A wind turbine gearbox fault diagnosis method and system

Abstract

This invention discloses a method and system for fault diagnosis of wind turbine gearboxes, belonging to the field of gearbox fault diagnosis technology. The method includes the following steps: collecting multi-source signal data during gearbox operation and preprocessing it; extracting feature parameters from the multi-source signal data; calculating the mutual information entropy value between the feature parameters of the multi-source signal data and the gearbox fault type in real time, and normalizing it with the feature parameters to obtain the correlation weight of the multi-source signal data; based on the correlation weight, performing fusion decision on the feature parameters through improved D-S evidence theory to obtain a new evidence body, and performing weighted averaging to obtain a comprehensive fault feature vector; inputting the comprehensive fault feature vector into a pre-trained fault classification model to obtain the gearbox fault type. This invention can solve the problem of existing technologies that rely on single vibration information and fixed-weight weighted fusion for fault diagnosis.

Description

Technical Field

[0001] This invention belongs to the field of gearbox fault diagnosis technology, specifically relating to a method and system for diagnosing gearbox faults in wind turbine generators. Background Technology

[0002] In the global transition to a clean and low-carbon energy structure, wind power, as a core component of renewable energy, continues to expand in terms of installed capacity and operational scale. The stable operation of wind turbines directly determines power generation efficiency and economic benefits. The gearbox, as a core component of the wind turbine's transmission system, plays a crucial role in converting the low-speed mechanical energy of the wind turbine into high-speed electrical energy from the generator. Because wind turbines are typically deployed in complex and harsh environments such as the field and offshore, gearboxes are subjected to complex conditions such as variable loads, strong impacts, and temperature fluctuations over long periods, making them prone to failures such as gear wear, bearing failure, shaft misalignment, and abnormal noises from the gearbox. Industry statistics show that gearbox failures account for more than 30% of total wind turbine downtime, and repair costs are high. Therefore, accurate and real-time fault diagnosis of wind turbine gearboxes is a core requirement for ensuring the safe and stable operation of wind turbines and reducing maintenance costs.

[0003] The existing Chinese invention patent with publication number CN118150157A provides a method, system, device and storage medium for diagnosing gearbox faults in wind turbines. This patent achieves multi-angle acquisition of vibration information through multiple channels, but it is still limited to the single physical dimension of vibration signal. Moreover, when vibration signals from different channels contradict each other, the fusion result is directly output by simply averaging with fixed weights, which is prone to decision-making bias due to signal conflicts, resulting in low accuracy of gearbox fault diagnosis results. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for diagnosing gearbox faults in wind turbines, in order to solve the problem of existing technologies that rely on single vibration information and perform fault diagnosis through fixed-weight fusion.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: Firstly, a method for diagnosing gearbox faults in wind turbine generators includes the following steps: After collecting and preprocessing multi-source signal data during gearbox operation, feature parameters of the multi-source signal data are extracted. The mutual information entropy value between the feature parameters of the multi-source signal data and the gearbox fault type is calculated in real time, and the correlation weight of the multi-source signal data is obtained by normalization in combination with the feature parameters. Based on the correlation weight, the feature parameters are fused and decided by the improved DS evidence theory to obtain a new evidence body, and a weighted average is performed to obtain a comprehensive fault feature vector. The comprehensive fault feature vector is input into a pre-trained fault classification model to obtain the fault type of the gearbox.

[0006] In some embodiments, the multi-source signal data includes: vibration signal, temperature signal, oil abrasive particle concentration signal, and rotational speed signal; The steps of acquiring and preprocessing multi-source signal data during gearbox operation specifically include: After the vibration signal is processed to remove the mean, the amplitude of the vibration signal is normalized using the maximum-minimum algorithm to obtain the processed vibration signal. The number of modes and the penalty factor of the VMD algorithm are determined by the particle swarm optimization algorithm. Based on the number of modes and the penalty factor, the VMD algorithm is used to decompose the processed vibration signal to obtain multiple modal components and the center frequency of each modal component. The modal components with center frequencies greater than a preset frequency threshold are deleted. The Pearson correlation coefficient between the remaining modal components and the processed vibration signal is calculated. The vibration signal is reconstructed using the remaining modal components with Pearson correlation coefficients greater than a preset correlation coefficient threshold to obtain a new vibration signal.

[0007] In some implementations, the method further includes a step of matching the new vibration signal, specifically including: The energy envelope is obtained by performing ITEO calculation on the new vibration signal, and the energy envelope is analyzed by Hilbert transform to obtain the analytical signal. The instantaneous frequency of the analytical signal is calculated by the phase difference method and denoted as the state characteristic frequency. The state characteristic frequency is matched with the theoretical state frequency, and the state label corresponding to the theoretical state frequency is output. The state label includes: normal, worn, and broken gear tooth.

[0008] In some implementations, the steps of fusing the feature parameters based on the correlation weights using an improved DS evidence theory to obtain a new evidence body, and then performing a weighted average to obtain a comprehensive fault feature vector, specifically include: Calculate the conflict coefficient between each piece of evidence in the improved DS evidence theory. When the conflict coefficient is greater than a preset conflict threshold, calculate the similarity between each piece of evidence to obtain a credibility factor. Based on the credibility factor, perform a weighted correction on the confidence of each piece of evidence to obtain a corrected basic probability allocation function. Normalize the corrected basic probability allocation function to obtain the corrected piece of evidence. The modified evidence body is fused using the DS synthesis rule to obtain a preliminary fusion result. The preliminary fusion result is then subjected to a consistency check. If the preset consistency condition is met, a new evidence body is obtained. Otherwise, the calculation weight of the credibility factor is readjusted, and the above steps of weighting and correcting the confidence of each evidence body based on the credibility factor are repeated. The feature parameters of the multi-source signal data corresponding to the new evidence are weighted and averaged according to the correlation weight to obtain a comprehensive fault feature vector.

[0009] In some implementations, the following steps are also included: The window length of the sliding window is calculated based on the rotational speed signal, and the correlation weight is dynamically updated using a sliding time window mechanism based on the window length.

