Fan gear box high speed shaft bearing fault diagnosis method and system based on VMD

Abstract

This invention belongs to the field of fault diagnosis technology and discloses a method and system for fault diagnosis of high-speed shaft bearings in wind turbine gearboxes based on Virtual Machine Diagnosis (VMD). The method includes the following steps: acquiring vibration signals from the high-speed shaft bearings of the wind turbine gearbox and preprocessing the vibration signals to obtain preprocessed vibration signals; using a multi-objective optimization VMD algorithm to decompose the preprocessed vibration signals to obtain several IMF components; filtering the IMF components to obtain the optimal IMF component and extracting the envelope spectrum of the optimal IMF component; using a fault diagnosis model constructed based on a deep learning algorithm to perform fault diagnosis on the envelope spectrum of the optimal IMF component to obtain the fault diagnosis result. This invention solves the problems of low accuracy and high labor costs in existing technologies.

Description

Technical Field

[0001] This invention belongs to the field of fault diagnosis technology, specifically relating to a method and system for fault diagnosis of high-speed shaft bearings in wind turbine gearboxes based on VMD. Background Technology

[0002] Energy is of paramount importance to the progress of human civilization, with fossil fuels making a significant contribution. However, with the depletion of fossil fuels and pollution problems, the energy trend is shifting from fossil fuels to pollution-free renewable energy. Wind energy is an important renewable and clean energy source, and wind power generation technology is relatively mature. my country is a major wind power country with abundant wind resources and its installed capacity continues to rise. However, this has also led to frequent wind turbine failures. Among the key components of wind turbines, gearbox failures account for a high proportion, and within gearboxes, bearing problems are the most prevalent.

[0003] To better diagnose bearing faults, researchers have proposed numerous methods, most of which are based on the analysis of detection signals such as vibration, sound, oil, and temperature. Among these, using vibration signals as the detection signal for bearing fault diagnosis is a relatively mature and reliable method, capable of collecting and detecting fault information caused by external factors. Common methods for vibration signal analysis include time-domain analysis, frequency-domain analysis, time-frequency-domain analysis, and mode decomposition. Variational Mode Decomposition (VMD) is an adaptive signal processing method that can decompose complex signals into a series of eigenmode functions with a finite bandwidth and a center frequency. VMD performs exceptionally well in processing non-stationary and nonlinear signals, capable of detecting potential anomalies and reducing losses, and has been successfully applied in the field of fault diagnosis.

[0004] Existing technologies still have many shortcomings, including:

[0005] 1) Low accuracy: The performance of the VMD algorithm heavily depends on two core parameters. In practical applications, these two core parameters usually need to be determined based on experience or repeated experiments, which is not only time-consuming and labor-intensive, but also highly subjective and difficult to guarantee the optimality of the decomposition effect. This may result in fault features being decomposed into different IMF components or being contaminated by noise components, thereby affecting the accuracy of subsequent diagnosis.

[0006] 2) High labor costs: Existing technologies often rely on manual analysis of the IMF components of vibration signals obtained from VMD decomposition to identify bearing faults. This method requires a large investment of manpower and is inefficient. Summary of the Invention

[0007] To address the issues of low accuracy and high labor costs in existing technologies, the present invention aims to provide a method and system for diagnosing faults in high-speed shaft bearings of wind turbine gearboxes based on VMD.

[0008] The technical solution adopted in this invention is as follows:

[0009] A method for fault diagnosis of high-speed shaft bearings in wind turbine gearboxes based on VMD includes the following steps:

[0010] Vibration signals of the high-speed shaft bearing of the wind turbine gearbox are collected and preprocessed to obtain preprocessed vibration signals.

[0011] The preprocessed vibration signal is decomposed using the VMD algorithm with multi-objective optimization to obtain several IMF components.

[0012] Several IMF components are screened to obtain the optimal IMF component, and the envelope spectrum of the optimal IMF component is extracted.

[0013] A fault diagnosis model based on deep learning algorithms is used to diagnose faults in the envelope spectrum of the optimal IMF component, and the fault diagnosis results are obtained.

