Wind driven generator gearbox fault detection method based on infrared thermal imaging XGBoost-GRU

Abstract

The invention discloses a wind driven generator gear box fault detection method based on infrared thermal imaging XGBoost-GRU, and relates to the field of gear box detection.The method comprises the steps that the temperature value of a wind driven generator gear box at any moment is obtained based on the infrared imaging technology, and the temperature value of the wind driven generator gear box at any moment is calculated according to the operation parameters of the wind driven generator gear box; establishing an association database between the temperature and the operation parameters; calculating the distance between the operation state of the wind driven generator gear box and the reference operation state by using an association database and a dynamic time rule technology, and obtaining a fault detection result of the wind driven generator gear box; and establishing a fatigue life evaluation model of dynamic characteristics, predicting a failure cycle of the wind driven generator gearbox on the premise of a fault inspection result, and performing operation adjustment on the wind driven generator gearbox. According to the method, the fatigue life and the failure cycle of the gearbox can be predicted, sudden fault shutdown of equipment is effectively avoided, and therefore economic losses caused by sudden faults are reduced.

Description

Technical Field

[0001] The present invention relates to the field of gearbox detection, and more specifically, to a method for detecting faults in a wind turbine gearbox based on infrared thermal imaging XGBoost-GRU. Background Art

[0002] As an important part of renewable energy, the operational reliability of wind turbines directly affects the stability of the power system. However, with the rapid increase in the installed capacity of wind power in various countries, design defects and insufficient operation and maintenance practices have been increasingly exposed, resulting in frequent component failures, causing significant economic losses and safety hazards. Effective operation and maintenance practices are crucial for ensuring the safety and reliability of wind turbines.

[0003] A wind turbine has a complex structure consisting of multiple components, among which the gearbox is a key subsystem connecting the blades and the generator. Faults in the gearbox will directly affect the operating states of various key components in the wind turbine. In addition, the downtime and economic losses caused by gearbox faults are often the primary problems among various components. Therefore, implementing condition monitoring and fault detection for the gearbox is of great significance for improving operational reliability and reducing operation and maintenance costs. The gearbox is a key component of a wind turbine, and its faults can lead to downtime and economic losses.

[0004] Existing gearbox fault detection methods are mainly divided into physical model-based and data-driven methods. Physical model methods are difficult to accurately describe complex gearbox systems, while data-driven methods, although not relying on accurate models, still have problems with insufficient detection accuracy when facing the changing external environment and operating states of wind turbines. Therefore, there is an urgent need for a high-precision gearbox fault detection method that can adapt to different operating states.

[0005] Regarding the problems in the related art, no effective solutions have been proposed yet. Summary of the Invention

[0006] In view of the problems in the related art, the present invention proposes a method for detecting faults in a wind turbine gearbox based on infrared thermal imaging XGBoost-GRU to overcome the above-mentioned technical problems existing in the existing related art.

[0007] To this end, the specific technical solution adopted by the present invention is as follows:

[0008] A method for detecting faults in a wind turbine gearbox based on infrared thermal imaging XGBoost-GRU, comprising:

[0009] Obtaining the temperature value of the wind turbine gearbox at any moment based on infrared imaging technology, and establishing an association database between the temperature and the operating parameters in combination with the operating parameters of the wind turbine gearbox;

[0010] Using the associated database and dynamic time rule technology, calculate the distance between the operating state of the wind turbine gearbox and the reference operating state, and obtain the fault detection result of the wind turbine gearbox;

[0011] Establish a fatigue life assessment model with dynamic characteristics, predict the failure cycle of the wind turbine gearbox based on the fault inspection result, and generate an operation adjustment strategy according to the failure cycle to adjust the operation of the wind turbine gearbox.

[0012] Preferably, using the associated database and dynamic time rule technology, calculating the distance between the operating state of the wind turbine gearbox and the reference operating state, and obtaining the fault detection result of the wind turbine gearbox includes:

[0013] Select monitoring features from the associated database according to the working principle and performance index standard of the wind turbine gearbox, and form an operating state data set by dividing the associated database through a sliding window;

[0014] Use the operating state data set as the dimension to represent the current operating characteristics of the wind turbine gearbox, and use the dynamic time rule technology to calculate the distance between the current operating characteristics and the reference operating state;

[0015] Randomly select several cluster centroids according to the time series data of the operating parameters of the wind turbine gearbox, and calculate the distance error from all arbitrary operating state data to their respective cluster centroids under any number of clusters;

[0016] Based on the distance error, perform unsupervised classification on the operating state of the wind turbine gearbox, and obtain the fault detection result of the wind turbine gearbox according to the classification result and machine learning technology.

[0017] Preferably, based on the distance error, perform unsupervised classification on the operating state of the wind turbine gearbox, and obtaining the fault detection result of the wind turbine gearbox according to the classification result and machine learning technology includes:

[0018] Plot the error distances corresponding to any number of clusters as a distance curve, observe the inflection point where the descent speed of the distance error in the distance curve reaches the requirement, and use the distance number corresponding to the inflection point as the optimal number of clusters;

[0019] Iteratively optimize the cluster assignment based on the optimal number of clusters, and assign the operating state data set to the nearest cluster according to the distance between the current operating characteristics and the reference operating state, and recalculate the centroid of each cluster;

[0020] Terminate the iteration when the cluster assignment does not change or reaches the maximum number of iterations, and form a consistency matrix according to the frequency of the operating state data set being assigned to the same cluster in all iterations;

[0021] Quantify the stability of the consistency matrix using the dispersion coefficient, judge the rationality of the clustering result based on the stability result, and use the consistency matrix as the classification result for any operating state according to the judgment result;

[0022] Combine the classification result, associated data set with machine learning to analyze the probability of any operating state category of the wind turbine gearbox, and use a gated recurrent unit to obtain the fault detection result of the wind turbine gearbox.

[0023] Preferably, combining the classification result, associated data set with machine learning to analyze the probability of any operating state category of the wind turbine gearbox, and using a gated recurrent unit to obtain the fault detection result of the wind turbine gearbox includes:

[0024] Train a machine learning classification model based on the sampling weight allocation strategy, and input the classification result and associated data set into the machine learning classification model to output the probability that the sample belongs to each clustering category;

[0025] Use a gated recurrent unit and the operating state category of the wind turbine gearbox to perform probability verification and post-processing on the probability result, and calculate the fault confidence of the wind turbine gearbox according to the processing result and the weighting parameter;

[0026] Compare the fault confidence with the fault threshold, and obtain the fault detection result of the wind turbine gearbox based on the comparison result.

[0027] Preferably, establish a fatigue life assessment model with dynamic characteristics, predict the failure period of the wind turbine gearbox under the premise of the fault inspection result, and generate an operation adjustment strategy according to the failure period to adjust the operation of the wind turbine gearbox, including:

[0028] Use the fault detection result and the wind turbine gearbox structure information to generate a three-dimensional gearbox model, and generate a numerical parameter set including temperature and friction and apply it to the three-dimensional gearbox model;

[0029] Determine the fatigue life assessment model according to the application result, and perform dynamic coupling simulation on the fatigue life assessment model to obtain the load history of the wind turbine gearbox;

[0030] Add the coupling shaft of the wind turbine gearbox as a node attribute to the fatigue life assessment model, and perform transient analysis to obtain the equivalent structural stress response history of the wind turbine gearbox;

[0031] Predict the working development trend of the wind turbine gearbox under the premise of the fault inspection result based on the load history and the equivalent structural stress response history, and judge the failure period of the wind turbine gearbox;

[0032] Analyze the real-time working state of the wind turbine gearbox in any time period according to the failure cycle, and generate an operation adjustment strategy based on the analysis results to adjust the operation of the wind turbine gearbox.

