Time frequency-mechanism fusion driving fan transmission chain part fault prediction method

Abstract

The invention discloses a fan transmission chain part fault prediction method based on time frequency-mechanism fusion driving, and the method comprises the steps: extracting time domain features and frequency domain features of a vibration signal of a fan transmission chain part, and analyzing power spectrum features through fast Fourier transform; the time domain features and the frequency domain features are screened, and a random forest algorithm is adopted to construct a transmission chain life prediction model based on the time-frequency features of the vibration signals; constructing a physical mechanism life prediction model in the damage process of different parts of the fan transmission chain; monitoring a vibration signal of a transmission chain of the wind driven generator in real time, extracting a current time-frequency feature and inputting the current time-frequency feature into the model for life prediction; and carrying out normalization processing on the time-frequency characteristic prediction result and the physical mechanism prediction result, carrying out fusion through a weighted average method to obtain a final residual life prediction value, and triggering graded fault early warning according to a threshold value. The method has the advantages of high sensitivity, strong anti-interference capability and the like, adapts to variable working condition dynamic correction, and improves the accuracy and reliability of fault prediction.

Description

Technical Field

[0001] The present invention relates to the field of wind turbine fault prediction, and in particular to a wind turbine transmission chain component fault prediction method driven by time-frequency-mechanism fusion. Background Art

[0002] In recent years, with the continuous depletion of non-renewable energy sources such as coal and oil, and the national goal of achieving carbon neutrality and peak carbon emissions, promoting the construction of a new power system based primarily on natural energy is crucial. Wind energy, as a widely distributed renewable energy source, has played a significant role in adjusting the energy mix, meeting electricity demand, and reducing carbon emissions. Wind turbines are essential equipment for converting wind energy into electrical energy and have become a key technology for achieving green and low-carbon development. However, the long-term operation of wind turbines in harsh environments significantly increases their probability of failure. Wind turbine failures not only increase wind farm operation and maintenance costs but can also lead to serious safety incidents. Therefore, predicting wind turbine failures is particularly important.

[0003] Drive train failure is a common fault in wind turbines. Wind turbine drive train failures primarily include blade, bearing, and gearbox failures. Blades, bearings, and other components experience fixed vibration patterns during damage, which are manifested through frequency-related characteristics. Furthermore, the damage process follows relevant physical laws, enabling the relationship between remaining service life and macroscopically observable parameters such as frequency to be established. Integrating the time-frequency domain characteristics of frequency with physical mechanisms can effectively predict wind turbine drive train failures.

[0004] Currently, various methods for predicting faults in wind turbine drive train components have been proposed both domestically and internationally. These methods are mainly based on physical theory and data-driven methods. The physical theory-based method uses existing physical principles to mathematically model the remaining lifespan; the data-driven method uses observable data from wind turbines for modeling and analysis. Both methods have significant drawbacks: the former, due to the complex electromechanical structure of wind turbines, makes it difficult to comprehensively consider factors affecting the lifespan of the generator, leading to inaccurate predictions; the latter, however, requires a large amount of real data to support the process. The shortcomings of the aforementioned methods result in low efficiency, poor timeliness, and insufficient ability to cope with complex faults in predicting faults in wind turbine drive train components. Summary of the Invention

[0005] In view of the defects in the prior art, the present invention proposes a method for predicting the faults of wind turbine transmission chain components driven by time-frequency-mechanism fusion, which includes the following steps: S1: Extract the time domain and frequency domain characteristics of the vibration signal of the fan transmission chain components, and analyze the power spectrum characteristics through fast Fourier transform; S2: Screen the time domain features and frequency domain features, and use the random forest algorithm to build a transmission chain life prediction model based on the time and frequency characteristics of the vibration signal; S3: Construct a life prediction model based on the physical mechanism of damage process of different components in the wind turbine transmission chain; S4: Real-time monitoring of the vibration signal of the wind turbine transmission chain, extraction of the current time-frequency characteristics and input into the model of step S2 and step S3 for life prediction; S5: Normalize the time-frequency feature prediction results and the physical mechanism prediction results, fuse them through the weighted average method to obtain the final remaining life prediction value, and trigger a graded fault warning based on the threshold.