[0010] In some implementations, the method further includes a step of predicting the remaining life of the gearbox, specifically including: The cumulative running time, design life parameters, and historical degradation data of the gearbox are obtained. Combined with the processed multi-source signal data and the fault type of the gearbox, a remaining life prediction model is constructed based on a long short-term memory network architecture, with the rate of change of the characteristic parameters of the multi-source signal data as the input variable and the remaining life value of the gearbox as the output variable. The feature parameters of the multi-source signal data are divided into time series to generate multiple complete degradation cycle samples. Each degradation cycle sample contains a full life cycle feature parameter sequence from normal state to fault state. The sliding window method is used to perform data augmentation on the full life cycle feature parameter sequence, extract the historical feature change rate of different degradation stages, and establish a mapping relationship between the historical feature change rate and the remaining lifetime. The remaining lifetime prediction model is trained using the mapping relationship between the historical feature change rate and the remaining lifetime to obtain the trained remaining lifetime prediction model. The mean value of the feature parameters of the multi-source signal data is calculated, and the mean value is used as the health status benchmark value. The deviation between the feature parameters of the multi-source signal data and the health status benchmark value is calculated. When the deviation exceeds a preset degradation threshold, the rate of change of the feature parameters of the multi-source signal data is input into the trained remaining life prediction model to obtain the remaining life prediction value. Acquire all multi-source signal data of the current gearbox and integrate them into real-time and historical sequences according to the order of acquisition time; The DTW algorithm is used to align the real-time sequence and the historical sequence to obtain the historical multi-source signal data corresponding to the last multi-source signal data in the real-time sequence. Based on the historical multi-source signal data corresponding to the historical sequence, the actual remaining lifetime is obtained. The deviation rate between the predicted remaining lifetime and the actual remaining lifetime is calculated. Based on the deviation rate, the Kalman filter algorithm is used to dynamically correct the output of the remaining lifetime prediction model and update the predicted remaining lifetime.

[0011] Secondly, a wind turbine gearbox fault diagnosis system includes: The data acquisition and preprocessing module is used to acquire multi-source signal data during the operation of the gearbox, preprocess the data, and extract the feature parameters of the multi-source signal data. The comprehensive fault feature vector construction module is used to calculate the mutual information entropy value between the feature parameters of the multi-source signal data and the gearbox fault type in real time, and to normalize the feature parameters to obtain the correlation weight of the multi-source signal data. Based on the correlation weight, the feature parameters are fused and decided by the improved DS evidence theory to obtain a new evidence body, and a weighted average is performed to obtain the comprehensive fault feature vector. The fault diagnosis module is used to input the comprehensive fault feature vector into a pre-trained fault classification model to obtain the fault type of the gearbox.

[0012] Thirdly, an electronic device includes a memory, a processor, and a computer program stored in the memory and executable in the processor, wherein the processor executes the computer program to implement the steps of the wind turbine gearbox fault diagnosis method.

[0013] Fourthly, a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the wind turbine gearbox fault diagnosis method.

[0014] Fifthly, a computer program product comprising a computer program that, when executed by a processor, implements the steps of the wind turbine gearbox fault diagnosis method.

[0015] Compared with the prior art, the present invention has the following beneficial effects: This invention provides a method for fault diagnosis of wind turbine gearboxes. By collecting and preprocessing multi-source signal data during gearbox operation, feature parameters are extracted from the multi-source signal data. This transforms the raw multi-source signal data into more meaningful information that better reflects the gearbox's operating status and fault characteristics, thereby uncovering hidden fault features within the signals. Extracting feature parameters reduces the amount of data while highlighting key fault-related information, improving the efficiency and accuracy of subsequent fault diagnosis. Real-time calculation of the mutual information entropy between the feature parameters of the multi-source signal data and the gearbox fault type quantifies the correlation between the feature parameters of each signal source and the fault type, revealing the importance and contribution of different signal sources in fault diagnosis. Then, the feature parameters are normalized to obtain the correlation weights of the multi-source signal data, converting the mutual information entropy values ​​into weight values. This allows different signal sources to be reasonably weighted according to their correlation with the fault type during subsequent feature fusion, avoiding the influence of subjective factors on feature fusion and improving the accuracy and reliability of the fusion results. Based on the aforementioned correlation weights, a new evidence body is obtained by fusing the feature parameters using an improved DS evidence theory. DS evidence theory is an effective method for handling uncertain information, capable of fusing information from different evidence sources to obtain more comprehensive and accurate conclusions. The improved DS evidence theory addresses the potential for unreasonable results in handling conflicting evidence by correcting the confidence level of conflicting evidence, thus improving the robustness and accuracy of the fusion decision. Through fusion decision, the feature information of multi-source signals can be comprehensively utilized to obtain a new evidence body, enhancing the ability to characterize gearbox fault features. A weighted average of the new evidence body yields a comprehensive fault feature vector, forming a more representative and discriminative feature representation, providing strong support for subsequent fault classification. Finally, the comprehensive fault feature vector is input into a pre-trained fault classification model to obtain the gearbox fault type, enabling rapid fault diagnosis and providing timely decision-making basis for equipment maintenance and repair.

[0016] Furthermore, after the vibration signal undergoes mean removal processing, the amplitude of the vibration signal is normalized using a minimax algorithm to obtain the processed vibration signal. By removing the mean of the vibration signal amplitude, the vibration signal can fluctuate around zero, eliminating any potential DC components or fixed offsets in the vibration signal, highlighting the dynamic change characteristics of the signal, and avoiding interference from DC components in feature extraction and algorithm analysis. The normalization process maps the amplitude of the mean-removed vibration signal to the range of 0-1, making signals of different magnitudes and dimensions comparable, unifying the signal scale, and improving the stability and convergence of subsequent algorithms. The particle swarm optimization algorithm is used to determine the mode number and penalty factor of the VMD algorithm, providing suitable parameters for the VMD algorithm. Suitable parameters enable the VMD algorithm to more accurately decompose the vibration signal and extract physically meaningful modal components. When determining the mode number and penalty factor of the VMD algorithm, the particle swarm optimization algorithm can perform a global search within a given parameter space, avoiding getting trapped in local optima. By continuously iterating and updating the position and velocity of the particles, it gradually approaches the optimal combination of mode number and penalty factor. Based on the modal number and penalty factor, the VMD algorithm is used to decompose the processed vibration signal, obtaining multiple modal components and the center frequency of each modal component. The VMD algorithm can decompose the processed vibration signal into multiple modal components with different center frequencies. These modal components represent vibrational components in different frequency ranges of the signal, enabling a more detailed characterization of the signal's frequency features. In gearbox fault diagnosis, different fault types (such as gear wear, bearing failure, etc.) may generate vibration signals of different frequencies. VMD decomposition can separate the frequency components corresponding to these different fault characteristics, facilitating the analysis and identification of specific frequency features related to the fault. In actual gearbox operation, high-frequency noise is often unrelated to faults and may interfere with the extraction of fault features. This invention removes high-frequency noise components from the signal by deleting modal components with center frequencies greater than a preset threshold, thereby improving the signal-to-noise ratio and allowing subsequent analysis to focus more on low-frequency or mid-frequency components related to the fault, reducing the interference of noise on fault diagnosis. Finally, the Pearson correlation coefficient between the remaining modal components and the processed vibration signal is calculated to evaluate the correlation between each modal component and the original signal. The higher the correlation, the more representative the modal component is of the main features of the original signal. The vibration signal is reconstructed using the remaining modal components with Pearson correlation coefficients greater than the preset correlation coefficient threshold to obtain a new vibration signal. Modal components with strong correlation to the original signal are selected for signal reconstruction, resulting in a purer vibration signal that better highlights fault features and is beneficial for subsequent fault classification and diagnosis.