[0033] Preferably, determine the fatigue life assessment model according to the application results, and perform dynamic coupling simulation on the fatigue life assessment model to obtain the load history of the wind turbine gearbox, including:

[0034] Determine the fatigue life assessment model based on the application results of the numerical parameter set and the three-dimensional gearbox model, and define the dynamic coupling simulation step size and simulation result output frequency of the fatigue life assessment model;

[0035] Perform dynamic coupling simulation of the wind turbine gearbox under any temperature and friction conditions according to the defined results and the numerical parameter set, and obtain the dynamic response curve of the wind turbine gearbox based on the simulation results;

[0036] Obtain the ratio between the change in thermal load and the meshing amplitude of the wind turbine gearbox according to the dynamic response curve, define the distribution evaluation index of the load space based on the ratio result, and generate the load distribution curve;

[0037] Use entropy value to analyze the distribution discreteness of the distribution curve, generate a line graph to obtain the dynamic change of the internal gear meshing force of the wind turbine gearbox, and use it as the load history of the wind turbine gearbox.

[0038] Preferably, predict the working development trend of the wind turbine gearbox based on the load history and the equivalent structural stress response history under the premise of the fault inspection results, and judge the failure cycle of the wind turbine gearbox, including:

[0039] Use a multi-layer perceptron to screen out the influencing factors strongly related to the failure rate of the wind turbine gearbox from the load history and the equivalent structural stress response history, and input the influencing factors into the single prediction model to output the initial working development trend of the wind turbine gearbox;

[0040] Generate a mean square error weighted combination model according to the weighted coefficient and discount factor of the single prediction model, integrate the initial working development trend, and output the secondary working development trend of the combination model;

[0041] Use the quantum harmony search algorithm to dynamically optimize the weighted coefficient of the mean square error weighted combination model to obtain a comprehensive prediction model, and use the comprehensive prediction model to predict the working development trend of the wind turbine gearbox;

[0042] Based on the predicted working development trend and the cyclic fatigue life criterion, judge the fatigue damage accumulation of the wind turbine gearbox, and obtain the failure cycle of the wind turbine gearbox according to the fatigue damage accumulation.

[0043] Preferably, the quantum harmony search algorithm is used to dynamically optimize the weighting coefficients of the mean square error weighted combination model, and the comprehensive prediction model obtained includes:

[0044] Based on the quantum coding harmony memory library, the weighting coefficients in the mean square error weighted combination model are defined as quantum harmonies, and the quantum superposition state of the quantum harmonies is determined according to the quantum state information coding technology;

[0045] The quantum superposition state is measured by the collapse theory to determine the corresponding ground state information, the perturbation bandwidth is determined by generating a perturbation adjustment frequency according to the ground state information, and the quantum harmony is slightly perturbed to dynamically optimize the quantum harmony;

[0046] Update the quantum coding harmony memory library according to the optimized quantum harmony, and select the quantum harmony that satisfies the weighting coefficient objective function from the quantum coding harmony memory library after reaching the maximum number of iterations;

[0047] The quantum harmony that meets the requirements is used as the optimal weighting coefficient, the optimal solution of the discount factor matrix is output according to the weighting coefficient, and the mean square error weighted combination model is updated using the optimal weighting coefficient and the optimal discount factor matrix to obtain the comprehensive prediction model.

[0048] Preferably, the calculation formula for the perturbation bandwidth is:

[0049]

[0050] In the formula, S h ′ represents the perturbation bandwidth after perturbation adjustment, S h represents the perturbation bandwidth without perturbation, rand(0,1) represents a random number in the range of 0 to 1, g represents the golden ratio, d represents the perturbation adjustment frequency, K represents the minor perturbation amount, S up represents the upper bound of the quantum harmony, S down represents the lower bound of the quantum harmony.

[0051] Preferably, the calculation formula for the optimal weighting coefficient is:

[0052]

[0053] In the formula, E o represents the optimal weighting coefficient of the o-th single prediction model, A represents the prediction period, c represents the total number of single prediction models, y a represents the true value of the initial working development trend of the wind turbine gearbox output at the a-th time, represents the discount factor of the o-th single prediction model at the a-th time, represents the predicted value of the initial working development trend of the wind turbine gearbox output by the o-th single prediction model at the a-th time.

[0054] The beneficial effects of the present invention are as follows:

[0055] 1. Through the infrared thermal imaging technology, the present invention can capture the temperature changes of the gearbox in real time. Compared with traditional vibration monitoring, it is more sensitive to slight temperature rise anomalies of the gearbox, can detect abnormal trends at an early stage of the fault. At the same time, by combining the powerful feature selection ability of XGBoost and the time series modeling ability of GRU (Gated Recurrent Unit), the prediction of the fault state is more accurate, reducing false alarms and missed alarms. It can not only detect the current operating state, but also calculate the deviation from the reference operating state based on the dynamic time rule technology, so as to issue a warning before the fault occurs, avoiding equipment downtime or damage caused by sudden faults. At the same time, combined with the dynamic characteristic fatigue life assessment model, it can not only evaluate the current operating state, but also predict the future health status, making the operation adjustment of the wind turbine more scientific and reasonable, and extending the service life.

[0056] 2. By using unsupervised clustering to divide the operating state, the present invention can adapt to the changing operating environment of the wind turbine, reducing the interference of the external environment on the detection of fault signals of the wind turbine; when classifying the fault state, a sampling weight allocation strategy is adopted to solve the problem of sample imbalance and improve the accuracy of the classification model; by combining the classification model and the regression model, high-precision detection of gearbox faults is achieved; when performing on-line detection of wind turbine gearbox faults, the fault detection results are generated through probability weighting, providing more detailed and accurate fault information, and thus being applicable to the fault detection of wind turbine gearboxes, capable of effectively identifying abnormal behaviors of the gearbox in different operating states and warning potential faults in advance.

[0057] 3. By combining technologies such as three-dimensional model simulation, fatigue life assessment, load history analysis, and equivalent structural stress response, the present invention can not only achieve precise monitoring of the wind turbine, but also dynamically adjust its operating state, enabling the combination of actual operating data, and combining key parameters such as the temperature, friction, and load of the gearbox to accurately predict the fatigue life and failure cycle of the gearbox, effectively avoiding sudden fault shutdown of the equipment, thereby reducing the economic losses caused by sudden faults. Description of the Drawings

[0058] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0059] Figure 1 is a flowchart of a method for detecting faults in a wind turbine gearbox based on infrared thermal imaging XGBoost-GRU according to an embodiment of the present invention;

[0060] Figure 2 It is a schematic structural diagram of a fault detection method for a wind turbine gearbox based on infrared thermal imaging XGBoost-GRU according to an embodiment of the present invention. Detailed implementation manners

[0061] To further illustrate each embodiment, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. They are mainly used to illustrate the embodiments and can be combined with the relevant descriptions in the specification to explain the operating principle of the embodiments. With reference to these contents, those of ordinary skill in the art should be able to understand other possible implementation manners and the advantages of the present invention.

[0062] According to an embodiment of the present invention, there is provided a fault detection method for a wind turbine gearbox based on infrared thermal imaging XGBoost-GRU.

[0063] Now, the present invention will be further described in conjunction with the accompanying drawings and specific implementation manners. As Figure 1 shown, the fault detection method for a wind turbine gearbox based on infrared thermal imaging XGBoost-GRU according to an embodiment of the present invention includes:

[0064] S1, based on infrared imaging technology, obtain the temperature value of the wind turbine gearbox at any time, and combine with the operating parameters of the wind turbine gearbox to establish an association database between temperature and operating parameters.