[0006] Preferably, the time domain characteristics and frequency domain characteristics of the vibration signals of the wind turbine transmission chain components include range, root mean square value, kurtosis, and skewness.

[0007] Preferably, the physical mechanism life prediction model of the damage process of different components of the wind turbine transmission chain includes a modified Pairs law model for blade cracks, a fatigue relief model for bearing spalling, and a bending fatigue model for gear tooth breakage.

[0008] Preferably, the S1 specifically includes: S1.1 Extract the maximum and minimum values ​​in the time characteristic sequence of the vibration signal, which are respectively denoted as , and get the number of peak occurrences , and on this basis calculate the range of the time series ,Right now ; S1.2 Calculate the RMS value of the time series to characterize the energy of the vibration signal, that is, the vibration intensity. The RMS value is recorded as: ; Where N is the number of sampling points in the time series, is the sampling value of the discrete vibration signal; When a wind turbine is in normal operation, the RMS value tends to be stable. Wear or loosening of components will cause the RMS to gradually increase. S1.3 Determine the kurtosis K of the vibration signal, denoted as ; in, is the mean value of the vibration signal; when the fan is in normal operation, K is about 3. When an early fault occurs, K will increase significantly due to the impact signal; S1.4 determines the skewness S of the vibration signal. Under normal operating conditions, the skewness is 0. When a fan fails, asymmetric shocks can cause the skewness to change. ; S1.5 uses Fast Fourier Transform (FFT) to transform the vibration signal in the time domain into the frequency domain. is the data after FFT transformation, is the signal to be transformed, is the weight function of discrete Fourier transform, then ; The theoretical fault frequency of wind turbine components when they fail is located, and the amplitude corresponding to the frequency is extracted as the frequency domain feature.

[0009] Preferably, said S2 specifically includes: S2.1 Feature data preprocessing: First, perform data cleaning to process outliers and missing values ​​in the data set; for missing data, use the mean to fill; then perform data standardization; suppose the original data set is , the standardized data set is ,but ; in, Dataset The mean of It is a dataset The standard deviation of ; ; S2.2 Feature selection: Select the optimal feature set from the original feature set; The information gain rate (IGR) is used to evaluate the contribution of features to classification tasks. By quantifying the information correlation between features and category labels, the optimal feature set is selected and the problem of information gain's preference for multi-valued features is solved.

[0010] in, The amount of information brought to feature A is increased; To measure the distribution complexity of the value of feature A itself, ; ; in, is the number of discrete values ​​of feature A, The value of feature A is A subset of For the dataset The entropy of is used to measure the category uncertainty; is the conditional entropy of the dataset given feature A,

[0011] ; in, yes Medium Category the proportion of After calculating the information gain rate of all time-domain and frequency-domain features, sort them from high to low according to IGR, and take the top 20 features for subsequent model training; S2.3 Training and testing of the fan transmission chain fault prediction model based on the random forest algorithm; First, the pre-processed vibration signal is divided into a training set and a test plan with a ratio of 7:3; The random forest model realizes prediction by integrating multiple regression decision trees. The CART algorithm is used when constructing a single tree, and the node splitting criterion is to minimize the mean square error (MSE). Specifically, ; in, is the sample label mean. Each tree is randomly selected features (m is the total number of features) for split optimization and enhance model diversity through bootstrap sampling; Finally, the final predicted lifespan value is output through the multi-tree prediction mean ; .