[0014] Furthermore, the VMD algorithm is improved using the NSGA-II algorithm to obtain a multi-objective optimized VMD algorithm.

[0015] Furthermore, the VMD algorithm is improved using the NSGA-II algorithm to obtain a multi-objective optimization VMD algorithm, which includes the following steps:

[0016] Determine the core parameters and their ranges for the VMD algorithm, and use the core parameters as independent variables to be optimized;

[0017] Set the objective function of the NSGA-II algorithm, and use the NSGA-II algorithm to iteratively optimize based on the objective function and independent variables to obtain the optimal solution;

[0018] The optimal solution is decoded to obtain the optimal core parameter values ​​of the VMD algorithm. Based on the optimal core parameter values, the VMD algorithm is improved to obtain a multi-objective optimization VMD algorithm.

[0019] Furthermore, the core parameters of the VMD algorithm include the number of modes and the penalty factor;

[0020] The objective functions include the permutation entropy function and the envelope entropy function.

[0021] Furthermore, the objective function of the NSGA-II algorithm is set, and based on the objective function and independent variables, the NSGA-II algorithm is used to iteratively optimize and obtain the optimal solution, including the following steps:

[0022] Set the objective function and algorithm parameters of the NSGA-II algorithm, and encode the independent variables as the solution vectors of individuals in the NSGA-II algorithm;

[0023] Based on the algorithm parameters, the NSGA-II population is initialized to obtain an initial NSGA-II population including several candidate solutions.

[0024] Based on the objective function, obtain the objective function value corresponding to each candidate solution in the initial NSGA-II population;

[0025] Based on the objective function value and the Pareto front sorting and crowding selection mechanism, the NSGA-II algorithm is used to perform iterative evolution to obtain a new NSGA-II population;

[0026] If the number of iterations reaches the maximum number of iterations, then output the Pareto front solution set in the final NSGA-II population based on the objective function value;

[0027] Using the curvature method, the optimal compromise solution is selected from the Pareto front solution set, and this optimal compromise solution is taken as the optimal solution.

[0028] Furthermore, based on the objective function, the objective function value corresponding to each candidate solution in the initial NSGA-II population is obtained, including the following steps:

[0029] Decode any candidate solution in the initial NSGA-II population to obtain candidate core parameters, and improve the VMD algorithm based on the candidate core parameters to obtain a candidate VMD algorithm;

[0030] Based on the candidate VMD algorithm, the sample vibration signal is decomposed to obtain several sample IMF components;

[0031] Based on the permutation entropy function in the objective function, calculate the sample permutation entropy of several sample IMF components and output the corresponding average sample permutation entropy.

[0032] Based on the envelope entropy function in the objective function, calculate the sample envelope entropy of several sample IMF components and output the corresponding average sample envelope entropy.

[0033] By combining the average entropy of the sample permutation and the average entropy of the sample envelope, the objective function value corresponding to the current candidate solution is obtained;

[0034] By traversing all candidate solutions in the initial NSGA-II population, several corresponding objective function values ​​are obtained.

[0035] Furthermore, several IMF components are screened to obtain the optimal IMF component, and the envelope spectrum of the optimal IMF component is extracted, including the following steps:

[0036] Calculate the kurtosis value of each IMF component and select the IMF component with the largest kurtosis value as the optimal IMF component;

[0037] Perform a Hilbert transform on the optimal IMF component to obtain the corresponding analytical signal;

[0038] The analytic signal is moduloed to obtain the corresponding envelope signal, and the envelope signal is subjected to spectral analysis to plot the corresponding envelope spectrum.

[0039] Furthermore, the fault diagnosis model is constructed based on the CNN-LSTM-Attention-MLP algorithm, and the fault diagnosis model includes a spatial feature extraction module based on the CNN algorithm, a temporal feature extraction module based on the LSTM algorithm, a weighted fusion module based on the Attention mechanism, and a fault diagnosis module based on the MLP algorithm, which are connected in sequence.

[0040] Furthermore, a fault diagnosis model based on deep learning algorithms is used to diagnose faults in the envelope spectrum of the optimal IMF component, obtaining the fault diagnosis results, including the following steps:

[0041] The envelope spectrum of the optimal IMF component is input into the fault diagnosis model constructed based on a deep learning algorithm;

[0042] The spatial feature extraction module of the fault diagnosis model is used to extract the spatial features of the envelope spectrum.