[0065] It should be explained that when establishing the association database between temperature and operating parameters, the temperature data of the wind turbine gearbox can be obtained through infrared thermal imaging technology, focusing on the temperature of each part of the gearbox (such as gears, bearings, couplings, etc.); the operating parameter data includes the operating state of the wind turbine, wind speed, power output, rotation speed, load, gearbox vibration, oil temperature, oil pressure, working time, etc. At the same time, according to the working environment of the gearbox, select a suitable infrared thermal imager to ensure that the temperature change on the surface of the gearbox can be clearly captured, determine the position of the gearbox components to be monitored, usually focusing on the areas with intensive heat sources such as gears, bearings, and couplings, and regularly or real-time collect the temperature data of the gearbox using the infrared thermal imager according to the operating cycle of the wind turbine; design the database architecture, clarify the storage method of each data item, data fields (such as temperature, wind speed, rotation speed, load, etc.), time stamp, etc., and a relational database (such as MySQL, PostgreSQL) or a non-relational database (such as MongoDB) can be used.

[0066] S2, use the association database and dynamic time rule technology to calculate the distance between the operating state of the wind turbine gearbox and the reference operating state, and obtain the fault detection result of the wind turbine gearbox.

[0067] In one embodiment, by using an association database and dynamic time rule technology, calculating the distance between the operating state of a wind turbine gearbox and the reference operating state, and obtaining the fault detection result of the wind turbine gearbox includes:

[0068] Select monitoring features from the association database according to the working principle and performance index standard of the wind turbine gearbox, and form an operating state data set by segmenting the association database through a sliding window;

[0069] Use the operating state data set as a dimension to represent the current operating characteristics of the wind turbine gearbox, and use dynamic time rule technology to calculate the distance between the current operating characteristics and the reference operating state;

[0070] Randomly select several cluster centroids according to the time series data of the operating parameters of the wind turbine gearbox, and calculate the distance error from all arbitrary operating state data to their respective cluster centroids under any number of clusters;

[0071] Perform unsupervised classification on the operating state of the wind turbine gearbox based on the distance error, and obtain the fault detection result of the wind turbine gearbox according to the classification result and machine learning technology.

[0072] Specifically, when performing unsupervised classification on the operating state of the wind turbine gearbox based on the distance error and obtaining the fault detection result of the wind turbine gearbox according to the classification result and machine learning technology, the error distance corresponding to any number of clusters can be plotted as a distance curve, and the inflection point where the descent speed of the distance error within the distance curve meets the requirement is observed, and the distance number corresponding to the inflection point is used as the optimal number of clusters; perform iterative optimization on the cluster assignment based on the optimal number of clusters, and allocate the operating state data set to the nearest cluster according to the distance between the current operating characteristics and the reference operating state, and recalculate the centroid of each cluster; terminate the iteration when the cluster assignment does not change or reaches the maximum number of iterations, and form a consistency matrix according to the frequency of the operating state data set being assigned to the same cluster in all iterations; use the coefficient of dispersion to quantify the stability of the consistency matrix, judge the rationality of the clustering result based on the stability result, and use the consistency matrix as the classification result of any operating state according to the judgment result; combine the classification result, the associated data set and machine learning to analyze the probability of any operating state category of the wind turbine gearbox, and use a gated recurrent unit to obtain the fault detection result of the wind turbine gearbox.

[0073] Specifically, when analyzing the probability of any operating state category of a wind turbine gearbox by combining classification results, associated datasets, and machine learning, and using a gated recurrent unit to obtain the fault detection result of the wind turbine gearbox, a machine learning classification model can be trained based on a sampling weight allocation strategy, and the classification results and associated datasets can be input into the machine learning classification model to output the probability that a sample belongs to each clustering category; the gated recurrent unit and the operating state category of the wind turbine gearbox are used to perform probability verification and post-processing on the probability results, and the fault confidence of the wind turbine gearbox is calculated according to the processing results and weighted parameters; the fault confidence is compared with the fault threshold, and the fault detection result of the wind turbine gearbox is obtained based on the comparison result.

[0074] It should be noted that as Figure 2 shown, in the process of fault detection of a wind turbine gearbox, it is mainly divided into four steps:

[0075] Step 1: Use infrared thermal imaging technology to monitor the temperature changes of different parts of the gearbox in real time. Infrared thermal imaging sensors are arranged at key parts such as the bearing seat and gear meshing area of the gearbox to continuously collect temperature data, record the temperature values of different parts at each moment, and at the same time synchronously collect operating parameters such as the load and speed of the gearbox to establish an associated database of temperature changes and parameters such as the load and speed of the gearbox.

[0076] Step 2: Propose an operating state classification model for the fan gearbox. Select the wind speed, generator speed, and output active power of the wind turbine as operating state parameters, calculate the distance between each sample and the reference sample through the dynamic time warping (DTW) algorithm, and use the K-means clustering algorithm to perform unsupervised classification on different operating states.

[0077] Step 3: Conduct classification model training. Based on the classification results, train the classification model through XGBoost-GRU to determine the probability that a real-time sample belongs to different operating state categories, and design a sampling weight allocation strategy to solve the problem of sample imbalance. For each operating state category, train a regression model to learn the normal behavior of the gearbox.

[0078] Step 4: Complete the online fault detection of the fan gearbox. For real-time data samples, first determine the probability that they belong to different operating state categories through the classification model, then generate predicted values based on the regression model of each operating state category, and perform fault judgment by comparing the residual with the threshold. Finally, the fault detection result is obtained through weighted aggregation.

[0079] In step 2, the sample set is the time series data collected for the operating parameters of the gearbox, including temperature, wind speed, generator speed, active power, etc. It is a time series sample set formed by dividing the data set through a sliding window. The reference sample is the operating state parameters of the gearbox measured when the wind turbine operates at the rated power under normal conditions.

[0080] Specifically, the calculation formula for the dynamic time warping (DTW) distance is:

[0081]

[0082] In the formula, i represents the corresponding sample number of the operating state data set, j represents the reference sample, W represents the warping path, K represents the length of the warping path, and X φ represents the normalized operating state parameters.

[0083] Among them, the temperature measured by the infrared thermal imaging sensor installed in the bearing can reflect the operating state of the gearbox. Therefore, the bearing temperature is regarded as the target variable; three variables, namely wind speed, active power, and generator speed, are selected as the parameters representing the operating state for clustering. For the baseline sample under steady wind conditions, the sliding window length is 10. By calculating the DTW distance between each sample and the baseline sample, the K-means clustering algorithm is used to achieve unsupervised partitioning. The implementation steps are as follows:

[0084] (1) Randomly select n initial clustering centers (cluster centroids) according to the time series data of the gearbox operating parameters. By calculating the sum of squared errors SSE of all data points to their respective cluster centroids under different numbers of clusters n:

[0085]

[0086] Among them, G z represents the z-th cluster, μ z represents the centroid of this cluster, and x represents the data point;

[0087] (2) Plot the SSE corresponding to different n values as a curve, observe the inflection point where the SSE decline rate significantly slows down in the graph, and the corresponding n is the optimal number of clusters;

[0088] (3) Iteratively optimize the cluster assignment. According to the DTW distance, assign each sample to the nearest cluster and recalculate the centroid (mean) of each cluster;

[0089] (4) Terminate the iteration when the cluster assignment no longer changes or reaches the maximum number of iterations;

[0090] (5) Evaluate the clustering stability by calculating the consistency matrix and dispersion coefficient of the classification results of the operating state parameters. According to the frequency of each pair of samples being assigned to the same cluster in all iterations, form the consistency matrix Cxy :

[0091]

[0092] Matrix element C xy represents the probability that samples x and y are in the same class, with a value range of [0, 1]. The closer the value is to 1, the more stable the clustering results of the two samples. The Dispersion Score is used to quantify the stability of the consistency matrix, and the formula is:

[0093]

[0094] where, the closer ρ is to 1, the more consistent the clustering results of all iterations; approaching 0 indicates randomness;

[0095] (6) Use the clustering results that pass the test as the class labels for different operating states, and use them for the training of the subsequent XGBoost classification model to solve the problem of sample imbalance.

[0096] By gradually increasing the number of clusters, count the sum of squared errors (SSE) of the clustering results to determine the final number of clusters. The number of clusters is determined as the smallest number with no significant change in SSE, which is set to 5 in this embodiment. According to the final clustering results on the power curve, even when the wind speeds are similar, differences in power generation due to power curtailment schemes, abnormal operations, etc. will lead to different classes.