[0012] Preferably, said S3 specifically includes: S3.1 For the short-term life prediction of blade cracks, a physical mechanism model for short-term life prediction of fan blade cracks based on the modified Pairs law is used; first, the critical length of the blade crack for the fan to maintain normal operation is determined. ; ; in, is the fracture toughness of the fan blade, here we take Take 1.15 to consider the blade curvature correction; Take the maximum stress in the current time window; Assume the current crack length is This length is measured by the sensor of the fan blade, and the remaining life It can be expressed as ; Now solve the current crack length The expansion increment under , corresponding to each stress block , extended increment for: ; Among them, C is , m is 3.3, is the stress intensity factor, and its expression is as follows: ; ; Among them, E is 40GPa, Take 0.3; S3.2 The contact fatigue model is used to model and analyze the spalling mechanism of the fan bearing. First, the contact stress is calculated based on the Hertz theory. Perform the calculation: ; Where Q is the bearing load and R is the equivalent radius of curvature of the rolling element and the raceway. Both can be found in the bearing manual. Taking 210GPa, the crack growth life can be calculated according to Pairs' law ,Right now ; in, is the critical crack length, which is 0.5 mm; n is the rotation speed; C is the Pairs constant, which is ; m is 3.5; is the subsurface shear stress, and its expression is ; S3.3 uses the bending fatigue model to predict the short-term life of the wind turbine transmission chain gear. In order for the gear to operate normally, it is necessary to ensure that the tooth root bending stress is less than the fatigue limit of the material, that is, ; The tooth root bending stress Perform calculations: ; Among them, b, m, and Y are the gear module, tooth width, and tooth form factor respectively. These three physical quantities can be obtained by looking up the gear design manual; is the application coefficient, which is taken as 1.5 here; is the stress concentration factor, which is taken as 1.8 here; is the tangential force on the tooth surface, and its calculation relationship is: ; Where T is the torque and d is the gear pitch circle radius; Under the premise of satisfying formula (15), the number of cycles predicted by the Basquin equation is solved. ; in, is the fatigue strength coefficient of the Basquin equation, which is taken as 1000 MPa here; b is the Basquin equation exponent, which is taken as -0.1 here; Finally calculate the remaining life , ; Where n is the gear speed.

[0013] Preferably, the S4 specifically includes real-time monitoring of the vibration signal of the wind turbine transmission chain, placing sensors at different components to capture the vibration signal of the wind turbine transmission chain, transmitting the vibration signal collected by the sensor back to the data acquisition system and performing feature extraction.

[0014] Preferably, said S5 specifically includes: S5.1 In step S2, the predicted value of the fan transmission chain life based on the time-frequency domain characteristics of the vibration signal is obtained In step S3, the remaining life prediction value of each part based on the physical mechanism modeling of the damage process is obtained. , because the dimensions of the two predicted values ​​are different, they need to be normalized first. ; in, is the historical maximum lifespan of similar parts, and then weighted average fusion is performed. ; 、 are the weights of the two lifespans, which are dynamically adjusted according to the prediction errors of the two models in historical data. The calculation method is: ; Among them, MSE1 and MSE2 are the mean square errors of T1 and T2 on the validation set respectively; Finally, the final prediction value is output through denormalization , ; S5.2 Fault warning and decision-making, set the remaining life threshold and make adjustments according to actual conditions; When the value falls below the threshold, a graded response is triggered: a yellow alert initiates status re-inspection and spare parts preparation, while a red alert results in an immediate shutdown for inspection. This combines the time-frequency characteristic sensitivity of the data-driven model with the damage evolution law of the physical model.

[0015] In a second aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0016] In a third aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the above method when executed by a processor.

[0017] Beneficial effects of the present invention: The time-frequency-mechanism fusion-driven wind turbine transmission chain component fault prediction method disclosed in the present invention comprehensively considers the time-frequency domain characteristics of the wind turbine transmission chain vibration signal and the physical mechanism of mechanical damage of the components to predict the faults of the wind turbine transmission chain components. Compared with other methods, it has the advantages of high sensitivity and strong anti-interference ability, adapts to dynamic correction of changing working conditions, and improves the accuracy and reliability of fault prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The present invention has the following accompanying drawings: Figure 1 This is a flow chart of the wind turbine transmission chain component fault prediction method driven by time-frequency-mechanism fusion according to the present invention. DETAILED DESCRIPTION

[0019] To make the purpose, advantages and features of the present invention more apparent, the following Figure 1 The present invention is further described in detail with reference to the following specific embodiments.

[0020] Step A: Extract the statistical characteristics and frequency domain characteristics of the vibration signal at different fault occurrence stages.