[0043] The time feature extraction module of the fault diagnosis model is used to extract the time features of the fault.

[0044] Based on the dynamic attention weights, the weighted fusion module of the fault diagnosis model is used to perform weighted fusion of spatial and temporal features to obtain fused features.

[0045] Based on the fusion characteristics, the fault diagnosis module of the fault diagnosis model is used to perform fault diagnosis and obtain the fault diagnosis results.

[0046] A VMD-based high-speed shaft bearing fault diagnosis system for wind turbine gearboxes is provided to implement a fault diagnosis method for high-speed shaft bearings in wind turbine gearboxes. The system includes a vibration signal acquisition unit, a vibration signal decomposition unit, an IMF component screening unit, and a fault diagnosis unit connected in sequence.

[0047] The beneficial effects of this invention are as follows:

[0048] This invention provides a method and system for fault diagnosis of high-speed shaft bearings in wind turbine gearboxes based on Virtual Machine Decomposition (VMD). The multi-objective optimization algorithm employed can quickly and accurately determine the core parameters of the VMD algorithm, overcoming the subjectivity and local optima problems of traditional parameter selection, reducing manual intervention, and significantly improving the scientific rigor and effectiveness of parameter determination. The resulting multi-objective optimized VMD algorithm takes into account both the complexity of the decomposed components and the energy concentration characteristics, achieving a more comprehensive and automated evaluation of the decomposition effect and improving the accuracy of fault diagnosis. The fault diagnosis model built based on deep learning algorithms enables automatic fault diagnosis, reducing labor costs and improving the efficiency of fault diagnosis.

[0049] Other beneficial effects of the present invention will be further explained in the specific embodiments. Attached Figure Description

[0050] Figure 1 This is a flowchart of the VMD-based high-speed shaft bearing fault diagnosis method for wind turbine gearboxes in this invention.

[0051] Figure 2 This is a structural block diagram of the wind turbine gearbox high-speed shaft bearing fault diagnosis system based on VMD in this invention. Detailed Implementation

[0052] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments.

[0053] Example 1:

[0054] like Figure 1 As shown in the figure, this embodiment provides a method for fault diagnosis of high-speed shaft bearings in wind turbine gearboxes based on VMD, including the following steps:

[0055] S1: Collect the vibration signal of the high-speed shaft bearing of the wind turbine gearbox, and preprocess the vibration signal to obtain the preprocessed vibration signal;

[0056] For example, an accelerometer is installed on the bearing housing of the high-speed shaft of a wind turbine gearbox to collect vibration signals at a set sampling frequency (e.g., 25600 Hz) and sampling duration (e.g., 10 s). The collected raw vibration signals usually contain interference such as DC components and environmental noise.

[0057] The collected vibration signals are preprocessed by denoising and normalization to suppress irrelevant frequency bands and environmental noise, thereby improving the analyzability and feature prominence of the signals.

[0058] For example, the signal is subjected to mean removal to eliminate the DC component; then, a bandpass filter (e.g., with cutoff frequencies of 1000 Hz and 8000 Hz) is used to filter the signal, retaining the effective frequency band containing the main fault characteristics and suppressing low-frequency and high-frequency noise to obtain a clean, denoised vibration signal; the denoised vibration signal is then normalized to eliminate the influence of dimensions, resulting in the final preprocessed vibration signal.

[0059] S2: The preprocessed vibration signal is decomposed using the VMD algorithm with multi-objective optimization to obtain several Intrinsic Mode Function (IMF) components;

[0060] The VMD algorithm is improved using the second-generation Non-dominated Sorting Genetic Algorithm II (NSGA-II) with an elitist strategy, resulting in a multi-objective VMD algorithm, which includes the following steps:

[0061] A-1: Determine the core parameters of the VMD algorithm and their range, and use the core parameters as independent variables to be optimized;

[0062] The core parameters of the VMD algorithm include the number of modes and the penalty factor; the number of modes determines the number of components in the decomposition, and the penalty factor controls the bandwidth of each component.

