A method for fault diagnosis of wind turbine bearings under varying operating conditions based on DBO-VMD

Abstract

This invention discloses a fault diagnosis method for wind turbine bearings under varying operating conditions based on DBO-VMD, specifically implemented according to the following steps: Step 1, data preprocessing stage; Step 2, DBO algorithm optimization of VMD parameters [K,α] stage; Step 3, VMD decomposition and signal reconstruction stage; Step 4, dataset generation stage; Step 5, fault diagnosis stage. This invention solves the problems of insufficient feature extraction capability under varying operating conditions and slow convergence speed of VMD decomposition optimization methods in existing technologies for wind turbine bearing fault diagnosis.

Description

Technical Field

[0001] This invention belongs to the field of wind turbine bearing fault diagnosis technology, specifically relating to a method for diagnosing wind turbine bearing faults under varying operating conditions based on DBO-VMD. Background Technology

[0002] Fans are fluid machines widely used in industry, agriculture, construction, transportation, and other fields. Their main function is to convert electrical energy or other forms of energy into the kinetic and pressure energy of gas flow, thereby achieving gas transport, pressurization, circulation, and ventilation. Fan bearings are a crucial component of fans, bearing the weight and unbalanced forces of the fan rotor to ensure its stable operation. However, fan bearings are susceptible to various factors during operation, such as load changes, poor lubrication, improper installation, material defects, and changes in ambient temperature and humidity, leading to various faults such as wear, cracks, spalling, and pitting. These faults can cause a decline in fan bearing performance, and even fan shutdown or accidents, resulting in serious losses to production and safety. Therefore, timely and effective detection and diagnosis of fan bearing faults are of great significance for improving fan operating efficiency and reliability, extending fan life, reducing maintenance costs and downtime losses, and ensuring production safety.

[0003] Fault feature extraction is a core step in wind turbine bearing fault diagnosis. Wind turbine bearings operate in complex environments, and the collected vibration signals are highly noisy, obscuring fault features. Therefore, effectively extracting fault features is one of the main research areas in wind turbine bearing fault diagnosis. Currently, bearing fault feature extraction mainly employs time-frequency combined analysis methods. Among these, Variational Mode Decomposition (VMD) has shown excellent performance in feature extraction from complex bearing signals since its inception. It can effectively eliminate background noise and significantly improve the accuracy of subsequent bearing fault detection. However, the parameters of VMD decomposition […]. , The parameters of VMD need to be set manually, which has a significant impact on decomposition performance. Many scholars have addressed this issue based on the theory of intelligent optimization algorithms. , Intelligent optimization was performed to find the optimal parameters, which effectively avoided potential modal aliasing issues. However, the optimization effects of different intelligent optimization algorithms vary. Therefore, it is necessary to find an intelligent optimization algorithm with fast convergence properties, good adaptability to wind turbine bearing faults, and the ability to effectively improve the accuracy of fault diagnosis. Summary of the Invention

[0004] The purpose of this invention is to provide a method for diagnosing wind turbine bearing faults under varying operating conditions based on DBO-VMD, which solves the problems of insufficient fault feature extraction capability under varying operating conditions and slow convergence speed of VMD decomposition optimization method in the existing technology of wind turbine bearing fault diagnosis.

[0005] The technical solution adopted in this invention is a method for diagnosing wind turbine bearing faults under variable operating conditions based on DBO-VMD, which is implemented according to the following steps:

[0006] Step 1: Data preprocessing stage;

[0007] Step 2: Optimize VMD parameters using the DBO algorithm. , ]stage;

[0008] Step 3: VMD decomposition and signal reconstruction stage;

[0009] Step 4: Dataset generation stage;

[0010] Step 5: Fault Diagnosis Stage.

[0011] The invention is further characterized in that,

[0012] Step 1 is implemented in the following steps:

[0013] Step 1.1: Select the driver-side data from the CWRU dataset;

[0014] Step 1.2: Create driver-side data tags, including 10 types of data, of which 9 are fault data: ba_7, ba_14, ba_21, ir_7, ir_14, ir_21, or_7, or_14, or_21, and 1 type of normal state data nc; ba_X is for rolling element fault, ir_X is for inner ring fault, and or_X is for outer ring fault.