[0021] A1. Extract the maximum and minimum values ​​in the time feature sequence of the vibration signal, denoted as And get the number of peak occurrences , and on this basis calculate the range of the time series ,Right now ; A2. Calculate the RMS value of the time series to represent the energy of the vibration signal, that is, the vibration intensity. The RMS value is expressed as: ; Where N is the number of sampling points in the time series, is the sampling value of the discrete vibration signal.

[0022] When a wind turbine is operating normally, the RMS value tends to be stable. Wear or looseness of components will cause the RMS value to gradually increase.

[0023] A3. Determine the kurtosis K of the vibration signal, denoted as ; in, is the mean value of the vibration signal. Under normal operating conditions, K is approximately 3. When an early fault occurs, K increases significantly due to the impact signal.

[0024] A4. Determine the skewness S of the vibration signal. Under normal operating conditions, the skewness is 0. When a fan fails, asymmetric shocks can cause the skewness to change.

[0025] ; A5. Use Fast Fourier Transform (FFT) to transform the vibration signal in the time domain into the frequency domain. is the data after FFT transformation, is the signal to be transformed, is the weight function of discrete Fourier transform, then ; The theoretical fault frequency of wind turbine components when they fail is located, and the amplitude corresponding to the frequency is extracted as the frequency domain feature.

[0026] Step B: Screen the time domain and frequency domain features of the vibration signals of the fan transmission chain components obtained in the previous step, and then use the random forest algorithm to predict the fan transmission chain fault.

[0027] B1. Feature data preprocessing. First, perform data cleaning to deal with outliers, missing values, etc. in the data set. For missing data, you can use the mean to fill it. Next, perform data standardization. Suppose the original data set is , the standardized data set is .but ; in, Dataset The mean of It is a dataset The standard deviation of ; ; B2. Feature Selection. The resulting feature dataset may contain irrelevant features, which can lead to structural redundancy in the fault prediction model and slow convergence. To optimize the prediction model's performance, feature selection must be performed based on information gain, selecting the optimal feature set from the original feature set.

[0028] The information gain rate (IGR) is used to evaluate the contribution of features to classification tasks. By quantifying the information correlation between features and category labels, the optimal feature set is selected and the problem of information gain's preference for multi-valued features is solved.

[0029] ; in, The amount of information brought to feature A is increased; It is used to measure the distribution complexity of the value of feature A itself.

[0030] ; ; in, is the number of discrete values ​​of feature A, The value of feature A is A subset of For the dataset The entropy of is used to measure the category uncertainty; is the conditional entropy of the dataset given feature A.

[0031] ; ; in, yes Medium category % of the total.

[0032] After calculating the information gain rate of all time-domain and frequency-domain features, they are sorted from high to low according to IGR, and the top 20 features are used for subsequent model training.

[0033] B3. Training and testing of a fan drive train fault prediction model based on the random forest algorithm. First, the preprocessed vibration signal is divided into a training set and a test set with a ratio of 7:3 to ensure data independence and generalization verification. The random forest model achieves prediction by integrating multiple regression decision trees. The CART algorithm is used to construct a single tree, with the minimum mean square error (MSE) as the node splitting criterion. Specifically, ; in, is the sample label mean. Each tree is randomly selected The features (m is the total number of features) are split and optimized, and the model diversity is enhanced by the bootstrap sampling method.

[0034] Finally, the final predicted lifespan value is output through the multi-tree prediction mean .

[0035] ; Step C. Construct a mathematical model of the damage process of different components in the wind turbine drive chain. Wind turbine drive chain failures mainly include blade cracks, bearing spalling, and gear tooth breakage.

[0036] C1. For the short-term life prediction of blade cracks, a physical mechanism model for short-term life prediction of fan blade cracks based on the modified Pairs law is used. First, the critical length of the blade crack for the fan to maintain normal operation is determined. .

[0037] ; in, is the fracture toughness of the fan blade, here we take Take 1.15 to consider the blade curvature correction; Take the maximum stress in the current time window.

[0038] Assume the current crack length is , the length is measured by the sensor of the fan blade, and the remaining life It can be expressed as ; Now solve the current crack length The expansion increment under . For each stress block , extended increment for: ; Among them, C is , m is 3.3, is the stress intensity factor, and its expression is as follows.

