 1 This is a flowchart illustrating a wind turbine gearbox fault diagnosis method according to an embodiment of the present invention.

[0064] Figure 2 This is a flowchart illustrating another wind turbine gearbox fault diagnosis method according to an embodiment of the present invention;

[0065] Figure 3 This is a flowchart illustrating another wind turbine gearbox fault diagnosis method according to an embodiment of the present invention;

[0066] Figure 4 This is a structural block diagram of a wind turbine gearbox fault diagnosis device according to an embodiment of the present invention;

[0067] Figure 5 This is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. Detailed Implementation

[0068] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0069] As the core transmission component of a wind turbine, the gearbox accounts for over 30% of all turbine failures. Current technological shortcomings include:

[0070] ① Poor data synchronization: Traditional methods fail to achieve strict time alignment of vibration, temperature, and SCADA data, resulting in the loss of feature correlation.

[0071] ② Insufficient noise suppression: Vibration signal noise reduction methods (such as traditional EMD) have mode aliasing problems, which affect the extraction of high-frequency impact components.

[0072] ③ Rigid feature fusion: Fixed weight allocation cannot adapt to changes in feature importance under varying speed and load conditions, reducing diagnostic sensitivity.

[0073] ④ Limitations of health assessment modeling: Health assessment based on static thresholds or shallow models is difficult to capture the nonlinear temporal characteristics of gearbox degradation processes.

[0074] This embodiment provides a method for diagnosing gearbox faults in wind turbines. By using high-precision synchronous acquisition and noise reduction of multi-source heterogeneous data (vibration, temperature, torque, speed, and power), dynamic extraction and fusion of fault-sensitive features under complex operating conditions, and real-time quantitative assessment of gearbox health status, it achieves the effect of accurate early fault identification.

[0075] According to an embodiment of the present invention, a method for diagnosing gearbox faults in wind turbine generators is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0076] This embodiment provides a method for diagnosing gearbox faults in wind turbine generators, which can be used in wind turbine generators. Figure 1 This is a flowchart of a wind turbine gearbox fault diagnosis method according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps:

[0077] Step S101: Collect multi-source data of the wind turbine gearbox.

[0078] Specifically, triaxial vibration sensors, temperature sensors, and torque sensors are arranged on the high-speed shaft, low-speed shaft, and planetary carrier of the wind turbine gearbox, with the sampling frequency adaptively adjusted (1kHz-50kHz). Vibration signals, temperature signals, torque signals, and SCADA operating parameters of the wind turbine are acquired through vibration acceleration sensors, infrared temperature sensors, torque sensors, and the SCADA system.

[0079] Step S102: Extract the time domain features, frequency domain features, temperature features, and operating condition features of the multi-source data respectively.

[0080] Specifically, time-domain and frequency-domain features are extracted from vibration signals, temperature features are extracted from temperature signals, and operating condition features are extracted from SCADA operating condition parameters.

[0081] Step S103: Perform weighted feature fusion on time domain features, frequency domain features, temperature features and operating condition features to obtain real-time fused features.

[0082] Specifically, weighted feature fusion refers to the process in wind turbine gearbox fault diagnosis that extracts time-domain, frequency-domain, temperature, and operating condition features from multi-source data such as vibration, temperature, and operating conditions, assigns corresponding weights to each feature based on the importance of the fault diagnosis, and then performs linear combination to generate a comprehensive fault sensitivity index.

[0083] That is, after normalizing the time domain features, frequency domain features, temperature features and operating condition features, the weight of each feature is dynamically calculated based on the information entropy theory, highlighting the contribution of fault-sensitive features, and the normalized features are linearly fused according to the weights to generate a comprehensive fault-sensitive index, which is the real-time fused feature.

[0084] Step S104: Obtain historical health data of the wind turbine gearbox, establish a health benchmark for the wind turbine gearbox based on the historical health data, and construct a health model based on real-time fusion features and the health benchmark.

[0085] Specifically, the health benchmark is a key reference standard used to measure the operating status of wind turbine gearboxes in fault diagnosis, providing a quantitative basis for fault diagnosis. In detail, the health benchmark refers to a quantitative standard for normal operating status established based on historical health operation data of the equipment within the wind turbine gearbox fault diagnosis system, used to measure the degree of deviation between the current operating status and the ideal health status.

[0086] A gated recurrent unit (GRU) model is trained based on historical health data to establish a health benchmark under normal gearbox operation. Real-time fused features are compared with the health benchmark to calculate a health score, and finally, a health model is constructed based on the health score.

[0087] Step S105: Use a health model to diagnose faults in the wind turbine gearbox.

[0088] Specifically, the multi-source data of the wind turbine gearbox collected in real time is input into the health model to obtain a health score. Based on the relationship between the health score and the preset score threshold, it is determined whether there is a fault. If a fault exists, the fault mode matching is used to diagnose the fault type, thereby achieving the purpose of accurately assessing the health status of the gearbox.

[0089] The wind turbine gearbox fault diagnosis method provided in this embodiment avoids the limitations of a single data source by collecting multi-source data from the wind turbine gearbox, providing a rich and comprehensive data foundation for fault diagnosis. Features are extracted from four dimensions: time domain, frequency domain, temperature, and operating condition, constructing a complete fault-sensitive feature system. Dynamic weighted feature fusion is achieved, and weights are automatically allocated according to the concentration of feature distribution, highlighting the contribution of fault-sensitive features. A health benchmark is established using historical health data, and a health model is constructed based on real-time fused features and the health benchmark. This method can accurately assess the health status of the gearbox, providing a scientific basis for fault diagnosis and solving the problem of low accuracy in wind turbine gearbox fault diagnosis caused by the lack of multi-source data fusion in existing technologies.

[0090] This embodiment provides a method for diagnosing gearbox faults in wind turbine generators, which can be used in wind turbine generators. Figure 2 This is a flowchart of a wind turbine gearbox fault diagnosis method according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following steps:

[0091] Step S201: Collect multi-source data of the wind turbine gearbox.