[0017] Furthermore, the energy envelope is obtained by performing ITEO calculation on the new vibration signal. The energy envelope is then analyzed using Hilbert transform to obtain an analytical signal. The instantaneous frequency of the analytical signal is calculated using the phase difference method and denoted as the state characteristic frequency. The state characteristic frequency is matched with the theoretical state frequency to output the state label corresponding to the theoretical state frequency. The state label includes: normal, worn, and broken gear teeth. By performing ITEO calculation on the new vibration signal, the energy changes of the vibration signal are captured, and the energy envelope is obtained. The energy envelope reflects the trend of energy change of the vibration signal over time. It removes high-frequency fluctuation details from the signal while retaining the overall outline of energy change. For example, during gearbox operation, when gears experience wear or broken teeth, the energy distribution of the vibration signal changes, and the energy envelope clearly shows this energy change. Subsequently, the energy envelope is analyzed using Hilbert transform to obtain the analytical signal. The analytical signal contains the amplitude and phase information of the original signal and has a single sideband characteristic, avoiding interference from negative frequency components. This allows for easier extraction of the instantaneous features of the signal, such as instantaneous amplitude and instantaneous phase. The instantaneous frequency of the analytical signal is calculated using the phase difference method and denoted as the state characteristic frequency. By performing a differential operation on the phase of the analytical signal, the rate of change of the phase over time can be obtained, and this rate of change is the instantaneous frequency. The state characteristic frequency reflects the frequency change of the vibration signal at different times, and it can capture the dynamic frequency characteristics of the signal. The theoretical state frequency is a frequency value pre-calculated based on the normal operating state of the gearbox and known fault modes. This application compares the calculated state characteristic frequency with the theoretical state frequency to determine their similarity. If the state characteristic frequency matches a certain theoretical state frequency within a certain error range, the match is considered successful, and the state label corresponding to the theoretical state frequency at the time of successful matching is output. This allows the actual collected vibration signal characteristics to be linked with known fault modes, providing a basis for fault diagnosis. The state labels include normal, worn, and broken gear teeth, which are a direct description of the gearbox's operating state. When the state characteristic frequency matches the theoretical state frequency successfully, the corresponding state label is output, clearly informing the equipment maintenance personnel of the current state of the gearbox.

[0018] Furthermore, the conflict coefficient is an indicator that measures the degree of contradiction between different pieces of evidence. By calculating the conflict coefficient, this application can quantify the consistency or conflict in the information expression of each piece of evidence. For example, in gearbox fault diagnosis, data collected by different sensors (such as vibration sensors and temperature sensors) serve as different pieces of evidence, and the fault information they reflect may differ or even conflict. Calculating the conflict coefficient can clearly identify these potential conflicts. When the conflict coefficient is greater than a preset conflict threshold, this application performs a weighted correction on the evidence based on the credibility factor of each piece of evidence. The credibility factor is calculated based on the similarity between the evidence and other pieces of evidence. The higher the similarity, the more consistent the evidence is in information expression with other pieces of evidence, and the higher its credibility. For example, if the data collected by a vibration sensor has a high similarity in fault characteristics to the data collected by several other sensors, then the credibility factor of the evidence corresponding to that sensor will be larger. By performing a weighted correction on the evidence, this application can adjust the weight of each piece of evidence in the fusion process. For evidence with high credibility, a larger weight is assigned to allow it to play a greater role in the fusion result; for evidence with low credibility, a smaller weight is assigned to reduce its adverse impact on the fusion result, thereby minimizing interference from conflicting evidence and improving the accuracy of the fusion result. Finally, this application fuses the preprocessed evidence using the DS synthesis rule to obtain a preliminary fusion result. This preliminary fusion result integrates the information from each piece of evidence after correction, reflecting to some extent the degree to which all evidence supports different propositions (such as the normal state, wear state, and broken tooth state of the gearbox). The preliminary fusion result is then subjected to a consistency check to verify whether it conforms to the expected logic and practical meaning, providing a basis for readjusting the calculation weights of the credibility factor. If the fusion result does not meet the consistency condition, it indicates that the current credibility factor calculation method may be unreasonable, and the calculation weights of the credibility factor need to be adjusted to improve the fusion result. If the consistency condition is not met, the calculation weights of the credibility factor are readjusted until a fusion result that meets the consistency condition is obtained. By continuously adjusting the calculation weights of the credibility factor, the weight allocation of each piece of evidence in the fusion process can be changed, thereby affecting the final fusion result.

[0019] Furthermore, the gearbox's rotational speed directly affects the frequency components and variation patterns of fault characteristic signals such as vibration signals. Calculating the window length based on the rotational speed signal takes into account that the meshing frequency of gears and the rotational frequency of bearings inside the gearbox change at different rotational speeds, resulting in different signal characteristic periods. By calculating the window length using the rotational speed signal, this application allows the window setting to better reflect the actual operating state of the gearbox, providing a foundation for accurate subsequent signal analysis. Based on the calculated window length, a sliding time window mechanism is used to dynamically update the correlation weights, reflecting changes in the correlation between signal sources in real time. This is because the gearbox may be in different operating conditions or fault development stages at different times, and the correlation between signal sources will change accordingly. By dynamically updating the correlation weights, this application can more accurately identify signal sources that play a key role in fault diagnosis at different times.