[0015] In step 2, firstly, bearing vibration data of one state is selected as the input data for the DBO algorithm; then, the fitness function of the DBO algorithm is constructed; the parameters of the DBO algorithm are initialized, and the parameters of the DBO algorithm are iterated; finally, the running results of the DBO algorithm are obtained and the results are substituted into the VMD algorithm.

[0016] Step 2 is implemented in the following steps:

[0017] Step 2.1: Extract bearing fault data segments;

[0018] Step 2.2: Construct the average envelope entropy function for the data segment;

[0019] Step 2.3: Establish the DBO algorithm model, initialize the parameters, and begin M algorithm iterations;

[0020] Step 2.4: In one iteration, start a loop with a loop size of N equal to the population size, and execute steps 2.5-2.8;

[0021] Step 2.5: Update the location information of the dung beetle;

[0022] Step 2.6: Update the location information of the incubation ball;

[0023] Step 2.7: Update the location information of adult dung beetles;

[0024] Step 2.8: Update the location information of the dung beetle thief;

[0025] Step 2.9: Increment the loop count by 1;

[0026] Step 2.10: After the current iteration ends, increment the main loop count by 1;

[0027] Step 2.11: Obtain the global optimal position and fitness function value.

[0028] In step 2,

[0029] Model establishment of DBO algorithm: The simulated dung beetle behaviors include the following:

[0030] (a) Rolling ball

[0031] Assuming the dung beetle moves along a given direction throughout the search space, using sunlight as its navigation, and assuming that light intensity also affects the dung beetle's path, the formula for calculating the dung beetle's position update during rolling is:

[0032] (1)

[0033] In the formula: The updated position of the rolling dung beetle; This represents the current iteration number; Location information for the dung beetle; is a constant representing the deflection coefficient, with a value range of (0, 0.2]. The value range is (0,1); The natural coefficient; This is the worst position globally. For changes in light intensity;

[0034] In equation (1), the parameter and The appropriate selection is crucial; generally speaking, The value is 0.1. The value is 0.3. Simulate the natural factors that cause dung beetles to deviate from their original direction, when A value of 1 indicates that the direction has not deviated. A value of -1 indicates that the direction deviates from the original direction. A higher value indicates a weaker light source. Control the size of the light source to expand the search range;

[0035] (b) Dancing

[0036] The new rolling direction is obtained using the tangent function. After obtaining the new direction, the dung beetle will roll the ball in this direction. The position update formula is as shown in equation (2):

[0037] (2)

[0038] In the formula: The updated location of the Dancing Dung Beetle; The deflection angle is [0, ...]. ];

[0039] In the above formula, The value is 0. / 2、 If the position is not updated, it is assumed that the position has not been updated; if other values ​​are taken, it is assumed that the position of the dung beetle has been updated.

[0040] (c) Reproduction

[0041] A boundary selection strategy is proposed, and the concept of the boundary selection strategy is defined as shown in equation (3):

[0042] (3)

[0043] In the formula: These are the lower and upper boundaries of the spawning area; This is the current local optimum position; To optimize the lower and upper bounds of the problem; , Indicates the maximum number of iterations;

[0044] Once the oviposition area is determined, the female dung beetle will choose to lay its eggs in this area. In the DBO algorithm, it is assumed that each female dung beetle lays only one egg in each iteration. According to equation (3), the boundary range of the oviposition area changes dynamically with the value. Therefore, the position of the egg also changes dynamically during the iteration process. Its update formula is as follows:

[0045] (4)

[0046] In the formula: The updated location of the oocyte; For the first Position information for the next iteration; They are two independent random vectors;

[0047] (d) Foraging

[0048] Based on the foraging process of adult dung beetles, an optimal foraging area model is established, as shown in equation (5):

[0049] (5)

[0050] In the formula: These are the lower and upper boundaries of the optimal foraging area; The globally optimal position;

[0051] Based on the principle of reproductive location update, the following formula for dung beetle foraging location update is obtained:

[0052] (6)

[0053] In the formula: The updated location of the foraging dung beetle; These are random numbers that follow a normal distribution.