[0092] Specifically, the multi-source data includes vibration signals, temperature signals, and operating parameters. Triaxial vibration sensors (vibration acceleration sensors) are installed on the high-speed shaft, low-speed shaft, and planetary carrier of the wind turbine gearbox. Temperature sensors (infrared temperature sensors) are installed on the bearing housings and near-meshing gearbox walls of the wind turbine gearbox. A torque sensor is installed at the low-speed shaft coupling of the wind turbine gearbox, with the sampling frequency adaptively adjusted (1kHz-50kHz). Raw data from the wind turbine gearbox is acquired through the vibration acceleration sensors, infrared temperature sensors, torque sensors, and the SCADA monitoring system, including:

[0093] Vibration signal: Sampling frequency = 12.8kHz;

[0094] Temperature signal: T(t), sampling interval = 1 second;

[0095] SCADA operating parameters: collect speed R (rpm), power P (MW), and load torque L (kN·m);

[0096] All the above data are marked with a unified timestamp. The time window length is Δt = 10 minutes.

[0097] The sensor selection and parameter configuration are shown in Table 1 below:

[0098] Table 1 Sensor Selection and Parameter Configuration

[0099]

[0100] The above step S201 includes:

[0101] Step S2011: Arrange triaxial vibration sensors on the high-speed shaft, low-speed shaft and planetary carrier of the wind turbine gearbox, and use triaxial vibration sensors to collect vibration signals of the wind turbine gearbox.

[0102] Specifically, the installation of the triaxial vibration sensor includes:

[0103] High-speed shaft:

[0104] Installation method: Magnetic base (neodymium iron boron permanent magnet) + threaded reinforcement (M6 stainless steel bolt).

[0105] Axial alignment: X-axis (radial), Y-axis (axial), and Z-axis (tangential) are aligned with the direction of gear rotation.

[0106] Anti-resonance design: The triaxial vibration sensor base is equipped with a rubber damping pad (Shore hardness 70A) to suppress high-frequency resonance.

[0107] Planetary carrier: Space-constrained solution: Miniature triaxial sensor (15×10×10mm), fixed with epoxy resin adhesive, with built-in temperature compensation module.

[0108] In step S2012, temperature sensors are arranged in the bearing housing and near-meshing gearbox wall of the wind turbine gearbox, and the temperature signals of the wind turbine gearbox are collected using temperature sensors.

[0109] Specifically, the temperature sensor installation includes:

[0110] Uses an armored PT100 temperature sensor;

[0111] Bearing housing installation location: Install the sensor near the rolling elements in the bearing housing. It is recommended to install it on the side or bottom of the bearing housing. Apply a thin layer of thermal grease (thermal conductivity) evenly to the temperature sensing part of the sensor. Gear meshing surface: Select the housing wall near the gear meshing area and apply thermally conductive adhesive (thermal conductivity...). The temperature sensor is tightly attached to the wall of the enclosure.

[0112] Step S2013: Use the SCADA monitoring system to collect the operating parameters of the wind turbine, including speed, power and torque.

[0113] Specifically, rotational speed is used Indicated, torque is expressed as express.

[0114] In step S2014, a torque sensor is installed at the low-speed shaft coupling of the wind turbine gearbox, and an adaptive sampling frequency adjustment strategy is determined based on the rotational speed and torque.

[0115] Specifically, the torque sensor is integrated:

[0116] Non-contact design: Employs magnetoelastic torque measurement technology to avoid signal interference from traditional slip ring structures.

[0117] Dynamic compensation: Built-in strain gauge temperature drift compensation circuit (temperature drift < 0.005% FS / ℃).

[0118] The adaptive sampling frequency adjustment strategy is determined using a torque sensor, including:

[0119] Reference rules: Low-speed shaft (speed < 30 rpm): vibration sampling rate 1 kHz. High-speed shaft (speed ≥ 1000 rpm): vibration sampling rate 50 kHz.

[0120] Dynamic adjustment algorithm:

[0121] Based on rotational speed and torque The critical frequency is calculated using the following formula:

[0122] (1);

[0123] in, Number of teeth For adjustment coefficients, Maximum torque, sampling frequency according to Dynamic adjustment (satisfying Nyquist's theorem, i.e., Nyquist sampling theorem).

[0124] Triggered sampling: When the energy of the vibration signal increases by more than 30% in a short period of time, it automatically switches to the 50kHz high-speed sampling mode for 200ms.

[0125] Step S202: The CEEMDAN algorithm is used to denoise the vibration signal. At the same time, the vibration signal, temperature signal and operating parameters are standardized using a preset formula. Specifically, when the CEEMDAN algorithm is used to denoise the vibration signal, a preset speed synchronization noise template is used to suppress the interference components related to the vibration signal and gear meshing frequency. The physical and statistical dual criteria are used to screen the intrinsic mode functions corresponding to the vibration signal.

[0126] Specifically, the improved CEEMDAN algorithm for vibration signal noise reduction processing follows:

[0127] Step 1: Signal Initialization:

[0128] Input vibration signal Set the overall average number of times (generally ), noise amplitude coefficient (default )in, The standard deviation is denoted as .

[0129] Step 2: Iterative decomposition:

[0130] For the The next iteration ( ), Directional noise injection:

[0131] Generate speed synchronization noise signal The formula is as follows:

[0132] (2);

[0133] Baseband: (Unit: Hz);

[0134] Gearbox meshing frequency ;

[0135] in, This is a speed synchronization noise signal, which is related to time. The function; , These are all summation indices used to represent the order of harmonics; It is a positive integer representing the highest order of the harmonics, used to determine the upper limit of the summation, that is, taking into account the highest order harmonic component in the signal; , The amplitude coefficient (determined through regression of historical data). , All are random initial phases.