[0097] Then, use the clustering results as labels to train the XGBoost classification model. Count the number of samples in different clusters and assign corresponding sampling weights. Based on each cluster, train a GRU regression model to describe the expected operating state of the gearbox. For each sample in this case, present the prediction result according to the class with the highest probability, and obtain the actual measured values, predicted values of the training set, and predicted values of the test set. Determine the probabilities of different operating state classes of the wind turbine gearbox according to the number of clusters. The specific calculation steps are as follows:

[0098] (1) Input the feature values, and input the state class labels obtained by clustering through the above steps, normalized operating parameters, the associated data of temperature change and load / speed into the XGBoost classification model for training;

[0099] (2) After the real-time data is input into the classification model, XGBoost outputs the probability P of each sample belonging to each clustering class m , and the formula is the Softmax function:

[0100]

[0101] where, f m(x) represents the prediction score of the m-th type of regression model, and R represents the total number of categories.

[0102] (3) Modify the probability by combining infrared thermal imaging temperature data: If the correlation between temperature and load / speed deviates from the historical pattern, dynamically adjust the probability weight.

[0103] (4) Probability verification and post-processing, calculate the fault confidence through weighted residuals and temperature outliers:

[0104]

[0105] Among them, I m represents the fault judgment result of the m-th type of regression model;

[0106] (5) Threshold determination and alarm triggering, the residual threshold is dynamically adjusted by the PauTa criterion:

[0107] Threshold m = μ m + 3σ m ;

[0108] Among them, μ m represents the mean of the residuals of the m-th type of regression model, and σ m represents the standard deviation of the residuals of the m-th type of regression model. When the proportion of residual or temperature outliers exceeds the threshold, probability correction and fault alarm are triggered.

[0109] In the model training step, the classification model uses the eXtreme Gradient Boosting (XGBoost) algorithm, and the sampling weight allocation strategy is calculated by the following formula:

[0110]

[0111] Among them, represents the number of classification categories, and N i represents the number of samples in the i-th category.

[0112] At the same time, the regression model uses a gated recurrent unit (GRU) network, and the calculation formulas for its update gate and reset gate are:

[0113]

[0114] Among them, z t , r t , h t represent the update gate output, reset gate output, candidate hidden state, and final hidden state respectively, and W z , W r , W h represent the weight matrices, and Uz , U r , U h represents the state transition matrix, b z , b r , b h represents the bias term.

[0115] In the online detection step, the confidence level of the fault detection result is calculated by the following formula:

[0116]

[0117] where, P m represents the probability that each sample belongs to each clustering category, I m represents the fault judgment result of the m-th regression model.

[0118] In the operating state classification step, the length W of the sliding window is set to 10 minutes according to the operating characteristics of the wind turbine.

[0119] It should be noted that the input variables of the regression model include the oil temperature of the gearbox, the inlet pressure of the gearbox, the outlet pressure of the gearbox, the ambient temperature, the wind speed, the wind direction, the system state, the active power, the reactive power, the generator speed, the power factor, the yaw angle, the yaw nacelle position, the nacelle control temperature, the nacelle temperature, and the pitch angles of blades 1, 2, and 3.

[0120] Based on the model, the predicted residuals and thresholds are obtained, aiming to display the model prediction results of each sample under the category with the highest probability. In fact, for each real-time sample, the models trained under each category will make real-time predictions and warning judgments, and the final detection result is the probability weighted sum divided by category. To avoid frequent false alarms caused by outliers, a percentage overrun criterion is designed, and the parameter Pd% is set to 25%. The first detection time is 17 days earlier than the confirmation time. The detailed result of the first alarm point has a final detection confidence level of 0.94, which represents a relatively high value identified as a fault.

[0121] To verify the superiority of the method proposed in the embodiment, a comparative analysis of the traditional method was carried out. The key to the fault diagnosis of the wind power gearbox is the state classification step and the sampling weight allocation strategy. These steps were deleted in this embodiment for verification. The XGBoost-GRU method means deleting the sampling weight allocation strategy, while the GRU method means training only one regression model for all samples. The root mean square error (RMSE) was used as the index for the prediction results of all 31 WTs. For the XGBoost-GRU method, the sample imbalance problem will affect the classification performance. For the categories with fewer samples, the increase in misclassification will lead to a decline in the prediction performance. For the GRU method, the model will pay more attention to the main operating states and ignore some states with fewer occurrences, thus affecting the prediction performance. The method proposed in this embodiment can achieve the optimal prediction results on all wind power gearbox fault data sets, reaching the optimal detection results. The improvement of the prediction accuracy helps to more effectively distinguish faults and noises, thereby improving the detection accuracy. In addition, the proposed method can evaluate each operating state more evenly, and thus better respond to unknown potential faults. The analysis and comparison experiments prove the effectiveness and superiority of the proposed method.

[0122] S3. Establish a fatigue life assessment model of dynamic characteristics, predict the failure cycle of the wind turbine gearbox based on the fault inspection results, and generate an operation adjustment strategy according to the failure cycle to adjust the operation of the wind turbine gearbox.

[0123] In one embodiment, establishing a fatigue life assessment model of dynamic characteristics, predicting the failure cycle of the wind turbine gearbox based on the fault inspection results, and generating an operation adjustment strategy according to the failure cycle to adjust the operation of the wind turbine gearbox includes:

[0124] Using the fault detection results and the structural information of the wind turbine gearbox to generate a three-dimensional gearbox model, and generating a numerical parameter set including temperature and friction and applying it to the three-dimensional gearbox model;

[0125] Determine the fatigue life assessment model according to the application results, and perform dynamic coupling simulation on the fatigue life assessment model to obtain the load history of the wind turbine gearbox;

[0126] Take the coupling shaft of the wind turbine gearbox as a node attribute and add it to the fatigue life assessment model, and perform transient analysis to obtain the equivalent structural stress response history of the wind turbine gearbox;

[0127] Based on the load history and the equivalent structural stress response history, predict the working development trend of the wind turbine gearbox based on the fault inspection results, and judge the failure cycle of the wind turbine gearbox;

[0128] Analyze the real-time working state of the wind turbine gearbox in any time period according to the failure cycle, and generate an operation adjustment strategy based on the analysis results to adjust the operation of the wind turbine gearbox.

[0129] Specifically, when generating a three-dimensional gearbox model using the fault detection results and the wind turbine gearbox structure information, and generating a set of numerical parameters including temperature and friction and applying them to the three-dimensional gearbox model, the detailed structure information of the gearbox can be obtained first, including the geometric shape, size, material properties, construction details (such as gears, bearings, gear transmission ratios, etc.) of the gearbox. The structure data can be obtained through CAD (Computer-Aided Design) software or according to the technical drawings provided by the manufacturer. Obtain the current or historical fault detection results of the gearbox from the fault detection results. The fault data may involve problems such as wear, cracks, and looseness of different components, and can be obtained through means such as vibration monitoring, temperature monitoring, and noise analysis. According to the aforementioned obtained gearbox structure information (such as gears, bearings, housings, etc.), use CAD modeling software (such as SolidWorks, CATIA, AutoCAD, etc.) to construct an accurate three-dimensional gearbox model. The model needs to include key details such as the tooth profile of the gear, gear clearance, bearing position, and housing structure. In addition to the geometric shape, it also needs to include the characteristics of different materials (such as gear steel, bearing materials, etc.) and the contact relationship and assembly method between components. Ensure the accuracy of the model for subsequent simulation and analysis.