[0020] Furthermore, this application monitors the characteristic parameters of multi-source signal data in real time and compares them with the health status benchmark. The remaining life prediction model is only triggered when the deviation exceeds a preset degradation threshold. This allows for flexible adjustment of the prediction timing based on the actual operating state of the gearbox, avoiding unnecessary prediction calculations and improving prediction efficiency. This application combines the gearbox's cumulative operating time, design life parameters, and historical degradation data to construct the remaining life prediction model, fully utilizing the equipment's historical operating information. Historical degradation data reflects the degradation patterns of the gearbox under different operating stages and conditions. Analyzing this data allows for a better understanding of the equipment's degradation process, providing rich samples and evidence for training the prediction model. This enables the model to more accurately capture the equipment's degradation trend and improve prediction accuracy. In gearbox remaining life prediction, the change of characteristic parameters over time is a typical sequential data problem. LSTM can learn the dependency relationship between characteristic parameters at different time steps, better capturing the long-term degradation trend of the equipment, thereby improving the accuracy of remaining life prediction. Calculating the deviation from the health status benchmark using the characteristic parameters of real-time multi-source signal data can reflect the degree of deviation between the gearbox's current state and its health status in real time. By setting a preset degradation threshold, it is possible to promptly determine whether the equipment exhibits obvious signs of degradation. By adopting the above scheme, this application can promptly detect potential equipment failures, providing accurate triggering conditions for subsequent remaining life prediction, avoiding premature or late predictions, and improving the timeliness and relevance of the prediction. Only when the deviation exceeds the preset degradation threshold is the rate of change of characteristic parameters of the real-time multi-source signal data input into the remaining life prediction model, outputting the remaining life prediction value, making the prediction results closer to the actual situation. The historical multi-source signal data of the same type of gearbox throughout its entire life cycle obtained by this application can cover the complete process of the gearbox from its normal state upon commissioning, through various operating conditions and potential failure development stages, until the final failure, providing rich and realistic basic information for accurate remaining life prediction and avoiding prediction deviations caused by incomplete data. Subsequently, this application extracts feature parameters from historical multi-source signal data and divides these parameters into time series, generating multiple complete degradation cycle samples. Each sample contains a full lifecycle feature parameter sequence from normal to fault state, thereby transforming complex multi-source signal data into a more representative and analyzable feature parameter sequence. Furthermore, by dividing the data into degradation cycle samples, the data becomes more structured and organized, facilitating a clearer observation of the gearbox's characteristic changes at different stages. This application employs a sliding window method to augment the feature parameter sequence, expanding the scale and diversity of the dataset without increasing the actual data acquisition cost. The sliding window can generate multiple subsequences for different time periods, containing information on the gearbox's characteristic changes within different local time ranges.Data augmentation helps improve the model's generalization ability, enabling it to better adapt to various complex operating conditions and data changes, and reducing the risk of overfitting. Subsequently, historical feature change rates at different degradation stages are extracted, and a mapping relationship between these historical feature change rates and remaining lifetime is established. Feature change rates more directly reflect the gearbox's degradation speed and trend, and establishing a mapping relationship with remaining lifetime provides a clear input-output correspondence rule for the prediction model. By analyzing the relationship between feature change rates and remaining lifetime in historical data, the inherent laws of gearbox degradation can be uncovered. This application uses the mapping relationship between historical feature change rates and remaining lifetime to train the remaining lifetime prediction model, enabling the model to learn the complex nonlinear relationship between feature change rates and remaining lifetime. Through training with a large amount of historical data, the model can continuously adjust its parameters to improve prediction accuracy. By calculating the mean values ​​of various feature parameters from historical multi-source signal data, a health status benchmark value is determined. This application fully utilizes the large amount of historical data accumulated during normal operation. Compared to manually setting benchmark values, the method based on the mean of historical data can more realistically reflect the feature parameter levels of the equipment in a healthy state, reducing the influence of subjective factors on the determination of the benchmark value. By comparing the characteristic parameters of real-time acquired multi-source signal data with a health status benchmark, the deviation between the current state and the health status of the equipment can be intuitively determined. When the deviation between the real-time characteristic parameters and the benchmark exceeds a certain threshold, it can be considered that the equipment may have a fault or abnormality, thereby triggering the fault detection mechanism. The benchmark-based comparison method can promptly detect potential faults in the equipment, providing timely information for subsequent fault diagnosis and maintenance. This application acquires all multi-source signal data of the current gearbox and integrates it into a real-time sequence according to the acquisition time sequence, which can promptly capture the current operating status information of the gearbox. Then, historical multi-source signal data is integrated into a historical sequence, which records the operating characteristics of the gearbox at different times. By comparing it with the real-time sequence, the changing trend of the gearbox status can be clearly observed, which helps to predict its remaining lifespan. This application uses the DTW (Dynamic Time Warping) algorithm to align the real-time sequence and the historical sequence, which can effectively solve the problem of inconsistency in length and local time offset that may exist between the two sequences on the time axis. The operating status of the gearbox may be affected by various factors, such as load changes and ambient temperature fluctuations. These factors can cause the real-time sequence and the historical sequence to not correspond completely in time. The DTW algorithm finds the best matching path between two sequences by dynamically adjusting the time axis of the sequences, so that real-time sequences and historical sequences can be effectively aligned in time.Subsequently, after obtaining the historical multi-source signal data corresponding to the last multi-source signal data in the real-time sequence, the actual remaining lifetime is obtained based on the corresponding historical multi-source signal data in the historical sequence. This application fully utilizes the empirical information of historical data. The historical sequence records the time interval between the gearbox's development from a similar current state to a fault state. By finding the corresponding position of the real-time sequence in the historical sequence, the remaining lifetime under the same or similar states in the historical data can be referenced, thereby estimating the actual remaining lifetime of the current gearbox more accurately. This application also intuitively reflects the accuracy of the remaining lifetime prediction model by calculating the deviation rate between the predicted remaining lifetime value and the actual remaining lifetime. If the deviation rate is large, it indicates that there may be some problems with the prediction model, which needs further adjustment and optimization; if the deviation rate is small, it indicates that the prediction model has high accuracy and can continue to be used. Based on the deviation rate, the Kalman filter algorithm is used to dynamically correct the output result of the remaining lifetime prediction model, which can update the remaining lifetime prediction value in real time and improve the accuracy and stability of the prediction. Attached Figure Description

[0021] Figure 1 A flowchart of a wind turbine gearbox fault diagnosis method provided in an embodiment of the present invention; Figure 2 This is a flowchart illustrating the preprocessing of vibration signals in a wind turbine gearbox fault diagnosis method provided by an embodiment of the present invention. Figure 3 This is a flowchart illustrating the prediction of the remaining life of a gearbox in a wind turbine gearbox fault diagnosis method provided in an embodiment of the present invention. Figure 4 This is a structural diagram of a wind turbine gearbox fault diagnosis system provided in an embodiment of the present invention. Detailed Implementation

[0022] To enable those skilled in the art to better understand the present invention, the technical solution of the present invention will be further described in detail below with reference to the accompanying drawings. The content described herein is for explanation rather than limitation of the present invention.