[0136] Generate a disturbance signal containing a speed noise template:

[0137] (3);

[0138] EMD decomposition (Empirical Mode Decomposition): [The following is a list of decomposition methods, not a direct translation] EMD decomposition was performed to obtain the first layer of intrinsic mode functions. .

[0139] Residual calculation: ;

[0140] Cyclic decomposition: for residuals ( Repeat the above process until the IMF (Intrinsic Mode Function) termination condition (std < 0.3) is met.

[0141] Step 3: Set average: for The results of each iteration are averaged to obtain the final IMF component:

[0142] ( (4).

[0143] Step 4: Effective IMF screening, i.e., screening IMFs based on both physical and statistical criteria:

[0144] Correlation coefficient criteria: (reserve );

[0145] Kurtosis Criteria: (reserve );

[0146] in, For the first d Individual eigenmode functions Compared with the original vibration signal The correlation coefficient between them ranges from [-1, +1]; , E[ ], Let these represent covariance, standard deviation, expected value, and mean, respectively. ; .

[0147] Step 5: Signal Reconstruction: The filtered IMFs are superimposed to obtain the denoised signal. :

[0148] in, For a valid IMF index set.

[0149] Step 6. Key parameter optimization: Noise amplitude coefficient (Dynamic adjustment strategy):

[0150] (5);

[0151] in, This is the frequency corresponding to the maximum permissible speed of the gearbox.

[0152] The number of iterations is adaptively selected based on signal complexity. :

[0153] (6);

[0154] The improved adaptive noise-complete set empirical mode decomposition described above is used to remove high-frequency noise from vibration signals.

[0155] Noise reduction vibration signal: =CEEMDAN( , K , ε );

[0156] in, K The number of intrinsic mode functions (IMFs) in the vibration signal decomposition is automatically determined by the number of peak values ​​in the signal spectrum; ε is the noise amplitude coefficient, with an initial value. =0.2, dynamically adjusted based on signal-to-noise ratio:

[0157] (7).

[0158] Vibration signals, temperature signals, and operating parameters are standardized using a preset formula to eliminate dimensions. The preset formula is as follows:

[0159] Standardized temperature , , , :

[0160] (8);

[0161] Formula (8) applies to all input data: ∈{ , T, ,L,P}; The signal mean (data normalization benchmark, calculated based on a sliding time window, recalculated every 10 minutes) is calculated in real time as the time window Δ. t Arithmetic mean of internal data:

[0162] = (9);

[0163] in, The signal standard deviation (data standardization benchmark, calculated based on a sliding time window, recalculated every 10 minutes) is calculated in real time as follows:

[0164] = (10);

[0165] in, This represents multi-source data sampled, where n represents the total number of sample points and k is the data from the kth sample point.

[0166] Step S203: Extract the time domain features, frequency domain features, temperature features, and operating condition features of the multi-source data respectively.

[0167] Specifically, step S203 includes:

[0168] Step S2031: Extract the time-domain and frequency-domain features of the vibration signal; the time-domain features include the root mean square value and kurtosis, and the frequency-domain features include the spectral energy integral and the sideband energy ratio.

[0169] Specifically, time-domain feature calculation:

[0170] 1. Root Mean Square (RMS): The average energy of the vibration signal, sensitive to uniform wear (such as tooth surface wear), the formula is as follows:

[0171] (11);

[0172] in, The vibration signal after noise reduction k Data from each sampling point, For time windows =Number of sampling points within 10 minutes × .

[0173] 2. Kurtosis: The intensity of the impact component of a vibration signal, sensitive to localized impacts (such as bearing spalling). The formula is as follows:

[0174] (12);

[0175] in, The average value of the vibration signal. ; The standard deviation of the vibration signal. .

[0176] Frequency domain feature calculation:

[0177] 1. Spectral energy integral, the formula is as follows:

[0178] (13);

[0179] in, i =1,2,…,M, The result is the Fourier transform of the vibration signal; For the first i The center frequency of each characteristic frequency band is determined by the physical parameters of the gearbox: = ( (Number of teeth on the gear); This refers to the bandwidth, which is the default value. =0.5× (Half octave); M is the number of meshing frequency orders monitored, usually M=3 (fundamental frequency, second harmonic, third harmonic). To characterize the vibration energy of a specific frequency band, it is sensitive to specific frequency components (such as broken teeth or eccentric faults).

[0180] 2. Sideband energy ratio: Reflects the degree of modulation fault and is sensitive to modulation phenomena (such as shaft misalignment). The formula is as follows:

[0181] (14);

[0182] in, Q represents the number of sidebands; the default value is Q=2.

[0183] Physical meaning: An increase in the sideband energy ratio indicates modulation phenomena (such as shaft misalignment or gear eccentricity).

[0184] Step S2032: Extract the temperature features of the temperature signal, including the temperature rise rate and the temperature distribution uniformity index.

[0185] Specifically, the temperature rise rate represents the rate of temperature rise after reducing its dimensions. It is directly related to the gearbox lubrication status and friction loss, and is used to detect abnormal heating (such as frictional heating caused by insufficient bearing lubrication). The formula is as follows:

[0186] (15);

[0187] in, The original temperature value before standardization (unit: °C); For time intervals, default. =5 minutes (set according to the thermal inertia characteristics of the gearbox); , : The mean and standard deviation of the temperature signal.

[0188] The Uniformity Index can detect localized overheating caused by cooling system failure. The formula is as follows:

[0189] (16);

[0190] Among them, the value of formula (16) is ∈ [0,1], and approaching 0 indicates uneven temperature distribution (such as local overheating). The temperature fluctuation range under healthy conditions is obtained from historical data statistics (e.g.) =2.5).

[0191] Step S2033: Extract the operating condition features of the operating condition parameters, including the load-speed coupling factor and power fluctuation entropy.