[0039] ; ; Among them, E is 40GPa, Take 0.3.

[0040] C2. Bearing spalling is a common fault in the fan transmission chain. The contact fatigue model is used to model and analyze the mechanism of fan bearing spalling. First, the contact stress is calculated based on the Hertz theory. Perform the calculation: ; in, is the bearing load, is the equivalent curvature radius of the rolling element and the raceway, both of which can be found in the bearing manual; Take 210GPa. According to Pairs' law, the crack growth life can be calculated ,Right now ; in, is the critical crack length, which is 0.5 mm; n is the rotation speed; C is the Pairs constant, which is ; m is 3.5; is the subsurface shear stress, which is expressed as

[0041] C3. Use bending fatigue model to predict the short-term life of wind turbine transmission chain gear. In order for the gear to operate normally, it is necessary to ensure that the tooth root bending stress is less than the fatigue limit of the material, that is, ; The tooth root bending stress Perform calculations.

[0042] ; Among them, b, m, and Y are the gear module, tooth width, and tooth form factor respectively. These three physical quantities can be obtained by looking up the gear design manual; is the application coefficient, which is taken as 1.5 here; is the stress concentration factor, which is taken as 1.8 here; is the tangential force on the tooth surface, and its calculation relationship is: ; Where T is the torque and d is the gear pitch circle radius.

[0043] Under the premise of satisfying Equation 15, the Basquin equation is solved to predict the number of cycles.

[0044] ; in, is the fatigue strength coefficient of the Basquin equation, which is taken as 1000 MPa here; b is the Basquin equation exponent, which is taken as -0.1 here.

[0045] Finally calculate the remaining life .

[0046] ; Where n is the gear speed.

[0047] Step D. Real-time monitoring of the vibration signal of the wind turbine transmission chain. To capture the vibration signal of the wind turbine transmission chain, sensors are placed at different components. The vibration signals collected by the sensors are transmitted back to the data acquisition system for feature extraction.

[0048] Step E: Use the weighted average method to obtain the remaining life prediction of the wind turbine transmission chain components, so as to perform fault prediction.

[0049] E1. The predicted value of the fan transmission chain life based on the time-frequency domain characteristics of the vibration signal is obtained in step B. In step C, the remaining life prediction value of each part based on the physical mechanism modeling of the damage process is obtained. Since the dimensions of the two predicted values ​​may be different, they need to be normalized first.

[0050] ; in, is the historical maximum lifespan of similar parts. Next, weighted average fusion is performed.

[0051] ; 、 are the weights of the two lifespans, which are dynamically adjusted based on the prediction errors of the two models in historical data. The calculation method is: ; Among them, MSE1 and MSE2 are the mean square errors of T1 and T2 on the validation set respectively. Finally, the final prediction value is output through denormalization .

[0052] ; E2. Fault warning and decision-making. Set the remaining life threshold, such as 7 days for yellow warning and 3 days for red warning, and adjust it according to the actual situation. When the value falls below the threshold, a graded response is triggered: a yellow alert initiates status re-inspection and spare parts preparation, while a red alert results in an immediate shutdown for inspection. By combining the time-frequency sensitivity of the data-driven model with the damage evolution patterns of the physical model, this effectively balances the risks of false alarms and missed alarms, providing a decision-making basis for safe and precise wind turbine operation and maintenance.

[0053] The above embodiments are only used to illustrate the present invention, and are not intended to limit the present invention. Ordinary technicians in the relevant technical field may make various changes and modifications without departing from the essence and scope of the present invention. Therefore, all equivalent technical solutions also fall within the scope of the present invention. The scope of patent protection of the present invention should be defined by the claims.

[0054] The contents not described in detail in this specification belong to the prior art known to those skilled in the art.