[0023] It should be noted that the terms "comprising" and "having" and any variations thereof in the specification and claims of this invention are intended to cover a non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such processes, methods, systems, products, or devices.

[0024] like Figure 1 As shown in the figure, this embodiment provides a method for diagnosing gearbox faults in wind turbines, including the following steps: S11, after collecting multi-source signal data during the operation of the gearbox and performing preprocessing, extract the feature parameters of the multi-source signal data; The multi-source signal data includes vibration signal, temperature signal, oil abrasive particle concentration signal, and rotational speed signal.

[0025] During operation, gearboxes generate vibrations due to gear meshing and bearing rotation. In this embodiment, by installing acceleration or velocity sensors in key parts of the gearbox (such as bearing housings and gearbox housings), vibration signals can be collected. These vibration signals contain fault information about gears and bearings. For example, wear or broken teeth on gear teeth, pitting or inner / outer ring faults in bearings can all cause changes in the characteristics of the vibration signals.

[0026] When a fault occurs, such as bearing damage leading to increased friction or poor lubrication, the local temperature will rise. In this embodiment, temperature sensors are installed in key parts of the gearbox (such as bearing positions, gearbox oil sump, etc.) to monitor temperature changes in real time.

[0027] During operation, the wear of gears and bearings in a gearbox generates abrasive particles, which enter the lubricating oil. In this embodiment, by detecting the concentration of abrasive particles in the lubricating oil, the wear level and fault development trend of the gearbox can be determined.

[0028] This embodiment measures the rotational speed of the input or output shaft to understand the gearbox's workload and operational stability. Then, characteristic parameters of the acquired multi-source signal data are extracted. The characteristic parameters of various multi-source signal data are as follows: The characteristic parameters of vibration signals include time-domain characteristic parameters such as mean, variance, peak value, peak-to-peak value, kurtosis, and impulse factor. The mean reflects the static component of the signal; the variance represents the degree of dispersion of the signal; the peak value and peak-to-peak value can reflect the impact degree of the signal; kurtosis is more sensitive to the impact component in the signal and is often used to detect early failures of gears and bearings; the impulse factor can highlight the impulse characteristics in the signal.

[0029] By performing a Fourier transform on the vibration signal, a spectrum is obtained, and then frequency domain characteristic parameters such as the dominant frequency, band energy, and sidebands are extracted. The dominant frequency reflects the main frequency components of the signal, and different fault types will cause changes in the dominant frequency; band energy reflects the energy distribution of the signal in different frequency bands, and the energy of certain bands will increase when a fault occurs; sidebands are one of the characteristics of gear faults, and the type of gear fault can be determined by analyzing the distribution and amplitude of sidebands.

[0030] Temperature signal characteristic parameters include: average temperature, maximum temperature, minimum temperature, and temperature change rate. The average temperature reflects the overall temperature level of the gearbox; the maximum and minimum temperatures indicate the temperature fluctuation range; and the temperature change rate reflects the speed at which the temperature rises or falls. An abnormal temperature change rate may indicate the occurrence of a fault.

[0031] The characteristic parameters of oil abrasive particle concentration signals include: the average value, standard deviation, and rate of change of abrasive particle concentration. The average value of abrasive particle concentration reflects the total content of abrasive particles in the oil; the standard deviation indicates the dispersion of abrasive particle concentration; and the rate of change of abrasive particle concentration can reflect the development trend of wear, indicating that wear is intensifying when the rate of change increases.

[0032] The characteristic parameters of the speed signal include: the average speed and speed fluctuation rate, which can be extracted. The average speed reflects the normal operating speed of the gearbox; the speed fluctuation rate reflects the stability of the speed. When the speed fluctuation rate is abnormal, it may be related to gear transmission failure or load changes.

[0033] S12, calculate the mutual information entropy value between the feature parameters of the multi-source signal data and the gearbox fault type in real time, and normalize the feature parameters to obtain the correlation weight of the multi-source signal data. Based on the correlation weight, perform fusion decision on the feature parameters by improving the DS evidence theory to obtain a new evidence body, and perform weighted averaging to obtain a comprehensive fault feature vector. Real-time calculation of the mutual information entropy values ​​between the characteristic parameters of each signal source and the fault type is performed. In gearbox fault diagnosis, by calculating the mutual information entropy values ​​between the characteristic parameters of each signal source and the fault type, the contribution of the characteristic parameter to the fault type can be evaluated. The larger the mutual information entropy value, the stronger the correlation between the characteristic parameter and the fault type, and the greater its role in fault diagnosis. The calculation model for the mutual information entropy of various characteristic parameters and fault types is as follows: ; in, The mutual information entropy value of feature parameter X and fault type Y. The probability that the feature parameter is x; Let y be the probability of fault type y; This represents the joint probability distribution of the characteristic parameter x and the fault type y. This embodiment estimates the probability distribution using statistical historical data. 、 and Then, the mutual information entropy value is calculated.

[0034] The correlation weight of each signal source is calculated based on the mutual information entropy value. The correlation weight of the i-th signal source can be obtained by normalizing the mutual information entropy value of the characteristic parameters of the i-th signal source and the fault type.

[0035] After calculating the correlation weights of each signal source based on the mutual information entropy value, the method further includes: calculating the window length based on the gearbox rotation speed signal, and dynamically updating the correlation weights based on the window length using a sliding time window mechanism.

[0036] Based on the relevance weights or dynamically updated relevance weights, the feature parameters are fused and decided using the improved DS evidence theory to correct the confidence of conflicting evidence and obtain a new body of evidence. This includes steps one through three, and the specific process is as follows: Step 1: Calculate the conflict coefficient between each piece of evidence. The conflict coefficient is used to quantify the degree of conflict between the pieces of evidence. The calculation model is as follows: ; Wherein, K is the conflict coefficient, which ranges from 0 to 1. The larger the value, the more intense the conflict. and Let A and B be the basic probability assignment functions for the two evidence bodies, and let A and B be subsets of the Frame of Discernment (FOD). The Frame of Discernment (FOD) is a set of mutually exclusive and complete propositions, and subsets A and B are any non-empty subsets of this set, representing possible combinations of propositions.

[0037] Step 2: When the conflict coefficient is greater than the preset conflict threshold, the evidence body is preprocessed. The preprocessing includes: calculating the credibility factor of each evidence body, calculating the similarity between each pair of evidence bodies, constructing a similarity matrix based on all the similarities, and calculating the similarity between each pair of evidence bodies can be done using methods such as cosine similarity, Jaccard coefficient or improved Tanimoto similarity coefficient.