[0192] Specifically, SCADA operating condition feature extraction: from standardized SCADA operating condition parameters (speed) ,load ,power Extracting operating condition coupling features from the data to characterize abnormal operating conditions includes:

[0193] 1. The load-speed coupling factor reflects the mechanical transmission efficiency and is strongly correlated with gear wear and shaft alignment. The formula is as follows:

[0194] (17);

[0195] in, To prevent the division by zero correction constant, =0.1MW; Physical meaning: Characterizes the efficiency of mechanical power transmission; an abnormal increase indicates slippage of the transmission chain or wear of gears.

[0196] 2. Power Fluctuation Entropy quantifies power stability and is sensitive to grid anomalies or intermittent gear failures. The formula is as follows:

[0197] (18);

[0198] in, , Divide the power value into bin ranges (to P ∈[0,8]MW are all divided into Y =10 intervals); n The number of power sampling points within a time window Δt = 10 minutes; This indicates increased randomness in power fluctuations (such as grid disturbances or intermittent gear jamming).

[0199] Feature extraction employs an anti-interference design: temperature rise rate calculation is performed using... , Restoring the dimensions avoids the ambiguity of physical meaning caused by standardization. Power binning (Y=10) balances computational efficiency and feature resolution.

[0200] Step S204 involves weighted feature fusion of time-domain features, frequency-domain features, temperature features, and operating condition features to obtain real-time fused features. For details, please refer to [link to relevant documentation]. Figure 1 Step S103 of the illustrated embodiment will not be described again here.

[0201] Step S205: Obtain historical health data of the wind turbine gearbox, establish a health benchmark for the wind turbine gearbox based on the historical health data, and construct a health model based on real-time fused features and the health benchmark. For details, please refer to [link to relevant documentation]. Figure 1 Step S104 of the illustrated embodiment will not be described again here.

[0202] Step S206: A health model is used to diagnose faults in the wind turbine gearbox. For details, please refer to [link to relevant documentation]. Figure 1 Step S105 of the illustrated embodiment will not be described again here.

[0203] The wind turbine gearbox fault diagnosis method provided in this embodiment utilizes multiple types of sensors arranged on the high-speed shaft, low-speed shaft, and planetary carrier of the gearbox to simultaneously collect multi-source data such as vibration and temperature. Combined with SCADA operating condition parameters, this provides comprehensive coverage of the gearbox's operating status information. Different types of data complement each other, avoiding the limitations of a single data source and providing a rich and comprehensive data foundation for fault diagnosis. The proposed sensor arrangement scheme, combined with miniature sensors, addresses the space constraints of the planetary carrier. A dynamic sampling frequency adjustment strategy based on speed and torque is designed to balance data accuracy and storage cost. A speed synchronization noise template is introduced to directionally suppress interference components related to gear meshing frequency. A physical statistical dual criterion (correlation coefficient + kurtosis) is used to screen effective IMFs, improving the ability to retain impact signals. Features are extracted from four dimensions: time domain, frequency domain, temperature, and operating condition, constructing a complete fault-sensitive feature system. This multi-dimensional feature extraction method can deeply mine the fault information contained in the data, improving the sensitivity and accuracy of fault diagnosis.

[0204] This embodiment provides a method for diagnosing gearbox faults in wind turbine generators, which can be used in wind turbine generators. Figure 3 This is a flowchart of a wind turbine gearbox fault diagnosis method according to an embodiment of the present invention, such as... Figure 3 As shown, the process includes the following steps:

[0205] Step S301: Collect multi-source data from the wind turbine gearbox. For details, please refer to [link / reference]. Figure 1 Step S201 of the illustrated embodiment will not be described again here.

[0206] Step S302 involves extracting the time-domain features, frequency-domain features, temperature features, and operating condition features from the multi-source data. For details, please refer to [link to relevant documentation]. Figure 1 Step S203 of the illustrated embodiment will not be described again here.

[0207] Step S303: Perform weighted feature fusion on time domain features, frequency domain features, temperature features and operating condition features to obtain real-time fused features.

[0208] Specifically, step S303 includes:

[0209] Step S3031: Normalize the time domain features, frequency domain features, temperature features, and operating condition features respectively.

[0210] Specifically, the time-domain features, frequency-domain features, temperature features, and operating condition features are normalized using the following formula:

[0211] (19);

[0212] in, , Features Historical average under healthy conditions (obtained through statistics of gearbox trouble-free operation data, updated quarterly); Features Historical standard deviation under healthy conditions (statistical range: at least 6 months of data, updated quarterly); These are the root mean square value, kurtosis, energy integral of the second harmonic spectrum, energy integral of the third harmonic spectrum, and sideband energy ratio, respectively. These are, respectively, the rate of temperature rise, the temperature distribution uniformity index, the load-speed coupling factor, and the power fluctuation entropy; The corresponding characteristics of vibration signals, These are temperature characteristics and operating condition characteristics.

[0213] Constraints: If (Features without fluctuation), forced =1 to avoid division by zero errors.

[0214] Step S3032: Calculate the feature information entropy corresponding to the normalized time domain feature, frequency domain feature, temperature feature and operating condition feature, and calculate the weights of each feature using the entropy weight method based on the feature information entropy of each feature to obtain the weights corresponding to each feature.

[0215] Specifically, the weights of each feature are dynamically calculated based on information entropy theory, highlighting the contribution of fault-sensitive features. The feature information entropy is calculated using the following formula:

[0216] (20);

[0217] in, , To feature The range of values ​​is divided into Y =10 intervals; The number of samples in the statistics window (default) =1440, corresponding to 24-hour data).

[0218] Physical meaning: The smaller, the more distinctive The more concentrated the distribution, the more sensitive it is to faults.

[0219] The feature fusion weights are calculated using the entropy weight method, as shown in the following formula:

[0220] (twenty one);

[0221] in, n’ Given the total number of features, the constraints are as follows: ;

[0222] Physical meaning: along Updated and adjusted in real time; weighting With entropy Negative correlation; the lower the entropy (the stronger the feature's ability to distinguish faults), the higher the weight.