[0063] For example, the modality number is set to an integer range of [3, 10], and the penalty factor is set to a range of [100, 3000]. These two parameters are used to form a two-dimensional vector as the independent variable to be optimized in the NSGA-II algorithm.

[0064] A-2: Set the objective function of the NSGA-II algorithm, and use the NSGA-II algorithm iteratively to find the optimal solution based on the objective function and independent variables, including the following steps:

[0065] A-2-1: Set the objective function and algorithm parameters of the NSGA-II algorithm, and encode the independent variables as the solution vectors of individuals in the NSGA-II algorithm;

[0066] The objective functions include the permutation entropy function and the envelope entropy function;

[0067] Permutation entropy is used to measure the complexity and randomness of each intrinsic mode function (IMF) component. The average permutation entropy of the IMF components is often used as the permutation entropy function value.

[0068] Envelope entropy is used to measure the energy distribution and complexity of the envelope of each intrinsic mode function (IMF) component. The average envelope entropy of the IMF components is often used as the envelope entropy function value.

[0069] The formula is: ;

[0070] In the formula, n is the permutation entropy value; n is the permutation length; i is the permutation indicator. Let be the probability of the i-th permutation appearing;

[0071] The formula is: ;

[0072] In the formula, The envelope entropy value; This is the normalized envelope value of the signal at time t;

[0073] ;

[0074] In the formula, The average entropy is the permutation of the average values. For IMF component indication; This represents the total number of IMF components. The permutation entropy function; For the first One IMF component;

[0075] ;

[0076] In the formula, The average envelope entropy; It is the envelope entropy function;

[0077] A-2-2: Based on the algorithm parameters, initialize the NSGA-II population to obtain an initial NSGA-II population including several candidate solutions;

[0078] A-2-3: Based on the objective function, obtain the objective function value corresponding to each candidate solution in the initial NSGA-II population, including the following steps:

[0079] A-2-3-1: Decode any candidate solution in the initial NSGA-II population to obtain candidate core parameters, and improve the VMD algorithm based on the candidate core parameters to obtain a candidate VMD algorithm;

[0080] A-2-3-2: Based on the candidate VMD algorithm, the sample vibration signal is decomposed to obtain several sample IMF components;

[0081] A-2-3-3: Based on the permutation entropy function in the objective function, calculate the sample permutation entropy of several sample IMF components and output the corresponding average sample permutation entropy.

[0082] A-2-3-4: Based on the envelope entropy function in the objective function, calculate the sample envelope entropy of several sample IMF components and output the corresponding average sample envelope entropy.

[0083] A-2-3-5: By combining the average entropy of the sample permutation and the average entropy of the sample envelope, the objective function value corresponding to the current candidate solution is obtained;

[0084] A-2-3-6: Traverse all candidate solutions in the initial NSGA-II population to obtain several corresponding objective function values;

[0085] A-2-4: Based on the objective function value and the Pareto front sorting and crowding selection mechanism, the NSGA-II algorithm is used to perform iterative evolution, and crossover, mutation and selection operations are performed to continuously generate a new generation of population, resulting in a new NSGA-II population.

[0086] A-2-5: If the number of iterations reaches the maximum number of iterations, then output the Pareto front solution set in the final NSGA-II population based on the objective function value. That is, the parameter solution set that is better under both objectives. All solutions in this solution set cannot be surpassed by other solutions at the same time under both objective functions.

[0087] A-2-6: Using the curvature method, select the optimal compromise solution from the Pareto front solution set, and take the optimal compromise solution as the optimal solution;

[0088] On the Pareto front, the point with the largest curvature is usually considered the optimal trade-off point between the two objectives. Calculate the curvature of each point on the Pareto front and select the solution corresponding to the point with the largest curvature as the optimal compromise solution.

[0089] A-3: Decode the optimal solution to obtain the optimal core parameter values ​​of the VMD algorithm, namely the optimal number of modes and the optimal penalty factor. Based on the optimal core parameter values, improve the VMD algorithm to obtain a multi-objective optimization VMD algorithm.