[0130] Combined with the temperature data of each part of the gearbox collected by devices such as infrared thermal imaging technology and temperature sensors, temperature data can be collected under different operating conditions (such as different loads, speeds, ambient temperatures, etc.) to form a numerical set of temperature changes. According to the information such as the materials of gears and bearings, lubrication conditions, and loads, estimate the friction mechanics data inside the gearbox. The friction parameters include but are not limited to the friction coefficient, contact stress, lubrication conditions (such as oil film thickness), sliding speed, etc. The friction data can be obtained through experimental data, material properties, or tribological models (such as the Berlin model, Reynolds equation, etc.). Introduce the heat conduction equation into the three-dimensional gearbox model to simulate the temperature field distribution inside the gearbox. The temperature distribution of the gearbox may change with different working states (such as load, speed, wind speed, etc.), especially the temperature changes at the gear transmission components and bearings. Input the temperature data obtained by infrared thermal imaging technology or other temperature measurement devices into the model, and set temperature boundary conditions or heat sources at the key parts of the gearbox (such as the gear meshing area, bearings, couplings, etc.).

[0131] According to the tribology principle, combined with friction coefficient, contact area, contact stress, lubrication conditions, etc., generate the friction parameters of the working components of the gearbox (such as gears, bearings, couplings, etc.), and use simulation methods such as finite element analysis (FEA) or multibody dynamics (MBD) to introduce the friction mechanics model into the 3D model of the gearbox to simulate the friction behavior under different working conditions.

[0132] Specifically, when determining the fatigue life assessment model based on the application results and performing dynamic coupling simulation on the fatigue life assessment model to obtain the load history of the wind turbine gearbox, the fatigue life assessment model can be determined based on the application results of the numerical parameter set and the 3D gearbox model, and the dynamic coupling simulation step size and simulation result output frequency of the fatigue life assessment model can be defined; perform dynamic coupling simulation of the wind turbine gearbox under any temperature and friction conditions according to the defined results and the numerical parameter set, and obtain the dynamic response curve of the wind turbine gearbox based on the simulation results; obtain the ratio between the change in thermal load and the meshing amplitude of the wind turbine gearbox according to the dynamic response curve, define the distribution evaluation index of the load space based on the ratio result, and generate the load distribution curve; use entropy value to analyze the distribution discreteness of the distribution curve, generate a line chart to obtain the dynamic change of the internal gear meshing force of the wind turbine gearbox, which is used as the load history of the wind turbine gearbox.

[0133] It should be explained that when obtaining the load history of the wind turbine gearbox, the simulation step size refers to the time interval at each simulation moment during the dynamic coupling simulation. Set an appropriate simulation step size to ensure that the dynamic changes during the simulation are accurately captured, and at the same time avoid the influence of too long or too short time steps on the result accuracy and calculation efficiency; according to the operating cycle, vibration frequency and other characteristics of the gearbox, set the simulation step size. For example, the meshing cycle of the gear and the periodicity of temperature change can be used as a reference for step size setting, and at the same time define the frequency of simulation result output, that is, how often to output the simulation results. A higher output frequency can analyze the dynamic behavior of the gearbox in more detail, but it will increase the calculation burden. Usually, the output frequency should be set as an integer multiple of the simulation step size.

[0134] During the process of defining the dynamic coupling simulation conditions, based on the previously obtained temperature and friction numerical parameter set, set conditions such as the operating temperature and friction coefficient of the gearbox to simulate the dynamic response of the gearbox under these conditions. At the same time, according to the operating conditions of the wind turbine gearbox, define the load changes (load, speed, etc.), and combine the temperature and friction conditions to perform dynamic coupling simulation, and use simulation software such as finite element analysis (FEA) or multibody dynamics simulation (MBD) (such as ANSYS, ABAQUS, Simulink, etc.) to perform coupling simulation to simulate the dynamic response of the gearbox under different temperature, friction and load conditions.

[0135] The simulation results will provide the dynamic response curves of the gearbox under different operating conditions, mainly including:

[0136] Gear meshing amplitude: The amplitude change of the gear meshing force can reflect the load fluctuation of the gearbox under different loads and friction conditions;

[0137] Heat load change: Under different operating temperatures, the heat load change of the gearbox may cause material expansion, stress change, etc., affecting the fatigue life of the gearbox;

[0138] Obtaining the dynamic response curve: The simulation software can output the dynamic response curve of the gearbox, and these curves can represent the changes of temperature, stress, friction force, meshing force, etc. of the gearbox over time during the simulation

[0139] By analyzing the dynamic response curve, obtain the ratio between the heat load change and the gear meshing amplitude during the operation of the gearbox. This ratio can reflect the relationship between temperature and load during the operation of the gearbox, helping to judge which operating conditions may cause overheating, overload or gear damage. The ratio is calculated by the following method:

[0140] Extract the heat load of each component (such as gears, bearings, etc.) of the gearbox at different time points from the simulation results, and at the same time extract the force or vibration amplitude during gear meshing, calculate the ratio of the heat load to the meshing amplitude, and form a ratio curve.

[0141] The load space refers to the distribution of forces and stresses on each component (such as gears, bearings, etc.) of the gear transmission system during the operation of the wind turbine gearbox. By analyzing the gear meshing force and heat load, the distribution of the load space can be defined. Through ratio analysis, an evaluation index is defined to quantify the operating state of the gearbox under different load conditions. Common evaluation indexes include:

[0142] Load concentration: Reflects the degree of concentration of the load distribution on each component inside the gearbox;

[0143] Load uniformity: Represents the uniformity of the load distribution of the gearbox;

[0144] Generate the load distribution curve: According to the ratio and evaluation index, generate the load distribution curve, reflecting the load change of different components of the gearbox during operation.

[0145] Use entropy value to analyze the discreteness of the distribution curve and generate a line graph. Entropy value analysis is a method used to measure the discreteness or uncertainty of the system distribution. By analyzing the entropy value of the load distribution curve, the degree of discreteness of the gearbox load distribution can be obtained.

[0146] A high entropy value indicates that the load distribution of the gearbox is relatively discrete, and there may be peak load points, leading to local overload; a low entropy value indicates that the load distribution of the gearbox is relatively uniform and the load change is relatively stable.

[0147] Convert the results of entropy value analysis into a line chart to show the dynamic changes of the load of the wind turbine gearbox under different working conditions. Based on the above dynamic response analysis, ratio calculation, load distribution curve and entropy value analysis, finally generate the load history of the wind turbine gearbox. This history reflects the load change trend of the wind turbine gearbox during long-term operation.

[0148] At the same time, in the process of obtaining the equivalent structural stress response history, a suitable fatigue life assessment method can be selected according to the structural characteristics of the wind turbine gearbox. For example, select the S-N curve method (a classic method based on the stress-life relationship), Miner's rule (based on the cumulative damage theory) or the stress-strain method (considering the specific stress-strain characteristics of the material). At the same time, determine the key parameters that need to be input, including: the material properties of the gearbox (such as yield strength, ultimate strength, elastic modulus, etc.); the geometric parameters of each component of the gearbox (such as the size and shape of gears, bearings, couplings, etc.); the working environment (temperature, load, etc.); kinematic parameters (rotation speed, angular velocity, bearing rotation, etc.).

[0149] As a key element connecting two rotating components, the coupling plays a crucial role in the power transmission of the gearbox. When adding the coupling as a node attribute to the model, it is necessary to clarify the working characteristics, material properties of the coupling and its impact on the dynamic response of the gearbox. In the fatigue life assessment model, the node attributes of the coupling include its stiffness, damping, transmission efficiency, torque transmission capacity, etc. It is necessary to model its stiffness and dynamic characteristics according to the actual structure and material. At the same time, the friction characteristics of the coupling and its impact on the gearbox system (such as the heat load and friction force that may be generated during torque transmission) also need to be considered in the model.