[0054] (e) theft

[0055] Within a dung beetle colony, some dung beetles steal the dung balls of other dung beetles; these are called thieving dung beetles. As can be seen from equation (5), It is the optimal food source, therefore, assuming The area around the dung beetle is the best location for competing for food, and the dung beetle's location information is updated according to the following formula:

[0056] (7)

[0057] In the formula: The updated location of the dung beetle; It is a constant; It is a random vector that follows a normal distribution;

[0058] Selection of fitness function: The envelope entropy of the modes after VMD decomposition is selected as the fitness function of DBO. Its calculation formula is as follows:

[0059] (8)

[0060] In the formula: for Normalized representation; The envelope signal after demodulation by Hilbert transform; Let be the envelope entropy.

[0061] Through iteration of the DBO algorithm, the optimal parameters for 10 types of data VMD decomposition can be obtained. , And the minimum value of the average envelope entropy.

[0062] Step 3 first involves running VMD decomposition to obtain the corresponding modes and mode spectra; then, fault feature extraction is performed to obtain the reconstructed signal. Step 3 is implemented specifically according to the following steps:

[0063] Step 3.1: Use the results of the DBO algorithm as the initialization parameters for VMD decomposition;

[0064] Step 3.2: Run VMD decomposition;

[0065] Step 3.3: Obtain the time-domain plot and spectrum of each mode;

[0066] Step 3.4: Calculate the kurtosis of each mode;

[0067] Step 3.5: Select the mode corresponding to the maximum kurtosis for envelope spectrum analysis;

[0068] Step 3.6: Select the mode corresponding to the maximum kurtosis as the reconstructed signal;

[0069] Step 3.7: Fault feature extraction complete.

[0070] The VMD decomposition stage in step 3 is as follows:

[0071] This stage includes using the results calculated by the DBO algorithm to perform VMD decomposition on the input data. The principle of VMD decomposition is: first, the process of establishing the variational model:

[0072] Signal Decomposed into several modal component signals As shown in equation (9), each mode is demodulated to obtain a single-sided spectrum.

[0073] (9)

[0074] In the formula: Unit impact function;

[0075] By introducing a center frequency The spectrum of each demodulated signal is modulated onto the corresponding baseband, as shown in equation (10):

[0076] (10)

[0077] The square norm of the gradient in equation (10) is calculated and the bandwidth corresponding to each modal component is estimated, resulting in the following constrained variational problem model.

[0078] (11)

[0079] In the formula: After decomposition One modal component; for The center frequency of each modal component;

[0080] Next, the variational model addresses the problem. To solve the constraint problem established by equation (11), a quadratic penalty factor is introduced. and Lagrange multiplication operators This transforms the problem into an unconstrained variational problem as follows:

[0081] (12)

[0082] In the formula: For Lagrange multiplication operators; To represent the inner product operation;

[0083] In equation (12), through continuous updating , , The optimal solution to the variational problem can be obtained by finding the values ​​of the three terms. , , The update formulas are respectively equations (13) to (15):

[0084] (13)

[0085] (14)

[0086] (15)

[0087] In the formula: for Fourier transform; for Fourier transform; for Fourier transform;

[0088] By following the steps above, the signal's mode, spectrum, and center frequency can be determined.

[0089] The signal reconstruction stage in step 3 is as follows:

[0090] The kurtosis of the modes in the DBO-VMD decomposition is calculated using the following formula:

[0091] (16)

[0092] The mode with the maximum kurtosis was selected as the extracted bearing fault feature.