[0090] The NSGA-II multi-objective optimization algorithm is adopted, with permutation entropy and envelope entropy as optimization objectives. The mode number and penalty factor of VMD are intelligently optimized. Minimizing permutation entropy ensures that the decomposed IMF components are more regular and less random, while minimizing envelope entropy ensures that the energy of the component envelope is more concentrated and can better highlight the characteristics of impact faults. The optimal compromise solution is selected by Pareto optimal solution set and curvature method, avoiding the one-sidedness of traditional empirical methods or single-objective optimization. It realizes the objective and adaptive selection of VMD parameters, laying a solid foundation for the accurate separation of fault features.

[0091] The VMD algorithm includes a constrained variational model and its constraints, as well as an augmented Lagrangian function, the formula of which is:

[0092] ;

[0093] In the formula, To minimize the set of variables { }and{ }; This represents the k-th component; k is the component indicator. It is an L2 norm; is the derivative with respect to time t; t is a time indicator. Here, j is the time partial derivative operator; j is the imaginary unit. This is the k-th component of time t; The total number of components; The center frequency of the k-th component;

[0094] ;

[0095] In the formula, It is a vibration signal;

[0096] ;

[0097] In the formula, To augment the Lagrange function; For Lagrange multipliers; For penalty parameters; Let be the Lagrange multiplier for time t;

[0098] S3: Filter several IMF components to obtain the optimal IMF component, and extract the envelope spectrum of the optimal IMF component, including the following steps:

[0099] S3-1: Calculate the kurtosis value of each IMF component to characterize the impulsiveness and non-Gaussianity of the component, and take the IMF component with the largest kurtosis value as the optimal IMF component. The optimal IMF component can best reflect the impact and abnormal components caused by mechanical failure and has the strongest failure characteristics.

[0100] The IMF components are screened using the kurtosis index. Kurtosis is very sensitive to the impact signal. Selecting the IMF component with the largest kurtosis as the optimal component can most effectively retain the periodic impact component caused by bearing failure, suppress the interference of noise and other irrelevant components, and make the subsequently extracted envelope spectrum clearer and the fault characteristics more prominent.

[0101] The formula is:

[0102] ;

[0103] In the formula, This represents the kurtosis value. For the first Data points (IMF components); For data point indication; The mean of the data is N; N is the total number of data points.

[0104] S3-2: Perform Hilbert transform on the optimal IMF component to obtain the corresponding analytical signal;

[0105] S3-3: Take the modulus of the analytic signal to obtain the corresponding envelope signal, and perform spectral analysis on the envelope signal to plot the corresponding envelope spectrum;

[0106] S4: Using a fault diagnosis model built based on deep learning algorithms, fault diagnosis is performed on the envelope spectrum of the optimal IMF component to obtain the fault diagnosis results;

[0107] The fault diagnosis model is constructed based on the Convolutional Neural Network (CNN) - Long Short-Term Memory (LSTM) - Attention - Multi-Layer Perceptron (MLP) algorithm. The fault diagnosis model includes a spatial feature extraction module based on the CNN algorithm, a temporal feature extraction module based on the LSTM algorithm, a weighted fusion module based on the Attention mechanism, and a fault diagnosis module based on the MLP algorithm, which are connected in sequence.

[0108] CNN excels at extracting local spatial features (such as fault frequency spectral lines) from the two-dimensional structure of the envelope spectrum; LSTM can capture the dynamic evolution of these features over time and their long-term and short-term dependencies; the attention mechanism can adaptively assign weights to different spatiotemporal features, thereby strengthening key features and suppressing redundant features; finally, MLP performs accurate predictions. This multi-module collaborative approach comprehensively and deeply mines the deep spatiotemporal correlation information in the envelope spectrum, and compared to a single model, it has stronger feature expression and discrimination capabilities.