[0223] Step S3033: The time domain features, frequency domain features, temperature features, and operating condition features, along with their corresponding weights, are linearly fused to obtain real-time fused features.

[0224] Specifically, weighted feature fusion: normalized features are linearly fused according to their weights to generate a comprehensive fault sensitivity index, as shown in the following formula:

[0225] (twenty two);

[0226] in, This is a dimensionless comprehensive index, i.e., a real-time fusion feature, with a value range typically in the range of [-3, 3]. Values ​​exceeding this threshold indicate a fault. The wind turbine gearbox fault diagnosis method provided in this embodiment achieves dynamic weighted feature fusion based on information entropy theory. Weights are automatically allocated according to the concentration of feature distribution, highlighting the contribution of fault-sensitive features. The weights are adjusted in real-time as the data changes, adapting to changes in feature importance under different operating conditions and fault scenarios. Compared to fixed-weight fusion methods, this method more accurately reflects the actual operating status of the gearbox. Simultaneously, by generating a comprehensive fault-sensitive index through normalization and weighted linear fusion, the influence of dimensional differences is reduced, enhancing the scientific rigor and practicality of feature fusion.

[0227] Step S304: Obtain historical health data of the wind turbine gearbox, establish a health benchmark for the wind turbine gearbox based on the historical health data, and construct a health model based on real-time fusion features and the health benchmark.

[0228] Specifically, step S304 includes:

[0229] Step S3041: Use historical health data to train the preset gated loop unit model to obtain the health benchmark of the wind turbine gearbox.

[0230] Specifically, the pre-defined gated recurrent unit model is a GRU (Gated Recurrent Unit) model. The gated recurrent unit model is trained based on historical health data to establish a health benchmark under normal gearbox operation. The health benchmark formula is as follows:

[0231] (twenty three);

[0232] in: Fusion feature sequences (time windows) in historical health data t ∈[ ]); Θ is the set of parameters for the GRU model, including the input gate weights. Reset gate weight Hidden layer weights Bias terms b etc.; Θ={ , , , b It is obtained through training with historical health data; it is updated once every 100 new sets of fault data.

[0233] Training constraints employ a loss function, the formula of which is as follows:

[0234] (twenty four);

[0235] Training data size: N≥ One sample (covering different seasons and working conditions).

[0236] Step S3042: Calculate the health score based on real-time fusion features, health benchmark and preset health status tolerance threshold, and calculate the trend of the health score change within a preset time period.

[0237] Specifically, real-time fusion features Calculate a health score by comparing the score to a health benchmark.

[0238] (25);

[0239] The preset time period is 10 minutes, meaning the health score is calculated every 10 minutes; Threshold is the health tolerance threshold, calculated as the 95th percentile of the historical health data residuals (recalculated monthly), and its formula is:

[0240] (26);

[0241] Rating range: ∈(-∞,1], health status is set to If it is less than 0.8, it is considered to be in an unhealthy state.

[0242] This embodiment collects new data and calculates a health score every 10 minutes. Forming a time series By using sliding window technology, the most recent m ratings (e.g., m=6, corresponding to 1 hour of data) are tracked in real time, providing a data foundation for trend analysis.

[0243] Step S3043: The health score and its changing trend, as well as the preset expert rule base, are used as the fault type decision rules, and a health model is constructed based on the fault type decision rules.

[0244] Specifically, the expert rule base can adopt a hierarchical structure, including a basic rule layer, a composite rule layer, and an extended rule layer. The basic rule layer defines single-indicator threshold rules, such as "trigger a primary warning when (St < 0.6)"; the composite rule layer combines multiple indicator logical relationships, such as "if (St < 0.6) and {..." If the load speed coupling factor exceeds the normal range, it is judged as a serious fault. The extended rule layer dynamically adds rules according to the new fault type, such as adding "When the oil temperature rises abnormally and the vibration signal shows low-frequency fluctuations, start the oil detection process" for gearbox oil leakage faults.

[0245] The system matches real-time health scores and trend characteristics with an expert rule base. A forward reasoning strategy is employed, retrieving rules from the rule base one by one that meet the conditions. For example, if a sudden drop in health score, an abnormal increase in power fluctuation entropy, and an increase in sideband energy ratio are detected, and the rule "local gear damage" is matched, a preliminary diagnosis of local gear damage fault is made. For cases involving multiple rule matching conflicts, evidence theory or fuzzy logic is introduced to resolve the conflicts. The final fault type is determined by combining the support of multiple rules, and the trained GPU model, along with the rule-matched model, is used as the health model.

[0246] This embodiment provides a method for fault diagnosis of wind turbine gearboxes. It utilizes historical health data to train a GRU model to establish a health benchmark. The GRU model can effectively process long-sequence data and capture the temporal variation patterns of gearbox operating states. Strict training constraints are set, including an appropriate loss function, sufficient training data, and overfitting control strategies, to ensure the model's reliability and generalization ability. By calculating real-time health scores and comparing real-time fused features with the health benchmark, combined with dynamically adjusted thresholds, the health status of the gearbox can be accurately assessed, providing a scientific basis for fault diagnosis.

[0247] Step S305: Use the health model to diagnose faults in the wind turbine gearbox.

[0248] Specifically, step S305 includes:

[0249] Step S3051: Obtain real-time multi-source data of wind turbine gearbox and input the real-time multi-source data of wind turbine gearbox into the health model to obtain the predicted health score.

[0250] For details, please refer to steps S301 to S304, which will not be repeated here.

[0251] Step S3052: When the predicted health score is less than the preset alarm threshold, an alarm is triggered and a fault risk is indicated.

[0252] Specifically, the formula for determining the alarm trigger condition is as follows:

[0253] < and (27);

[0254] in, Alarm threshold, default =0.6; The threshold for the rate of health decline, default. =-0.1 / minute.

[0255] Step S3053: Dynamically update the real-time feature weights based on the weight learning rate, historical fusion features, and real-time fusion features, and calculate the real-time feature weight vector.