[0093] Step 4: First, perform overlap sampling and CQT transform on the reconstructed signal to obtain the time-frequency diagram. Then, divide the dataset. Step 4 is implemented in the following steps:

[0094] Step 4.1: Import the data after fault feature extraction;

[0095] Step 4.2: Calculate the number of overlapping samples;

[0096] Step 4.3: Perform CQT transformation on each sample;

[0097] Step 4.4: Create a bearing dataset with 10 states;

[0098] Step 4.5: Swap the dataset according to the 8:1:1 ratio;

[0099] Dataset generation stage: The extracted fault features are used to generate CQT time-frequency maps using the overlapping sampling method. Assuming the data length of the overlapping sampling is L, and CQT transformation is performed on 1024 points of length T, with a step size of S=200 for each movement, the number of samples N that can be generated from one data file is determined by the following formula.

[0100] (17)

[0101] The generated dataset is divided into training, validation, and test sets in an 8:1:1 ratio.

[0102] Step 5 is implemented in the following steps:

[0103] Step 5.1: Train the RepVGG network using the partitioned dataset and save the model weights;

[0104] Step 5.2: Load the trained RepVGG model weights and use the test set for prediction;

[0105] Step 5.3: Obtain the fault diagnosis results;

[0106] The training and validation sets are input into the RepVGG model. After 60 training iterations, the weights are saved. During testing, the weights are loaded to perform bearing fault diagnosis.

[0107] The beneficial effects of this invention are that the DBO-VMD-based method for fault diagnosis of wind turbine bearings under varying operating conditions, taking the fault diagnosis of wind turbine bearings under varying operating conditions as the background, uses the VMD decomposition algorithm as the basis, and optimizes the parameters based on the DBO algorithm, ultimately achieving high-precision bearing fault diagnosis results. It eliminates the need for manually setting VMD parameters in the original method. , This method utilizes the DBO algorithm to accurately calculate the VMD parameters of fault data, effectively avoiding modal aliasing problems. It achieves intelligent and automated extraction of fault features from wind turbine bearings under varying operating conditions, improving the control level of wind turbine bearing operation and maintenance, reducing human resource costs, significantly enhancing the results of wind turbine bearing fault diagnosis, and reducing the economic expenditure of wind turbine management. If promoted and applied domestically, it will generate significant direct and indirect economic and social benefits. Attached Figure Description

[0108] Figure 1 The present invention provides a method for diagnosing wind turbine bearing faults under varying operating conditions based on DBO-VMD. (Flowchart illustrating the fault diagnosis process for wind turbine bearings under varying operating conditions.)

[0109] Figure 2 The flowchart of the DBO algorithm for the variable operating condition wind turbine bearing fault diagnosis method based on DBO-VMD is shown in this invention.

[0110] Figure 3 This is a flowchart of the DBO-VMD fault feature extraction method for wind turbine bearing fault diagnosis under varying operating conditions, based on the present invention.

[0111] Figure 4 This is a flowchart of the dataset creation for the variable operating condition wind turbine bearing fault diagnosis method based on DBO-VMD of this invention;

[0112] Figure 5 The diagram shows the RepVGG network structure of the variable operating condition wind turbine bearing fault diagnosis method based on DBO-VMD of this invention. Detailed Implementation

[0113] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0114] This invention relates to a method for diagnosing wind turbine bearing faults under varying operating conditions based on DBO-VMD. The flowchart is as follows: Figure 1 As shown, please follow these steps:

[0115] Step 1: Data preprocessing stage;

[0116] Step 1 is implemented in the following steps:

[0117] Step 1.1: Select the driver-side data from the CWRU dataset;

[0118] Step 1.2: Create driver-side data tags, including 10 types of data, of which 9 are fault data: ba_7, ba_14, ba_21, ir_7, ir_14, ir_21, or_7, or_14, or_21, and 1 type of normal state data nc; ba_X is for rolling element fault, ir_X is for inner ring fault, and or_X is for outer ring fault.

[0119] Step 2: Optimize VMD parameters using the DBO algorithm. , ]stage;

[0120] Combination Figures 2-4 In step 2, firstly, bearing vibration data of a certain state is selected as the input data for the DBO algorithm; then, the fitness function of the DBO algorithm is constructed; the parameters of the DBO algorithm are initialized, and the parameters of the DBO algorithm are iterated; finally, the running results of the DBO algorithm are obtained and the results are substituted into the VMD algorithm.