[0109] A fault diagnosis model based on deep learning algorithms is used to diagnose faults in the envelope spectrum of the optimal IMF component, and the fault diagnosis results are obtained. The steps include:

[0110] S4-1: Input the envelope spectrum of the optimal IMF component into the fault diagnosis model constructed based on the deep learning algorithm;

[0111] S4-2: Use the spatial feature extraction module of the fault diagnosis model to extract the spatial features of the envelope spectrum;

[0112] S4-3: Use the time feature extraction module of the fault diagnosis model to extract the time features of the fault features;

[0113] S4-4: Based on the dynamic attention weights, the weighted fusion module of the fault diagnosis model is used to perform weighted fusion of spatial and temporal features to obtain fused features.

[0114] S4-5: Based on the fusion characteristics, use the fault diagnosis module of the fault diagnosis model to perform fault diagnosis and obtain the fault diagnosis results;

[0115] The fault diagnosis results include the fault type, location, and severity level of the high-speed shaft bearing in the wind turbine gearbox;

[0116] Fault types include bearing inner ring failure, bearing outer ring failure, rolling element failure, cage failure, shaft misalignment, gear tooth breakage / wear, etc.

[0117] The faulty parts include the high-speed shaft bearing of the wind turbine gearbox, the drive end / non-drive end, and the specific bearing number;

[0118] The severity levels include mild, moderate, and severe.

[0119] Example 2:

[0120] like Figure 2 As shown, this embodiment provides a VMD-based high-speed shaft bearing fault diagnosis system for wind turbine gearboxes, which is used to realize a fault diagnosis method for high-speed shaft bearings in wind turbine gearboxes. The system includes a vibration signal acquisition unit, a vibration signal decomposition unit, an IMF component screening unit, and a fault diagnosis unit connected in sequence.

[0121] The vibration signal acquisition unit is used to acquire the vibration signal of the high-speed shaft bearing of the wind turbine gearbox and to preprocess the vibration signal to obtain the preprocessed vibration signal.

[0122] The vibration signal decomposition unit is used to decompose the preprocessed vibration signal using the VMD algorithm with multi-objective optimization to obtain several IMF components.

[0123] The IMF component screening unit is used to screen several IMF components to obtain the optimal IMF component and extract the envelope spectrum of the optimal IMF component.

[0124] The fault diagnosis unit is used to perform fault diagnosis on the envelope spectrum of the optimal IMF component using a fault diagnosis model built based on deep learning algorithms, and obtain the fault diagnosis result.

[0125] This invention provides a method and system for fault diagnosis of high-speed shaft bearings in wind turbine gearboxes based on Virtual Machine Decomposition (VMD). The multi-objective optimization algorithm employed can quickly and accurately determine the core parameters of the VMD algorithm, overcoming the subjectivity and local optima problems of traditional parameter selection, reducing manual intervention, and significantly improving the scientific rigor and effectiveness of parameter determination. The resulting multi-objective optimized VMD algorithm takes into account both the complexity of the decomposed components and the energy concentration characteristics, achieving a more comprehensive and automated evaluation of the decomposition effect and improving the accuracy of fault diagnosis. The fault diagnosis model built based on deep learning algorithms enables automatic fault diagnosis, reducing labor costs and improving the efficiency of fault diagnosis.

[0126] This invention is not limited to the optional embodiments described above, and anyone can derive other various forms of products based on the inspiration of this invention. The specific embodiments described above should not be construed as limiting the scope of protection of this invention; the scope of protection of this invention should be determined by the claims, and the specification can be used to interpret the claims.