Claims

1. A wind turbine transmission chain component fault prediction method driven by time-frequency-mechanism fusion, characterized by: The steps include: S1: Extract the time domain and frequency domain characteristics of the vibration signal of the fan transmission chain components, and analyze the power spectrum characteristics through fast Fourier transform; S2: Screen the time domain features and frequency domain features, and use the random forest algorithm to build a transmission chain life prediction model based on the time and frequency characteristics of the vibration signal; S3: Construct a life prediction model based on the physical mechanism of damage process of different components in the wind turbine transmission chain; S4: Real-time monitoring of the vibration signal of the wind turbine transmission chain, extraction of the current time-frequency characteristics and input into the model of step S2 and step S3 for life prediction; S5: Normalize the time-frequency feature prediction results and the physical mechanism prediction results, fuse them through the weighted average method to obtain the final remaining life prediction value, and trigger a graded fault warning based on the threshold.

2. The wind turbine transmission chain component fault prediction method driven by time-frequency-mechanism fusion as claimed in claim 1 is characterized in that: The time domain characteristics and frequency domain characteristics of the vibration signals of the wind turbine transmission chain components include range, root mean square value, kurtosis, and skewness.

3. The wind turbine transmission chain component fault prediction method driven by time-frequency-mechanism fusion as claimed in claim 1 is characterized in that: The physical mechanism life prediction model of the damage process of different components of the wind turbine transmission chain includes a modified Pairs law model for blade cracks, a fatigue relief model for bearing spalling, and a bending fatigue model for gear tooth breakage.

4. The wind turbine transmission chain component fault prediction method driven by time-frequency-mechanism fusion as claimed in claim 1 is characterized in that: Said S1 specifically includes: S1.1 Extract the maximum and minimum values ​​in the time characteristic sequence of the vibration signal, which are respectively denoted as , And get the number of peak occurrences And on this basis, calculate the range of the time series ,Right now ； S1.2 Calculate the RMS value of the time series to characterize the energy of the vibration signal, that is, the vibration intensity. The RMS value is recorded as: ； Where N is the number of sampling points in the time series, is the sampling value of the discrete vibration signal; when the wind turbine is in normal operation, the RMS value tends to be stable, and the wear or looseness of the components will cause the RMS value to gradually increase; S1.3 Determine the kurtosis K of the vibration signal, denoted as ； in, is the mean value of the vibration signal; when the fan is in normal operation, K is 3. When an early fault occurs, K will increase significantly due to the impact signal; S1.4 determines the skewness S of the vibration signal. Under normal operating conditions, the skewness is 0. When a fan fails, asymmetric shocks can cause the skewness to change. ； S1.5 uses fast Fourier transform to transform the vibration signal in the time domain into the frequency domain. is the data after FFT transformation, is the signal to be transformed, is the weight function of discrete Fourier transform, then ； The theoretical fault frequency of wind turbine components when they fail is located, and the amplitude corresponding to the frequency is extracted as the frequency domain feature.