[0150] Accurately set the geometric dimensions and connection methods (such as keyway connection, press-fit connection, etc.) of the coupling in the three-dimensional model, and ensure that the dynamic characteristics of the coupling can be accurately described. Transient analysis refers to the process of calculating the system response at different time points, mainly used to simulate the dynamic behavior of the system in the short term (such as instantaneously or within several cycles). In the case of a wind turbine gearbox, transient analysis can be used to simulate the short-term dynamic response of the gearbox under conditions such as load, rotation speed, temperature, friction, etc., and determine the transient load of the gearbox, including the load, rotation speed change, wind speed change, etc. of the wind turbine. Since the working load and rotation speed of the wind turbine will change at any time, transient analysis needs to calculate the stress response under these transient loads, and at the same time, the dynamic response of the coupling needs to be considered, especially the torque fluctuations and vibrations that the coupling may cause during high-speed operation.

[0151] During the process of dynamic simulation and stress response calculation, simulation software such as multi-body dynamics (MBD) or finite element analysis (FEA) (such as ANSYS, Abaqus, Simulink, etc.) is used for dynamic simulation to simulate the dynamic behavior of the gearbox under different transient loads. During the simulation, the coupling is set as the node for power transmission, and through its interaction with other components (gears, bearings, etc.), the response of the entire system is calculated.

[0152] The steps of transient analysis are as follows:

[0153] Loading conditions: Input transient loads (such as torque, output load of wind turbines, rotational speed fluctuations, etc.);

[0154] Time step setting: According to the simulation requirements, select an appropriate time step for dynamic simulation to ensure that the dynamic response can be accurately captured at each time step;

[0155] Calculate the dynamic response: The simulation software calculates the dynamic response results such as stress, strain, vibration, and thermal load of the wind turbine gearbox, especially in the coupling and gear meshing areas.

[0156] After the simulation is completed, the obtained results may include the stress data of different components of the gearbox. In order to evaluate the fatigue life, it is necessary to calculate the equivalent structural stress, which is the comprehensive stress after considering different stress components (such as shear stress, normal stress, bending stress, etc.). The von Mises stress (or equivalent stress) can be used to represent the comprehensive stress under multi-axial stress. Through the results of transient analysis, the stress response curves of each component of the gearbox (especially the gear and coupling parts) can be plotted. The stress history can help understand the working state of the gearbox under different transient loads, especially the response of the coupling. Based on the obtained equivalent stress response history, methods such as Miner's rule are used to evaluate the fatigue damage accumulation of the gearbox under different loads. According to the stress history under different working conditions, calculate the fatigue damage accumulation of each load cycle on the structure of the gearbox and predict the fatigue life.

[0157] Specifically, when predicting the working development trend of a wind turbine gearbox based on the load history and equivalent structural stress response history and judging the failure cycle of the wind turbine gearbox under the premise of the fault inspection result, a multi-layer perceptron can be used to screen out the influencing factors strongly related to the failure rate of the wind turbine gearbox from the load history and equivalent structural stress response history, and input the influencing factors into a single prediction model to output the initial working development trend of the wind turbine gearbox; generate a mean square error weighted combination model according to the weighted coefficients and discount factors of the single prediction model, integrate the initial working development trend, and output the secondary working development trend of the combination model; use the quantum harmony search algorithm to dynamically optimize the weighted coefficients of the mean square error weighted combination model to obtain a comprehensive prediction model, and use the comprehensive prediction model to predict the working development trend of the wind turbine gearbox; based on the predicted working development trend and the cyclic fatigue life criterion, judge the fatigue damage accumulation of the wind turbine gearbox, and obtain the failure cycle of the wind turbine gearbox according to the fatigue damage accumulation.

[0158] It should be explained that the single prediction model can include a Long Short-term Memory (LSTM) model, a Multivariable Linear Regression (MLR) model, a Least Squares Support Vector Machine (LSSVM) model, and a Random Forest (RF) model.

[0159] Specifically, when using the quantum harmony search algorithm to dynamically optimize the weighted coefficients of the mean square error weighted combination model to obtain a comprehensive prediction model, the weighted coefficients in the mean square error weighted combination model can be defined as quantum harmonies based on quantum coding and harmony memory library, and the quantum superposition state of the quantum harmonies can be determined according to the quantum state information coding technology; measure the quantum superposition state through the collapse theory to determine the corresponding ground state information, generate a perturbation adjustment frequency according to the ground state information to determine the perturbation bandwidth, and perform a micro-perturbation on the quantum harmonies to dynamically optimize the quantum harmonies; update the quantum coding and harmony memory library according to the optimized quantum harmonies, and select the quantum harmony that meets the weighted coefficient objective function from the quantum coding and harmony memory library after reaching the maximum number of iterations; use the quantum harmony that meets the requirements as the optimal weighted coefficient, output the optimal solution of the discount factor matrix according to the weighted coefficient, and update the mean square error weighted combination model with the optimal weighted coefficient and the optimal discount factor matrix to obtain a comprehensive prediction model.

[0160] Specifically, the calculation formula for the perturbation bandwidth is:

[0161]

[0162] Where S h ′ represents the disturbance bandwidth after disturbance adjustment, S h represents the undisturbed disturbance bandwidth, rand(0,1) represents a random number in the range of 0 to 1, g represents the golden ratio, d represents the disturbance adjustment frequency, K represents the micro disturbance amount, S up represents the upper bound of the quantum harmony, S down represents the lower bound of the quantum harmony.

[0163] Specifically, the calculation formula for the optimal weighting coefficient is:

[0164]

[0165] Where E o represents the optimal weighting coefficient of the o-th single prediction model, A represents the prediction period, c represents the total number of single prediction models, y a represents the true value of the initial working development trend of the wind turbine gearbox output for the a-th time, represents the discount factor of the o-th single prediction model for the a-th time. The discount factor represents the importance degree of the predicted value of the o single prediction models to the actual value of the current working development trend, and its range is from 0 to 1, represents the predicted value of the initial working development trend of the wind turbine gearbox output for the a-th time by the o-th single prediction model.

[0166] To facilitate the understanding of the above technical solutions of the present invention, the following will detail the operation method for predicting the failure period of the wind turbine gearbox in the actual process of the present invention.

[0167] Step 1: Obtain the gearbox structure information and establish a 3D model;

[0168] (1) Obtain the gearbox structure information:

[0169] Geometric shape and size: Obtain the geometric information of the wind turbine gearbox through software (such as SolidWorks). Assume the gear dimensions of the gearbox are as follows:

[0170] The outer diameter of the gear is 120 mm; the module of the gear is 6 mm; the gear transmission ratio is 3:1; the number of gears is 2 pairs of gears; the outer diameter of the bearing is 50 mm; the inner diameter of the bearing is 30 mm; the housing size is 200 mm * 150 mm * 100 mm;

[0171] Material properties:

[0172] The yield strength in the gear steel is 1200 MPa, the ultimate strength is 1500 MPa, and the elastic modulus is 200 GPa; the yield strength in the bearing steel is 1000 MPa, the ultimate strength is 1200 MPa, and the elastic modulus is 210 GPa.

[0173] (2) Detection result application:

[0174] Vibration monitoring: Use vibration sensors to monitor the vibration of the gearbox. Assume that slight wear is detected in the gear bearings and the vibration acceleration reaches 2.5 mm / s.

[0175] Temperature monitoring: Obtain the temperature data (30°C to 90°C) of the gearbox working area through temperature sensors.

[0176] Noise monitoring: Noise tests show that the noise of the gearbox increases under specific load conditions, indicating possible gear meshing problems.

[0177] (3) 3D model generation:

[0178] Use CAD software (such as SolidWorks) to generate a 3D gearbox model, and ensure that all details of the gears, bearings, housing, and couplings are included in the model. Use the gear module to model the gears, and set the meshing relationship between the gears through the gear transmission ratio (3:1). Determine the connection relationships between the gears and bearings, and between the gears and the housing to ensure the accuracy of the model.

[0179] Step 2: Generation of temperature and friction parameters;

[0180] (1) Temperature data collection:

[0181] Use infrared thermal imaging technology and temperature sensors to collect the temperature data of each component of the gearbox. Assume the temperatures under different load conditions are as follows: low load: 40°C; medium load: 60°C; high load: 90°C.