[0256] Specifically, weights are adjusted based on real-time diagnostic feedback to improve model adaptability. The update rule is expressed as follows:

[0257] (28);

[0258] in, , The learning rate for the weights, with initial values... =0.01, dynamically adjusted based on convergence (e.g., adjusted once every diagnostic cycle (10 minutes)): ; These are fusion feature values ​​(i.e., historical fusion features) that are manually labeled or historically diagnosed faults.

[0259] Step S3054: Construct a standard library of fault types for wind turbine gearbox faults, and calculate the cosine similarity between the real-time feature weight vector and the standard weight pattern of each fault type in the standard library. The fault type with the highest cosine similarity is taken as the fault diagnosis result.

[0260] Specifically, the formula for representing fault type is: .

[0261] in, = It is a dynamic weight vector; For the first k Standard weighting model for fault types (preset based on expert experience and historical fault data analysis; new fault types are added quarterly; such as bearing fault models). =[0.15,0.25,0.1,…]).

[0262] The formula for calculating cosine similarity is: The wind turbine gearbox fault diagnosis method provided in this embodiment collects data in real time and calculates a health score, enabling timely detection of abnormal changes in the gearbox's operating status and achieving early warning of faults. Combining threshold alarms and cosine similarity-based fault mode matching, and utilizing an expert rule base for fault type decision-making, it considers both the absolute value of the health score and its changing trend. Furthermore, by comparing with a standard weighted model, it improves the accuracy and reliability of fault diagnosis.

[0263] Step S306: Update the fault type decision rule and health model using the newly diagnosed fault types.

[0264] Specifically, the formula for fine-tuning the parameters of the GRU model is as follows:

[0265] (29);

[0266] (30);

[0267] Where M represents the amount of new fault data; For a new set of parameters for the GRU model, For the old parameter set of the GRU model, For The function with parameters is a gradient vector.

[0268] Dynamic adjustment of health status tolerance threshold:

[0269] (31);

[0270] in, =0.9: Threshold update smoothing coefficient.

[0271] The wind turbine gearbox fault diagnosis method provided in this embodiment has an online self-optimization mechanism that can use newly diagnosed fault data to update model parameters and decision rules, forming a diagnostic closed loop, continuously improving diagnostic accuracy, and adapting to changes in equipment performance and new fault types.

[0272] The wind turbine gearbox fault diagnosis method provided in this embodiment has the following beneficial effects:

[0273] 1. Improved diagnostic accuracy: Multi-source data fusion increases fault detection accuracy to 98.2% (compared to 85.3% for single vibration analysis).

[0274] 2. Enhanced anti-interference capability: After improving the noise reduction of CEEMDAN, the signal-to-noise ratio (SNR) is increased by 6-8dB, effectively preserving the fault impact components.

[0275] 3. Dynamic Adaptive Optimization: An adaptive sampling strategy reduces data volume by 30% while ensuring high-speed 50kHz capture of transient impacts (>30% energy surge); weighting With feature information entropy Dynamically adjusts to automatically focus on key features at different stages of gearbox life (early wear, severe failure). Learning rate. The adaptive mechanism avoids model overfitting or slow convergence.

[0276] 4. Early warning capability: Health score The dual mechanism of matching weighted patterns can trigger early warnings 24-48 hours before a fault occurs, with a false alarm rate of <2%; descent rate threshold Filter out transient interference.

[0277] 5. Noise reduction adaptability: Through dynamic adjustment of ε, the fault impact component can still be retained under strong noise conditions (such as sudden wind speed changes).

[0278] 6. Enhanced fault sensitivity: The entropy weight method is used to suppress high-entropy and inefficient features (such as environmental temperature interference) and strengthen the decision weight of low-entropy features (such as vibration kurtosis).

[0279] 7. Model Evolution Capability: Online updates of GRU parameters Θ and the threshold Threshold enable the system to adapt to gearbox performance degradation. Fault Mode Library It is scalable and supports novel fault diagnosis methods. This embodiment also provides a wind turbine gearbox fault diagnosis device, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0280] This embodiment provides a wind turbine gearbox fault diagnosis device, such as... Figure 4 As shown, it includes:

[0281] The data acquisition module 401 is used to acquire multi-source data from the wind turbine gearbox.

[0282] The feature extraction module 402 is used to extract time-domain features, frequency-domain features, temperature features, and operating condition features from multi-source data.

[0283] The feature fusion module 403 is used to perform weighted feature fusion on time domain features, frequency domain features, temperature features and operating condition features to obtain real-time fused features.

[0284] Model 404 is used to acquire historical health data of wind turbine gearboxes, establish a health benchmark for wind turbine gearboxes based on historical health data, and construct a health model based on real-time fused features and the health benchmark.

[0285] The fault diagnosis module 405 is used to perform fault diagnosis on the gearbox of the wind turbine using a health model.

[0286] In some optional implementations, the multi-source data includes vibration signals, temperature signals, and operating parameters; the data acquisition module 401 includes:

[0287] The vibration signal acquisition unit is used to arrange triaxial vibration sensors on the high-speed shaft, low-speed shaft and planetary carrier of the wind turbine gearbox, and to acquire the vibration signals of the wind turbine gearbox using triaxial vibration sensors.

[0288] The temperature signal acquisition unit is used to arrange temperature sensors in the bearing housing and near-meshing gearbox wall of the wind turbine gearbox, and to acquire the temperature signal of the wind turbine gearbox using the temperature sensors.

[0289] The operating parameter acquisition unit is used to acquire the operating parameters of the wind turbine gearbox using the SCADA monitoring system. The operating parameters include speed, power and torque.

[0290] An adaptive sampling frequency adjustment unit is used to place a torque sensor at the low-speed shaft coupling of the wind turbine gearbox and determine the adaptive sampling frequency adjustment strategy based on the speed and torque.