Claims

1. A method for fault diagnosis of high-speed shaft bearings in wind turbine gearboxes based on VMD, characterized in that: Includes the following steps: Vibration signals of the high-speed shaft bearing of the wind turbine gearbox are collected and preprocessed to obtain preprocessed vibration signals. The preprocessed vibration signal is decomposed using the VMD algorithm with multi-objective optimization to obtain several IMF components. The VMD algorithm is improved using the NSGA-II algorithm to obtain a multi-objective optimization VMD algorithm, which includes the following steps: Determine the core parameters and their ranges for the VMD algorithm, and use the core parameters as independent variables to be optimized; The core parameters of the VMD algorithm include the number of modes and the penalty factor; Set the objective function of the NSGA-II algorithm, and use the NSGA-II algorithm iteratively to find the optimal solution based on the objective function and independent variables, including the following steps: Set the objective function and algorithm parameters of the NSGA-II algorithm, and encode the independent variables as the solution vectors of individuals in the NSGA-II algorithm; The objective function includes the permutation entropy function and the envelope entropy function; Based on the algorithm parameters, the NSGA-II population is initialized to obtain an initial NSGA-II population including several candidate solutions. Based on the objective function, the objective function value corresponding to each candidate solution in the initial NSGA-II population is obtained, including the following steps: Decode any candidate solution in the initial NSGA-II population to obtain candidate core parameters, and improve the VMD algorithm based on the candidate core parameters to obtain a candidate VMD algorithm; Based on the candidate VMD algorithm, the sample vibration signal is decomposed to obtain several sample IMF components; Based on the permutation entropy function in the objective function, calculate the sample permutation entropy of several sample IMF components and output the corresponding average sample permutation entropy. Based on the envelope entropy function in the objective function, calculate the sample envelope entropy of several sample IMF components and output the corresponding average sample envelope entropy. By combining the average entropy of the sample permutation and the average entropy of the sample envelope, the objective function value corresponding to the current candidate solution is obtained; Traverse all candidate solutions in the initial NSGA-II population to obtain several corresponding objective function values; Based on the objective function value and the Pareto front sorting and crowding selection mechanism, the NSGA-II algorithm is used to perform iterative evolution to obtain a new NSGA-II population; If the number of iterations reaches the maximum number of iterations, then output the Pareto front solution set in the final NSGA-II population based on the objective function value; Using the curvature method, the optimal compromise solution is selected from the Pareto front solution set, and the optimal compromise solution is taken as the optimal solution; The optimal solution is decoded to obtain the optimal core parameter values ​​of the VMD algorithm. Based on the optimal core parameter values, the VMD algorithm is improved to obtain a multi-objective optimization VMD algorithm. The optimal IMF component is obtained by screening several IMF components, and its envelope spectrum is extracted. This process includes the following steps: Calculate the kurtosis value of each IMF component and select the IMF component with the largest kurtosis value as the optimal IMF component; Perform a Hilbert transform on the optimal IMF component to obtain the corresponding analytical signal; The analytic signal is moduloed to obtain the corresponding envelope signal, and the envelope signal is subjected to spectral analysis to plot the corresponding envelope spectrum; A fault diagnosis model based on deep learning algorithms is used to diagnose faults in the envelope spectrum of the optimal IMF component, and the fault diagnosis results are obtained.

2. The method for fault diagnosis of high-speed shaft bearings in wind turbine gearboxes based on VMD according to claim 1, characterized in that: The fault diagnosis model is constructed based on the CNN-LSTM-Attention-MLP algorithm, and includes a spatial feature extraction module based on the CNN algorithm, a temporal feature extraction module based on the LSTM algorithm, a weighted fusion module based on the Attention mechanism, and a fault diagnosis module based on the MLP algorithm, which are connected in sequence.

3. The method for fault diagnosis of high-speed shaft bearings in wind turbine gearboxes based on VMD according to claim 2, characterized in that: A fault diagnosis model based on deep learning algorithms is used to diagnose faults in the envelope spectrum of the optimal IMF component, and the fault diagnosis results are obtained. The steps include: The envelope spectrum of the optimal IMF component is input into the fault diagnosis model constructed based on a deep learning algorithm; The spatial feature extraction module of the fault diagnosis model is used to extract the spatial features of the envelope spectrum. The time feature extraction module of the fault diagnosis model is used to extract the time features of the fault. Based on the dynamic attention weights, the weighted fusion module of the fault diagnosis model is used to perform weighted fusion of spatial and temporal features to obtain fused features. Based on the fusion characteristics, the fault diagnosis module of the fault diagnosis model is used to perform fault diagnosis and obtain the fault diagnosis results.

4. A VMD-based high-speed shaft bearing fault diagnosis system for wind turbine gearboxes, used to implement the fault diagnosis method for high-speed shaft bearings of wind turbine gearboxes as described in any one of claims 1-3, characterized in that: The system includes a vibration signal acquisition unit, a vibration signal decomposition unit, an IMF component filtering unit, and a fault diagnosis unit connected in sequence.

Images