[0038] Based on the similarity matrix, a credibility factor is assigned to each piece of evidence, the value of which can be defined as: ; in, The degree of consistency between the i-th piece of evidence and other pieces of evidence, i.e., the credibility factor of the i-th piece of evidence; N is the total number of pieces of evidence; Let be the similarity between the i-th piece of evidence and the j-th piece of evidence.

[0039] Based on the credibility factor of each piece of evidence, the evidence is weighted and adjusted to obtain the adjusted basic probability allocation function. The calculation model is as follows: ; in, The basic probability assignment function is modified for evidence A; Let A be the basic probability assignment function for evidence body A; To identify the frame.

[0040] The modified basic probability assignment functions are normalized to ensure that the sum of all modified basic probability assignment functions equals 1.

[0041] Step 3: Fuse the preprocessed evidence using the DS synthesis rule to obtain a preliminary fusion result. Verify whether the preliminary fusion result meets the preset consistency conditions (such as the convergence of the trust function, decision stability, etc.). If not, the calculation weights of the credibility factor need to be readjusted.

[0042] If the consistency condition is not met, the weight of evidence with higher consistency is increased and the weight of conflicting evidence is decreased based on the consistency test results until a fusion result that meets the consistency condition is obtained. A weighted average method is then used to aggregate the new evidence to obtain a comprehensive fault feature vector.

[0043] S13, input the comprehensive fault feature vector into the pre-trained fault classification model to obtain the fault type of the gearbox.

[0044] The fault classification model can be a support vector machine (SVM), a neural network (such as a backpropagation neural network, a convolutional neural network (CNN), a recurrent neural network (RNN) and its variant LSTM), a decision tree, a random forest, etc. This embodiment can collect, process, fuse, and classify multi-source signal data during gearbox operation, thereby accurately diagnosing the fault type of the gearbox and providing a basis for gearbox maintenance and repair.

[0045] like Figure 2 As shown, S11 also includes: S21 vibration signal denoising involves calculating the mean of the original vibration signal and subtracting it to obtain a zero-mean signal, ensuring that the signal energy is concentrated in the AC component. A maximum-minimum algorithm is then used to de-normalize the amplitude of the vibration signal, resulting in the processed vibration signal.

[0046] The number of modes and penalty factor for the VMD algorithm are determined using the particle swarm optimization algorithm. The number of modes determines the number of subbands in the signal decomposition, directly affecting the risk of mode aliasing and over-decomposition. The penalty factor controls the strength of the bandwidth constraint; a larger value results in a narrower mode bandwidth but may lead to the loss of high-frequency components.

[0047] Based on the number of modes and the penalty factor, the VMD algorithm is used to decompose the processed vibration signal to obtain multiple modal components and the center frequency of each mode. Modal components with center frequencies greater than a preset frequency threshold are deleted. The Pearson correlation coefficient between the remaining modal components and the processed vibration signal is calculated. The vibration signal is reconstructed using the remaining modal components with Pearson correlation coefficients greater than a preset correlation coefficient threshold to obtain a new vibration signal. The new vibration signal eliminates irrelevant modes from the original vibration signal and retains components that are strongly correlated with fault characteristics, thereby improving the signal-to-noise ratio of the vibration signal.

[0048] S22 matching was used to perform ITEO calculations on the new vibration signal to obtain the energy envelope. An improved Teager energy operator was employed in the calculation process. The instantaneous energy change of the signal is highlighted by nonlinear transformation. The calculation model is as follows: ; in, Let be the amplitude of the vibration signal at time n; Let be the amplitude of the vibration signal at time n-1; Let be the amplitude of the vibration signal at time n+1.

[0049] The energy envelope is analyzed using Hilbert transform to obtain an analytic signal, thus converting the real signal into a complex signal, facilitating the extraction of instantaneous phase and frequency. The instantaneous frequency of the analytic signal is calculated using the phase difference method and denoted as the state characteristic frequency. This state characteristic frequency is then matched with the theoretical state frequency. If a match is successful, a state label corresponding to the theoretical state frequency at the time of the successful match is output. These state labels include: normal, worn, and broken gear tooth. This method enables fully automated analysis of the entire process from the original vibration signal to the fault state label, providing reliable technical support for intelligent operation and maintenance.

[0050] like Figure 3 As shown, after completing the gearbox fault diagnosis, the following steps are also included: S31 modeling is based on the fault type and multi-source signal data output by the fault classification model. Combined with the cumulative running time of the gearbox, design life parameters and historical degradation data, a remaining life prediction model is constructed. The remaining life prediction model uses the rate of change of characteristic parameters of historical multi-source signal data as input variables and the remaining life value as output variables. The Long Short-Term Memory network is used to capture the degradation trend of characteristic parameters over time.

[0051] The remaining lifetime prediction model architecture includes an input layer, a data preprocessing layer, an LSTM degradation trend capture layer, and a fully connected layer. The input layer is used to input data. The data preprocessing layer performs Z-score standardization on input features, such as the rate of change of feature parameters and cumulative runtime, to eliminate dimensional differences. Then, the data is divided into fixed-length time-series windows, with a window length L=50 and a step size S=10. Each window contains feature vectors from L time steps, forming the input sequence G={gt L+1,…,gt}.

[0052] The LSTM degradation trend capture layer comprises multiple layers of LSTM units, each containing 128 neurons, including input, forget, and output gates. These gates are used for selectively remembering and forgetting historical information. Stacking 2-3 layers of LSTM units enhances the remaining lifetime prediction model's ability to capture long-term dependencies. A fully connected layer maps the hidden states of the last LSTM layer to remaining lifetime values.

[0053] S32 model training acquires historical multi-source signal data for the entire lifecycle of the same type of gearbox. The entire lifecycle is divided into: Normal state: From the time the equipment is put into use until the first slight degradation characteristics appear, such as the vibration RMS value exceeding the initial value by 10%. Degradation state: From slight degradation to fault warning, such as the vibration RMS value exceeding the threshold μ+3σ, where μ is the historical mean and σ is the standard deviation. Fault state: From fault warning to equipment shutdown, such as the vibration amplitude exceeding the safety limit or a sudden temperature rise.

[0054] Feature parameters of historical multi-source signal data are extracted, and the feature parameters of historical multi-source signal data are divided into time series to generate multiple complete degradation cycle samples. Each sample contains a full life cycle feature parameter sequence from normal state to fault state.