5. The wind turbine transmission chain component fault prediction method driven by time-frequency-mechanism fusion as claimed in claim 4 is characterized in that: Said S2 specifically includes: S2.1 Feature data preprocessing: First, perform data cleaning to process outliers and missing values ​​in the data set; for missing data, use the mean to fill; then perform data standardization; suppose the original data set is , the standardized data set is , but ； in, Dataset The mean of It is a dataset The standard deviation of ； ； S2.2 Feature selection: Select the optimal feature set from the original feature set; The information gain rate (IGR) is used to evaluate the contribution of features to classification tasks. By quantifying the information correlation between features and category labels, the optimal feature set is selected and the problem of information gain's preference for multi-valued features is solved. ； in, The amount of information brought to feature A is increased; To measure the distribution complexity of the value of feature A itself, ； ； in, is the number of discrete values ​​of feature A, The value of feature A is A subset of For the dataset The entropy of is used to measure the category uncertainty; is the conditional entropy of the dataset given feature A, ； ； in, yes Medium category the proportion of After calculating the information gain rate of all time-domain and frequency-domain features, sort them from high to low according to IGR, and take the top 20 features for subsequent model training; S2.3 Training and testing of the fan transmission chain fault prediction model based on the random forest algorithm; First, the pre-processed vibration signal is divided into a training set and a test plan with a ratio of 7:3; the random forest model realizes prediction by integrating multiple regression decision trees, and the CART algorithm is used when constructing a single tree, with the minimum mean square error as the node splitting criterion, specifically: ； in, is the sample label mean; each tree is randomly selected features, m is the total features The number of features is optimized by splitting, and the model diversity is enhanced by bootstrapping method; Finally, the final predicted lifespan value is output through the multi-tree prediction mean ; 。 6. The wind turbine transmission chain component fault prediction method driven by time-frequency-mechanism fusion as claimed in claim 5 is characterized in that: Said S3 specifically includes: S3.1 For the short-term life prediction of blade cracks, a physical mechanism model for short-term life prediction of fan blade cracks based on the modified Pairs law is used; first, the critical length of the blade crack for the fan to maintain normal operation is determined. , ； in, is the fracture toughness of the fan blade, here we take Take 1.15 to consider the blade curvature correction; Take the maximum stress in the current time window; Assume the current crack length is , the length is measured by the sensor of the fan blade, then the remaining life Expressed as ； Now solve the current crack length The expansion increment under , corresponding to each stress block , extended increment for: ； Among them, C is , m is 3.3, is the stress intensity factor, and its expression is as follows: ； ； Among them, E is 40GPa, Take 0.3; S3.2 The contact fatigue model is used to model and analyze the spalling mechanism of the fan bearing. First, the contact stress is calculated based on the Hertz theory. Perform the calculation: ； Where Q is the bearing load, R is the equivalent curvature radius of the rolling element and the track; Take 210GPa and calculate the crack growth life according to Pairs law ,Right now ； in, is the critical crack length, which is 0.5 mm; n is the rotation speed; C is the Pairs constant, which is ; m is 3.5; is the subsurface shear stress, and its expression is ； S3.3 uses the bending fatigue model to predict the short-term life of the wind turbine transmission chain gear. In order for the gear to operate normally, it is necessary to ensure that the tooth root bending stress is less than the fatigue limit of the material, that is, ； The tooth root bending stress Perform calculations: ； Among them, b, m, and Y are the gear module, tooth width, and tooth form coefficient respectively; is the application coefficient, which is taken as 1.5 here; is the stress concentration factor, which is taken as 1.8 here; is the tangential force on the tooth surface, and its calculation relationship is: ； Where T is the torque and d is the gear pitch circle radius; Under the premise of satisfying formula (15), the number of cycles predicted by the Basquin equation is solved. ； in, is the fatigue strength coefficient of the Basquin equation, which is taken as 1000 MPa here; b is the Basquin equation exponent, which is taken as -0.1 here; Finally calculate the remaining life , ； Where n is the gear speed.

7. The wind turbine transmission chain component fault prediction method driven by time-frequency-mechanism fusion as claimed in claim 6 is characterized in that: The S4 specifically includes real-time monitoring of the vibration signal of the wind turbine transmission chain, placing sensors at different components to capture the vibration signal of the wind turbine transmission chain, transmitting the vibration signal collected by the sensor back to the data acquisition system and performing feature extraction.

8. The wind turbine transmission chain component fault prediction method driven by time-frequency-mechanism fusion as claimed in claim 7 is characterized in that: Said S5 specifically includes: S5.1 In step S2, the predicted value of the fan transmission chain life based on the time-frequency domain characteristics of the vibration signal is obtained In step S3, the remaining life prediction value of each part based on the physical mechanism modeling of the damage process is obtained. , because the dimensions of the two predicted values ​​are different, they need to be normalized first. ； in, is the historical maximum lifespan of similar parts, and then weighted average fusion is performed. ; 、 are the weights of the two lifespans, which are dynamically adjusted according to the prediction errors of the two models in historical data. The calculation method is: ； Among them, MSE1 and MSE2 are the mean square errors of T1 and T2 on the validation set respectively; Finally, the final prediction value is output by denormalization , ； S5.2 Fault warning and decision-making, set the remaining life threshold and make adjustments according to actual conditions; When the value falls below the threshold, a graded response is triggered: a yellow alert initiates status re-inspection and spare parts preparation, while a red alert results in an immediate shutdown for inspection. This combines the time-frequency characteristic sensitivity of the data-driven model with the damage evolution law of the physical model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images