[0182] (2) Friction parameter estimation:

[0183] According to tribology principles and experimental data, estimate the friction parameters inside the gearbox. Assume the friction coefficient is 0.05 under low load and 0.08 under high load. At the same time, use the Berlin model to simulate the friction behavior between the gears and bearings, set the contact stress and lubrication conditions, and the friction coefficient changes during the working process.

[0184] (3) Influence of temperature and friction on the model:

[0185] In the 3D gearbox model, simulate the change of the temperature field through the heat conduction equation, set the temperature boundary conditions in the gear meshing area and bearing area (for example, the temperature change range in the gear meshing area is 40°C - 90°C), combine the temperature and friction parameters with the gearbox material properties, and calculate the influence of temperature on the gear material (such as geometric shape changes caused by expansion) and its influence on the meshing force.

[0186] Step 3: Dynamic coupling simulation and load history generation;

[0187] (1) Establish a fatigue life assessment model:

[0188] Through finite element analysis (FEA) and multibody dynamics (MBD), combined with parameters such as temperature, friction, and load, establish a fatigue life assessment model and set the dynamic simulation conditions for the gearbox:

[0189] During load variation, the load of the wind turbine fluctuates at different wind speeds (e.g., the load varies between 1000N and 3000N), during speed variation, the speed of the wind turbine changes from 1000 rpm to 2000 rpm, and during the simulation step, set the simulation step to 0.01 seconds and the simulation period to 60 seconds.

[0190] (2) Dynamic coupling simulation:

[0191] Use ANSYS or Abaqus as the simulation software for dynamic coupling simulation to simulate the dynamic response of the gearbox. Through simulation calculations, obtain the load history and thermal load changes of the wind turbine gearbox, and calculate the stress response of each component based on the gear meshing force.

[0192] (3) Load history and stress response:

[0193] The load history obtained from the simulation includes the load changes of the gearbox at different time points. Example data is as follows:

[0194] During load variation: 1500N (medium load) → 2500N (high load) → 1000N (low load);

[0195] During speed variation: 1200 rpm → 1800 rpm → 1500 rpm;

[0196] During the calculation of the equivalent stress, use von Mises stress to calculate the stress response history of the gearbox components. Assume that the equivalent stress of the gearbox varies under different load conditions, and the calculated stress response curve will reflect the stress states of components such as gears, bearings, and couplings at different time points.

[0197] Step 4: Failure cycle prediction and working development trend;

[0198] Based on the load history and equivalent stress response history, use a multi-layer perceptron (MLP) to predict the failure cycle of the gearbox. The input parameters are the load history, equivalent stress history, operating temperature of the gearbox, vibration data, etc., and the output is the predicted failure cycle of the gearbox. For example, the predicted failure cycle is 12 months.

[0199] The mean square error weighted combination model is used to perform weighted combination on the prediction results to obtain the optimal failure cycle prediction. The quantum harmony search algorithm is used to optimize the weighting coefficients to improve the prediction accuracy of the model. The weighting coefficients are encoded by qubits to ensure that the model can be optimized in each update. The weighting coefficients are slightly perturbed by adjusting the frequency of perturbation to optimize the model, and a comprehensive prediction model is obtained. Based on the prediction results, the failure cycle is judged, and it is determined that the failure cycle prediction of the wind turbine gearbox is 12 months.

[0200] Step Five: Generate operation adjustment strategies;

[0201] Based on the real-time operation data (temperature, load, speed, etc.) of the gearbox, the current working state of the gearbox is judged. Current temperature: 60°C (medium load); current load: 2000N (medium load);

[0202] Based on the failure cycle prediction (12 months) and the current working state, corresponding operation adjustment strategies are generated:

[0203] Reduce the load: Reduce the load by 10% to reduce fatigue damage, and adjust it to 1800N;

[0204] Reduce the speed: Reduce the speed by 5%, and adjust it to 1700 rpm;

[0205] Optimize the wind speed condition: When the wind speed is low, reduce the load of the wind turbine to extend the service life of the equipment.

[0206] Output the operation adjustment strategies of the wind turbine gearbox:

[0207] Reduce 10% in the adjusted load; reduce 5% in the adjusted speed. By adjusting the load and speed, it is expected to extend the failure cycle of the wind turbine by 3 months and reduce the maintenance cost.

[0208] In summary, with the above technical solutions of the present invention, the present invention can capture the temperature change of the gearbox in real time through infrared thermal imaging technology. Compared with traditional vibration monitoring, it is more sensitive to slight temperature rise anomalies of the gearbox and can detect abnormal trends at an early stage of the fault. At the same time, it combines the powerful feature selection ability of XGBoost and the time series modeling ability of GRU (Gated Recurrent Unit), making the prediction of the fault state more accurate, reducing false alarms and missed alarms. It can not only detect the current operation state, but also calculate the deviation from the benchmark operation state based on dynamic time rule technology, so as to issue a warning before the fault occurs, avoiding equipment shutdown or damage caused by sudden faults. At the same time, combined with the dynamic characteristic fatigue life assessment model, it can not only evaluate the current operation state, but also predict the future health status, making the operation adjustment of the wind turbine more scientific and reasonable and extending the service life.

[0209] The present invention divides the operating state through unsupervised clustering, can adapt to the variable operating environment of the wind turbine, and reduces the interference of the external environment on the detection of the fault signal of the wind turbine. When classifying the fault state, a sampling weight allocation strategy is adopted to solve the problem of sample imbalance and improve the accuracy of the classification model. By combining the classification model and the regression model, high-precision detection of the gearbox fault is realized. When performing online detection of the wind turbine gearbox fault, the fault detection result is generated through probability weighting, providing more detailed and accurate fault information, and thus being applicable to the fault detection of the wind turbine gearbox, capable of effectively identifying the abnormal behavior of the gearbox in different operating states and giving early warnings of potential faults. The present invention combines technologies such as three-dimensional model simulation, fatigue life assessment, load history analysis, and equivalent structural stress response, can not only achieve precise monitoring of the wind turbine, but also dynamically adjust its operating state, enabling the precise prediction of the fatigue life and failure cycle of the gearbox by combining the actual operating data and key parameters such as the temperature, friction, and load of the gearbox, effectively avoiding sudden fault shutdowns of the equipment, and thus reducing the economic losses caused by sudden faults.

[0210] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A fault detection method for a wind turbine gearbox based on infrared thermal imaging XGBoost-GRU, characterized in that, Including: Obtain the temperature value of the wind turbine gearbox at any moment based on infrared imaging technology, and establish an association database between temperature and operating parameters by combining the operating parameters of the wind turbine gearbox; Utilize the association database and dynamic time rule technology to calculate the distance between the operating state of the wind turbine gearbox and the reference operating state, and obtain the fault detection result of the wind turbine gearbox; Establish a fatigue life assessment model with dynamic characteristics, predict the failure cycle of the wind turbine gearbox under the premise of the fault inspection result, and generate an operation adjustment strategy according to the failure cycle to adjust the operation of the wind turbine gearbox.

2. The method for fault detection of a wind turbine gearbox based on infrared thermal imaging XGBoost-GRU according to claim 1, wherein The step of utilizing the association database and dynamic time rule technology to calculate the distance between the operating state of the wind turbine gearbox and the reference operating state, and obtain the fault detection result of the wind turbine gearbox includes: Select monitoring features from the association database according to the working principle and performance index standard of the wind turbine gearbox, and form an operating state data set by dividing the association database through a sliding window; Use the operating state data set as the dimension to represent the current operating characteristics of the wind turbine gearbox, and calculate the distance between the current operating characteristics and the reference operating state by using dynamic time rule technology; Randomly select several cluster centroids according to the time series data of the operating parameters of the wind turbine gearbox, and calculate the distance error from all arbitrary operating state data to their respective cluster centroids under any number of clusters; Perform unsupervised classification on the operating state of the wind turbine gearbox based on the distance error, and obtain the fault detection result of the wind turbine gearbox according to the classification result and machine learning technology.