[0055] The sliding window method was used to augment the feature parameter sequence, extract the historical feature change rate at different degradation stages, establish the mapping relationship between the historical feature change rate and the remaining lifetime, and simulate the degradation process using the Wiener process model. The calculation model of the Wiener process model is as follows: ; in, Let be the function value of the Wiener process model at time t; For Brownian motion; parameters are estimated using maximum likelihood estimation. and .

[0056] The remaining lifetime (RUL) is divided into m intervals, and the distribution of historical characteristic change rates within each interval is statistically analyzed. m sampling points are sampled from the Wiener process model, and the mapping relationship between historical characteristic change rates and remaining lifetime is established sequentially.

[0057] The remaining lifetime prediction model is trained by using the mapping relationship between historical characteristic change rate and remaining lifetime, and the trained remaining lifetime prediction model is obtained.

[0058] S33 prediction calculates the mean of various characteristic parameters of historical multi-source signal data, uses the mean as the health status benchmark, and calculates the deviation from the health status benchmark using the characteristic parameters of real-time multi-source signal data. When the deviation exceeds the preset degradation threshold, the change rate of the characteristic parameters of real-time multi-source signal data is input into the trained remaining life prediction model, and the remaining life prediction value is output.

[0059] S34 corrects the gearbox by acquiring all multi-source signal data of the current gearbox and integrating them into a real-time sequence according to the order of acquisition time, and integrating historical multi-source signal data into a historical sequence.

[0060] The DTW algorithm is used to align the real-time and historical sequences, outputting the optimal alignment path and cumulative distance matrix. This locates the mapping position of the real-time sequence within the historical trajectory, obtaining the historical multi-source signal data corresponding to the last multi-source signal data in the real-time sequence. Based on the corresponding historical multi-source signal data in the historical sequence, the RUL distribution of the same stage is queried. Kernel density estimation is used to construct the probability density function. Combined with Euclidean distance, five nearest-neighbor historical cases are selected to calculate the weighted average RUL. ; ; in, This is the weighted average RUL, i.e., actual remaining lifetime. Let the remaining lifetime be the historical case of the i-th nearest neighbor. This is the distance metric between the real-time feature sequence and the i-th historical case after DTW alignment.

[0061] The deviation rate between the predicted remaining lifetime and the actual remaining lifetime is calculated. Based on the deviation rate, the Kalman filter algorithm is used to dynamically correct the output of the remaining lifetime prediction model and update the remaining lifetime prediction value. The process is as follows: The state vector C of the Kalman filter algorithm is as follows: ; in, This represents the rate of change of remaining lifetime.

[0062] Considering the accelerated degradation effect of wind turbine gearboxes, this embodiment adopts a nonlinear state transition model, and the calculation formula is as follows: ; in, The degradation acceleration coefficient; For shape parameters; The actual remaining lifetime at time k; Let be the rate of change of remaining lifetime at time k-1.

[0063] The computational model for the observation model is as follows: ; in, Let be the observation vector at time k; For the predicted remaining lifespan; To observe noise.

[0064] The Kalman gain calculation model is as follows: ; ; ; ; ; in, The observation matrix; The covariance of the observed noise is obtained through historical prediction error statistics; To observe the covariance of the noise; This is the transition state matrix.

[0065] The status update process is as follows: ; ; ; in, The strength coefficient is adjusted to a value ranging from 0.1 to 0.5.

[0066] The final predicted value is the value in the state vector. Quantity.

[0067] This embodiment achieves prediction and correction of the gearbox by setting a remaining life prediction model, reducing the prediction error of the model and improving the accuracy of the remaining life prediction value.

[0068] like Figure 4 As shown, this embodiment provides a wind turbine gearbox fault diagnosis system, including: The data acquisition and preprocessing module is used to acquire multi-source signal data during the operation of the gearbox, preprocess the data, and extract the feature parameters of the multi-source signal data. The comprehensive fault feature vector construction module is used to calculate the mutual information entropy value between the feature parameters of the multi-source signal data and the gearbox fault type in real time, and to normalize the feature parameters to obtain the correlation weight of the multi-source signal data. Based on the correlation weight, the feature parameters are fused and decided by the improved DS evidence theory to obtain a new evidence body, and a weighted average is performed to obtain the comprehensive fault feature vector. The fault diagnosis module is used to input the comprehensive fault feature vector into a pre-trained fault classification model to obtain the fault type of the gearbox.

[0069] In summary, this embodiment offers the following advantages: By collecting multi-source signal data during gearbox operation, it obtains gearbox operating status information from multiple dimensions. Then, feature parameters are extracted from the collected multi-source signal data, transforming the original signals into more meaningful information that better reflects the gearbox's operating status and fault characteristics. This allows for the discovery of hidden fault features within the signals. Extracting feature parameters reduces the amount of data while highlighting key fault-related information, improving the efficiency and accuracy of subsequent fault diagnosis. Real-time calculation of the mutual information entropy between the feature parameters of each signal source and the fault type quantifies the correlation between the feature parameters of each signal source and the fault type, revealing the importance and contribution of different signal sources in fault diagnosis. Subsequently, the correlation weight of each signal source is calculated based on the mutual information entropy value, transforming the mutual information entropy value into a weight value. This ensures that different signal sources are reasonably weighted according to their correlation with the fault type during subsequent feature fusion, avoiding the influence of subjective factors on feature fusion and improving the accuracy and reliability of the fusion results.

[0070] The module division in this embodiment of the invention is illustrative and represents only one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional modules in the various embodiments of the invention can be integrated into a single processor, exist as separate physical entities, or be integrated into a single module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0071] This embodiment also provides a computer device, which includes a processor and a memory. The memory is used to store a computer program (in this embodiment, the computer program includes a computing component and an iterative component, capable of model calculation and model updating). The computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, and is suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in the computer storage medium to realize the corresponding method flow or corresponding function. The processor described in this embodiment can be used in the operation of a wind turbine gearbox fault diagnosis method.

[0072] This embodiment also provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the wind turbine gearbox fault diagnosis method in the above embodiment.

[0073] This embodiment also provides a computer program product, which includes a computer program that, when executed by a processor, implements the corresponding steps of the wind turbine gearbox fault diagnosis method described in the above embodiment.