[0291] In some optional implementations, the wind turbine gearbox fault diagnosis device further includes:

[0292] The noise reduction and standardization module is used to perform noise reduction on the vibration signal using the CEEMDAN algorithm, and at the same time, it uses a preset formula to standardize the vibration signal, temperature signal and operating parameters. Specifically, when using the CEEMDAN algorithm to reduce the noise of the vibration signal, a preset speed synchronization noise template is used to suppress the interference components related to the vibration signal and gear meshing frequency, and a physical and statistical dual criterion is used to screen the intrinsic mode functions corresponding to the vibration signal.

[0293] In some alternative implementations, the feature extraction module 402 includes:

[0294] The time-frequency domain feature extraction unit is used to extract the time-domain and frequency-domain features of the vibration signal; the time-domain features include the root mean square value and kurtosis, and the frequency-domain features include the spectral energy integral and the sideband energy ratio.

[0295] The temperature feature extraction unit is used to extract the temperature features of the temperature signal, including the temperature rise rate and the temperature distribution uniformity index.

[0296] The operating condition feature extraction unit is used to extract the operating condition features of the operating condition parameters, including the load-speed coupling factor and power fluctuation entropy.

[0297] In some alternative implementations, the feature fusion module 403 includes:

[0298] The normalization processing unit is used to normalize the time domain features, frequency domain features, temperature features, and operating condition features respectively.

[0299] The weight calculation unit is used to calculate the feature information entropy corresponding to the normalized time-domain features, frequency-domain features, temperature features, and operating condition features, and to perform entropy weight calculation based on the feature information entropy of each feature to obtain the weight corresponding to each feature.

[0300] The feature fusion unit is used to perform weighted linear fusion of time-domain features, frequency-domain features, temperature features, and operating condition features, along with their corresponding weights, to obtain real-time fused features.

[0301] In some optional implementations, the health model construction model 404 includes:

[0302] The health baseline acquisition unit is used to train a preset gated loop unit model using historical health data to obtain the health baseline of the wind turbine gearbox.

[0303] The health score calculation unit is used to calculate the health score based on real-time fusion features, health benchmark and preset health status tolerance threshold, and calculate the trend of health score change within a preset time period.

[0304] The health model construction unit is used to construct a health model based on the health score and its changing trend, as well as the preset expert rule base, as fault type decision rules.

[0305] In some alternative implementations, the fault diagnosis module 405 includes:

[0306] Acquire real-time multi-source data of wind turbine gearboxes and input the real-time multi-source data of wind turbine gearboxes into the health model to obtain the predicted health score.

[0307] An alarm is triggered when the predicted health score is lower than the preset alarm threshold, indicating a potential fault risk.

[0308] The real-time feature weights are dynamically updated based on the weight learning rate, historical fusion features, and real-time fusion features, and the real-time feature weight vector is calculated.

[0309] A standard library of fault types for wind turbine gearbox faults is constructed, and the cosine similarity between the real-time feature weight vector and the standard weight pattern of each fault type in the standard library is calculated. The fault type with the highest cosine similarity is taken as the fault diagnosis result.

[0310] In some optional implementations, the wind turbine gearbox fault diagnosis device further includes:

[0311] The update module is used to update the fault type decision rules and health model based on newly diagnosed fault types.

[0312] Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0313] In this embodiment, the wind turbine gearbox fault diagnosis device is presented in the form of a functional unit. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0314] This invention also provides a computer device having the above-described features. Figure 4 The wind turbine gearbox fault diagnosis device shown is shown.

[0315] Please see Figure 5 , Figure 5 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention, such as... Figure 5As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 5 Take a processor 10 as an example.

[0316] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0317] The memory 20 stores instructions executable by at least one processor 10 to cause the at least one processor 10 to perform the method shown in the above embodiments.

[0318] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0319] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0320] The computer device also includes an input device 30 and an output device 40. The processor 10, memory 20, input device 30, and output device 40 can be connected via a bus or other means. Figure 5 Taking the example of a connection between China and Israel via a bus.

[0321] Input device 30 can receive input numerical or character information, and generate key signal inputs related to user settings and function control of the computer device, such as a touchscreen, keypad, mouse, trackpad, touchpad, joystick, one or more mouse buttons, trackball, joystick, etc. Output device 40 may include display devices, auxiliary lighting devices (e.g., LEDs), and haptic feedback devices (e.g., vibration motors). The aforementioned display devices include, but are not limited to, liquid crystal displays, light-emitting diodes, displays, and plasma displays. In some alternative embodiments, the display device may be a touchscreen.

[0322] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.

[0323] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0324] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for diagnosing gearbox faults in wind turbine generators, characterized in that, The method includes: Collect multi-source data from the wind turbine gearbox; the multi-source data includes operating parameters, including speed and torque; the collection of multi-source data from the wind turbine gearbox includes: A torque sensor is installed at the low-speed shaft coupling of the wind turbine gearbox, and an adaptive sampling frequency adjustment strategy is determined based on the speed and torque. The adaptive sampling frequency adjustment strategy is determined using the torque sensor, including: The baseline rule is established: for low-speed shafts with a rotational speed of <30 rpm, the vibration sampling frequency is 1 kHz; for high-speed shafts with a rotational speed of ≥1000 rpm, the vibration sampling frequency is 50 kHz. The critical frequency is calculated using a dynamic adjustment algorithm based on rotational speed R(t) and torque L(t), as shown in the following formula: ； in, Number of teeth For adjustment coefficients, For maximum torque, sampling frequency according to Dynamically adjusted to satisfy the Nyquist sampling theorem; When the vibration signal energy suddenly increases by more than 30% in a short period of time, the system automatically switches to a 50kHz high-speed sampling mode using a trigger sampling method, which lasts for 200ms. The time-domain features, frequency-domain features, temperature features, and operating condition features of the multi-source data are extracted respectively. The time-domain features, frequency-domain features, temperature features, and operating condition features are weighted and fused to obtain real-time fused features; Historical health data of wind turbine gearboxes are obtained, a health benchmark for wind turbine gearboxes is established based on the historical health data, and a health model is constructed based on the real-time fusion features and the health benchmark. The aforementioned health model is used to diagnose faults in the gearbox of the wind turbine.