3. The fault detection method for a wind turbine gearbox based on infrared thermal imaging XGBoost-GRU according to claim 2, wherein, The step of performing unsupervised classification on the operating state of the wind turbine gearbox based on the distance error, and obtaining the fault detection result of the wind turbine gearbox according to the classification result and machine learning technology includes: Plot the error distances corresponding to any number of clusters as a distance curve, observe the inflection point where the decline rate of the distance error within the distance curve meets the requirement, and take the distance number corresponding to the inflection point as the optimal number of clusters; Iteratively optimize the cluster assignment based on the optimal number of clusters, and allocate the operating state data set to the nearest cluster according to the distance between the current operating characteristics and the reference operating state, and recalculate the centroid of each cluster; Terminate the iteration when the cluster assignment does not change or reaches the maximum number of iterations, and form a consistency matrix according to the frequency of the operating state data set being assigned to the same cluster in all iterations; Quantify the stability of the consistency matrix by using the coefficient of dispersion, judge the rationality of the clustering result based on the stability result, and use the consistency matrix as the classification result of any operating state according to the judgment result; Combine the classification result, the associated data set and machine learning to analyze the probability of any operating state category of the wind turbine gearbox, and use a gated recurrent unit to obtain the fault detection result of the wind turbine gearbox.

4. A fault detection method for a wind turbine gearbox based on infrared thermal imaging XGBoost-GRU according to claim 3, characterized in that, The step of combining the classification result, the associated data set and machine learning to analyze the probability of any operating state category of the wind turbine gearbox, and using a gated recurrent unit to obtain the fault detection result of the wind turbine gearbox includes: Train a machine learning classification model based on a sampling weight allocation strategy, and input the classification results and the associated dataset into the machine learning classification model to output the probability that the sample belongs to each clustering category; Use a gated recurrent unit and the operating state categories of the wind turbine gearbox to perform probability verification and post-processing on the probability results, and calculate the fault confidence level of the wind turbine gearbox according to the processing results and the weighting parameters; Compare the fault confidence level with the fault threshold, and obtain the fault detection result of the wind turbine gearbox based on the comparison result.

5. A fault detection method for a wind turbine gearbox based on infrared thermal imaging XGBoost-GRU according to claim 1, characterized in that, The fatigue life assessment model with dynamic characteristics is established to predict the failure period of the wind turbine gearbox under the premise of the fault inspection result, and an operation adjustment strategy is generated according to the failure period to adjust the operation of the wind turbine gearbox, including: Generate a three-dimensional gearbox model using the fault detection result and the wind turbine gearbox structure information, and generate a set of numerical parameters including temperature and friction and apply them to the three-dimensional gearbox model; Determine the fatigue life assessment model according to the application result, and perform dynamic coupling simulation on the fatigue life assessment model to obtain the load history of the wind turbine gearbox; Add the coupling shaft of the wind turbine gearbox as a node attribute to the fatigue life assessment model, and perform transient analysis to obtain the equivalent structural stress response history of the wind turbine gearbox; Predict the working development trend of the wind turbine gearbox under the premise of the fault inspection result based on the load history and the equivalent structural stress response history, and judge the failure period of the wind turbine gearbox; Analyze the real-time working state of the wind turbine gearbox at any time period according to the failure period, and generate an operation adjustment strategy based on the analysis result to adjust the operation of the wind turbine gearbox.

6. The method for fault detection of a wind turbine gearbox based on infrared thermal imaging XGBoost-GRU according to claim 5, wherein, The fatigue life assessment model is determined according to the application result, and the dynamic coupling simulation is performed on the fatigue life assessment model to obtain the load history of the wind turbine gearbox, including: Determine the fatigue life assessment model based on the application result of the numerical parameter set and the three-dimensional gearbox model, and define the dynamic coupling simulation step size and the simulation result output frequency of the fatigue life assessment model; Perform dynamic coupling simulation of the wind turbine gearbox under any temperature and friction conditions according to the defined result and the numerical parameter set, and obtain the dynamic response curve of the wind turbine gearbox based on the simulation result; Obtain the ratio between the change of the thermal load and the meshing amplitude of the wind turbine gearbox according to the dynamic response curve, define the distribution evaluation index of the load space based on the ratio result, and generate the load distribution curve; Use entropy value to analyze the distribution discreteness of the distribution curve, and generate a line chart to obtain the dynamic change of the gear meshing force in the wind turbine gearbox as the load history of the wind turbine gearbox.

7. A fault detection method for a wind turbine gearbox based on infrared thermal imaging XGBoost-GRU according to claim 6, characterized in that, Predict the working development trend of the wind turbine gearbox under the premise of the fault inspection result based on the load history and the equivalent structural stress response history, and judge the failure period of the wind turbine gearbox, including: Use a multi-layer perceptron to screen out the influencing factors strongly related to the failure rate of the wind turbine gearbox from the load history and the equivalent structural stress response history, and input the influencing factors into the single-item prediction model to output the initial working development trend of the wind turbine gearbox; Generate a mean square error weighted combination model based on the weighted coefficients and discount factors of the single-item prediction model, integrate the initial work development trend, and output the secondary work development trend of the combination model; Use the quantum harmony search algorithm to dynamically optimize the weighted coefficients of the mean square error weighted combination model to obtain a comprehensive prediction model, and use the comprehensive prediction model to predict the work development trend of the wind turbine gearbox; Based on the predicted work development trend and the cyclic fatigue life criterion, judge the fatigue damage accumulation of the wind turbine gearbox, and obtain the failure cycle of the wind turbine gearbox according to the fatigue damage accumulation.

8. A fault detection method for a wind turbine gearbox based on infrared thermal imaging XGBoost-GRU according to claim 7, characterized in that, The dynamic optimization of the weighted coefficients of the mean square error weighted combination model by using the quantum harmony search algorithm to obtain a comprehensive prediction model includes: Based on quantum coding and the harmony memory library, define the weighted coefficients in the mean square error weighted combination model as quantum harmony, and determine the quantum superposition state of the quantum harmony according to the quantum state information coding technology; Measure the quantum superposition state through the collapse theory to determine the corresponding ground state information, generate a perturbation adjustment frequency according to the ground state information to determine the perturbation bandwidth, and perform a micro-perturbation on the quantum harmony to dynamically optimize the quantum harmony; Update the quantum coding and harmony memory library according to the optimized quantum harmony, and select the quantum harmony that meets the weighted coefficient objective function from the quantum coding and harmony memory library after reaching the maximum number of iterations; Use the quantum harmony that meets the requirements as the optimal weighted coefficient, output the optimal solution of the discount factor matrix according to the weighted coefficient, and update the mean square error weighted combination model with the optimal weighted coefficient and the optimal discount factor matrix to obtain a comprehensive prediction model.

9. A fault detection method for a wind turbine gearbox based on infrared thermal imaging XGBoost-GRU according to claim 8, characterized in that, The calculation formula for the perturbation bandwidth is: Where S h ′ represents the disturbance bandwidth after disturbance adjustment, S h represents the undisturbed disturbance bandwidth, rand(0,1) represents a random number in the range of 0 to 1, g represents the golden ratio, d represents the disturbance adjustment frequency, K represents the amount of micro disturbance, S up represents the upper bound of the quantum harmony, S down represents the lower bound of the quantum harmony.

10. A fault detection method for a wind turbine gearbox based on infrared thermal imaging XGBoost-GRU according to claim 9, characterized in that, The calculation formula for the optimal weighted coefficient is: Where, E o represents the optimal weighting coefficient of the o-th single prediction model, A represents the prediction period, c represents the total number of single prediction models, and y a represents the true value of the initial working development trend of the wind turbine gearbox output for the a-th time, represents the discount factor of the o-th single prediction model for the a-th time, represents the predicted value of the initial working development trend of the wind turbine gearbox output for the a-th time by the o-th single prediction model.

Images