[0074] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0075] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0076] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0077] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0078] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for diagnosing gearbox faults in wind turbine generators, characterized in that, Includes the following steps: After collecting and preprocessing multi-source signal data during gearbox operation, feature parameters of the multi-source signal data are extracted. The mutual information entropy value between the feature parameters of the multi-source signal data and the gearbox fault type is calculated in real time, and the correlation weight of the multi-source signal data is obtained by normalization in combination with the feature parameters. Based on the correlation weight, the feature parameters are fused and decided by the improved DS evidence theory to obtain a new evidence body, and a weighted average is performed to obtain a comprehensive fault feature vector. The comprehensive fault feature vector is input into a pre-trained fault classification model to obtain the fault type of the gearbox.

2. The method for diagnosing gearbox faults in a wind turbine as described in claim 1, characterized in that, The multi-source signal data includes: vibration signal, temperature signal, oil abrasive particle concentration signal, and rotational speed signal; The steps of acquiring and preprocessing multi-source signal data during gearbox operation specifically include: After the vibration signal is processed to remove the mean, the amplitude of the vibration signal is normalized using the maximum-minimum algorithm to obtain the processed vibration signal. The number of modes and the penalty factor of the VMD algorithm are determined by the particle swarm optimization algorithm. Based on the number of modes and the penalty factor, the VMD algorithm is used to decompose the processed vibration signal to obtain multiple modal components and the center frequency of each modal component. The modal components with center frequencies greater than a preset frequency threshold are deleted. The Pearson correlation coefficient between the remaining modal components and the processed vibration signal is calculated. The vibration signal is reconstructed using the remaining modal components with Pearson correlation coefficients greater than a preset correlation coefficient threshold to obtain a new vibration signal.

3. The method for diagnosing gearbox faults in a wind turbine generator according to claim 2, characterized in that, It also includes a step of matching the new vibration signal, specifically including: The energy envelope is obtained by performing ITEO calculation on the new vibration signal, and the energy envelope is analyzed by Hilbert transform to obtain the analytical signal. The instantaneous frequency of the analytical signal is calculated by the phase difference method and denoted as the state characteristic frequency. The state characteristic frequency is matched with the theoretical state frequency, and the state label corresponding to the theoretical state frequency is output. The state label includes: normal, worn, and broken gear tooth.

4. The method for diagnosing gearbox faults in a wind turbine as described in claim 1, characterized in that, Based on the aforementioned correlation weights, the steps of fusing and deciding on the feature parameters using improved DS evidence theory to obtain a new evidence body, and then performing a weighted average to obtain a comprehensive fault feature vector, specifically include: Calculate the conflict coefficient between each piece of evidence in the improved DS evidence theory. When the conflict coefficient is greater than a preset conflict threshold, calculate the similarity between each piece of evidence to obtain a credibility factor. Based on the credibility factor, perform a weighted correction on the confidence of each piece of evidence to obtain a corrected basic probability allocation function. Normalize the corrected basic probability allocation function to obtain the corrected piece of evidence. The modified evidence body is fused using the DS synthesis rule to obtain a preliminary fusion result. The preliminary fusion result is then subjected to a consistency check. If the preset consistency condition is met, a new evidence body is obtained. Otherwise, the calculation weight of the credibility factor is readjusted, and the above steps of weighting and correcting the confidence of each evidence body based on the credibility factor are repeated. The feature parameters of the multi-source signal data corresponding to the new evidence are weighted and averaged according to the correlation weight to obtain a comprehensive fault feature vector.

5. The method for diagnosing gearbox faults in a wind turbine generator according to claim 2, characterized in that, It also includes the following steps: The window length of the sliding window is calculated based on the rotational speed signal, and the correlation weight is dynamically updated using a sliding time window mechanism based on the window length.

6. The method for diagnosing gearbox faults in a wind turbine generator according to claim 1, characterized in that, It also includes the step of predicting the remaining life of the gearbox, specifically including: The cumulative running time, design life parameters, and historical degradation data of the gearbox are obtained. Combined with the processed multi-source signal data and the fault type of the gearbox, a remaining life prediction model is constructed based on a long short-term memory network architecture, with the rate of change of the characteristic parameters of the multi-source signal data as the input variable and the remaining life value of the gearbox as the output variable. The feature parameters of the multi-source signal data are divided into time series to generate multiple complete degradation cycle samples. Each degradation cycle sample contains a full life cycle feature parameter sequence from normal state to fault state. The sliding window method is used to perform data augmentation on the full life cycle feature parameter sequence, extract the historical feature change rate of different degradation stages, and establish a mapping relationship between the historical feature change rate and the remaining lifetime. The remaining lifetime prediction model is trained using the mapping relationship between the historical feature change rate and the remaining lifetime to obtain the trained remaining lifetime prediction model. The mean value of the feature parameters of the multi-source signal data is calculated, and the mean value is used as the health status benchmark value. The deviation between the feature parameters of the multi-source signal data and the health status benchmark value is calculated. When the deviation exceeds a preset degradation threshold, the rate of change of the feature parameters of the multi-source signal data is input into the trained remaining life prediction model to obtain the remaining life prediction value. Acquire all multi-source signal data of the current gearbox and integrate them into real-time and historical sequences according to the order of acquisition time; The DTW algorithm is used to align the real-time sequence and the historical sequence to obtain the historical multi-source signal data corresponding to the last multi-source signal data in the real-time sequence. Based on the historical multi-source signal data corresponding to the historical sequence, the actual remaining lifetime is obtained. The deviation rate between the predicted remaining lifetime and the actual remaining lifetime is calculated. Based on the deviation rate, the Kalman filter algorithm is used to dynamically correct the output of the remaining lifetime prediction model and update the predicted remaining lifetime.

7. A wind turbine gearbox fault diagnosis system, characterized in that, include: The data acquisition and preprocessing module is used to acquire multi-source signal data during the operation of the gearbox, preprocess the data, and extract the feature parameters of the multi-source signal data. The comprehensive fault feature vector construction module is used to calculate the mutual information entropy value between the feature parameters of the multi-source signal data and the gearbox fault type in real time, and to normalize the feature parameters to obtain the correlation weight of the multi-source signal data. Based on the correlation weight, the feature parameters are fused and decided by the improved DS evidence theory to obtain a new evidence body, and a weighted average is performed to obtain the comprehensive fault feature vector. The fault diagnosis module is used to input the comprehensive fault feature vector into a pre-trained fault classification model to obtain the fault type of the gearbox.

8. An electronic device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable in the processor. When the processor executes the computer program, it implements the steps of the wind turbine gearbox fault diagnosis method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the wind turbine gearbox fault diagnosis method according to any one of claims 1 to 6.

10. A computer program product, the computer program product comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the wind turbine gearbox fault diagnosis method according to any one of claims 1 to 6.

Images