2. The method according to claim 1, characterized in that, The multi-source data also includes vibration signals and temperature signals; The collection of multi-source data from wind turbine gearboxes also includes: A triaxial vibration sensor is arranged on the high-speed shaft, low-speed shaft and planetary carrier of the wind turbine gearbox, and the vibration signal of the wind turbine gearbox is collected by the triaxial vibration sensor. Temperature sensors are arranged in the bearing housing and near-meshing gearbox wall of the wind turbine gearbox, and the temperature signals of the wind turbine gearbox are collected using the temperature sensors. The SCADA monitoring system is used to collect the operating parameters of the wind turbine, including power.

3. The method according to claim 2, characterized in that, Before extracting the time-domain features, frequency-domain features, temperature features, and operating condition features of the multi-source data respectively, the method further includes: The CEEMDAN algorithm is used to denoise the vibration signal, and a preset formula is used to standardize the vibration signal, temperature signal, and operating parameters, wherein: When using the CEEMDAN algorithm to denoise the vibration signal, a preset speed synchronization noise template is used to suppress the interference components related to the gear meshing frequency of the vibration signal, and a physical-statistical dual criterion is used to screen the intrinsic mode functions corresponding to the vibration signal.

4. The method according to claim 2, characterized in that, The extraction of time-domain features, frequency-domain features, temperature features, and operating condition features from the multi-source data includes: Extract the time-domain and frequency-domain features of the vibration signal; the time-domain features include the root mean square value and kurtosis, and the frequency-domain features include the spectral energy integral and the sideband energy ratio; Extract the temperature features of the temperature signal, including the temperature rise rate and the temperature distribution uniformity index; The operating condition features of the operating parameters are extracted, including the load-speed coupling factor and the power fluctuation entropy.

5. The method according to claim 1, characterized in that, The time-domain features, frequency-domain features, temperature features, and operating condition features are weighted and fused to obtain real-time fused features, including: The time-domain features, frequency-domain features, temperature features, and operating condition features are each normalized. Calculate the feature information entropy corresponding to the normalized time domain features, frequency domain features, temperature features and operating condition features, and calculate the weights of each feature based on the feature information entropy of each feature. The time-domain features, frequency-domain features, temperature features, and operating condition features, along with their corresponding weights, are linearly fused to obtain real-time fused features.

6. The method according to claim 1, characterized in that, The process of establishing a health benchmark for the wind turbine gearbox based on the historical health data, and constructing a health model based on the real-time fused features and the health benchmark, includes: The historical health data is used to train the preset gated loop unit model to obtain the health benchmark of the wind turbine gearbox; A health score is calculated based on the real-time fusion features, health benchmark, and preset health status tolerance threshold, and the trend of the health score change within a preset time period is calculated. The health score and its changing trend, along with a preset expert rule base, are used as fault type decision rules. A health model is then constructed based on these fault type decision rules.

7. The method according to claim 1, characterized in that, The method of using the health model to diagnose faults in the wind turbine gearbox includes: Acquire real-time multi-source data of wind turbine gearbox and input the real-time multi-source data of wind turbine gearbox into the health model to obtain a predicted health score; An alarm is triggered when the predicted health score is less than a preset alarm threshold, indicating a potential fault risk. The real-time feature weights are dynamically updated based on the weight learning rate, historical fusion features, and real-time fusion features, and the real-time feature weight vector is calculated. A standard library of fault types for wind turbine gearbox faults is constructed, and the cosine similarity between the real-time feature weight vector and the standard weight pattern of each fault type in the standard library is calculated. The fault type with the highest cosine similarity is taken as the fault diagnosis result.

8. The method according to claim 6, characterized in that, The method further includes: The newly diagnosed fault types are used to update the fault type decision rules and the health model.

9. A fault diagnosis device for a wind turbine gearbox, characterized in that, The device includes: The data acquisition module is used to collect multi-source data from the wind turbine gearbox; the multi-source data includes operating parameters, including speed and torque; the collection of multi-source data from the wind turbine gearbox includes: A torque sensor is installed at the low-speed shaft coupling of the wind turbine gearbox, and an adaptive sampling frequency adjustment strategy is determined based on the speed and torque. The adaptive sampling frequency adjustment strategy is determined using the torque sensor, including: The baseline rule is established: for low-speed shafts with a rotational speed of <30 rpm, the vibration sampling frequency is 1 kHz; for high-speed shafts with a rotational speed of ≥1000 rpm, the vibration sampling frequency is 50 kHz. The critical frequency is calculated using a dynamic adjustment algorithm based on rotational speed R(t) and torque L(t), as shown in the following formula: ； in, Number of teeth For adjustment coefficients, For maximum torque, sampling frequency according to Dynamically adjusted to satisfy the Nyquist sampling theorem; When the vibration signal energy suddenly increases by more than 30% in a short period of time, the system automatically switches to a 50kHz high-speed sampling mode using a trigger sampling method, which lasts for 200ms. The feature extraction module is used to extract the time-domain features, frequency-domain features, temperature features, and operating condition features of the multi-source data, respectively. The feature fusion module is used to perform weighted feature fusion on the time domain features, frequency domain features, temperature features and operating condition features to obtain real-time fused features; A health model is constructed to obtain historical health data of wind turbine gearboxes, establish a health benchmark for wind turbine gearboxes based on the historical health data, and construct a health model based on the real-time fusion features and the health benchmark. The fault diagnosis module is used to perform fault diagnosis on the wind turbine gearbox using the health model.

10. A computer device, characterized in that, include: The system includes a memory and a processor, which are interconnected. The memory stores computer instructions, and the processor executes the computer instructions to perform the wind turbine gearbox fault diagnosis method according to any one of claims 1 to 8.

Images