[0121] Step 2 is implemented in the following steps:

[0122] Step 2.1: Extract bearing fault data segments;

[0123] Step 2.2: Construct the average envelope entropy function for the data segment;

[0124] Step 2.3: Establish the DBO algorithm model, initialize the parameters, and begin M algorithm iterations;

[0125] Step 2.4: In one iteration, start a loop with a loop size of N equal to the population size, and execute steps 2.5-2.8;

[0126] Step 2.5: Update the location information of the dung beetle;

[0127] Step 2.6: Update the location information of the incubation ball;

[0128] Step 2.7: Update the location information of adult dung beetles;

[0129] Step 2.8: Update the location information of the dung beetle thief;

[0130] Step 2.9: Increment the loop count by 1;

[0131] Step 2.10: After the current iteration ends, increment the main loop count by 1;

[0132] Step 2.11: Obtain the global optimal position and fitness function value.

[0133] In step 2,

[0134] Model establishment of the DBO algorithm: The DBO algorithm mainly simulates the dung beetle's behaviors in nature, including rolling, dancing, foraging, stealing, and reproduction. The specific behaviors of the simulated dung beetle include the following:

[0135] (a) Rolling ball

[0136] To simulate the behavior of the rolling ball, the dung beetle is assumed to move along a given direction throughout the search space, using sunlight as its navigation. It is also assumed that the intensity of the light source affects the dung beetle's path. The formula for calculating the dung beetle's position update during the rolling process is:

[0137] (1)

[0138] In the formula: The updated position of the rolling dung beetle; This represents the current iteration number; Location information for the dung beetle; is a constant representing the deflection coefficient, with a value range of (0, 0.2]. The value range is (0,1); The natural coefficient; This is the worst position globally. For changes in light intensity;

[0139] In equation (1), the parameter and The appropriate selection is crucial; generally speaking, The value is 0.1. The value is 0.3. Simulate the natural factors that cause dung beetles to deviate from their original direction, when A value of 1 indicates that the direction has not deviated. A value of -1 indicates that the direction deviates from the original direction. A higher value indicates a weaker light source. By controlling the size of the light source and expanding the search range, the DBO algorithm can search the entire problem space as much as possible during the optimization process, while reducing the possibility of getting trapped in local optima.

[0140] (b) Dancing

[0141] When a dung beetle encounters an obstacle and cannot move forward, it will dance to reorient itself and obtain a new route. To simulate the dancing behavior, the tangent function is used to obtain the new rolling direction. After obtaining the new direction, the dung beetle will roll the ball in this direction. The position update formula is as shown in equation (2):

[0142] (2)

[0143] In the formula: The updated location of the Dancing Dung Beetle; The deflection angle is [0, ...]. ];

[0144] In the above formula, The value is 0. / 2、 If the position is not updated, it is assumed that the position has not been updated; if other values ​​are taken, it is assumed that the position of the dung beetle has been updated.

[0145] (c) Reproduction

[0146] Choosing a suitable oviposition site is crucial for dung beetles to provide a safe environment for their offspring. To simulate the oviposition area of ​​female dung beetles, a boundary selection strategy is proposed, defined as shown in equation (3):

[0147] (3)

[0148] In the formula: These are the lower and upper boundaries of the spawning area; This is the current local optimum position; To optimize the lower and upper bounds of the problem; , Indicates the maximum number of iterations;

[0149] Once the oviposition area is determined, the female dung beetle will choose to lay its eggs in this area. In the DBO algorithm, it is assumed that each female dung beetle lays only one egg in each iteration. According to equation (3), the boundary range of the oviposition area changes dynamically with the value. Therefore, the position of the egg also changes dynamically during the iteration process. Its update formula is as follows:

[0150] (4)

[0151] In the formula: The updated location of the oocyte; For the first Position information for the next iteration; They are two independent random vectors;

[0152] (d) Foraging

[0153] Based on the foraging process of adult dung beetles, an optimal foraging area model is established, as shown in equation (5):

[0154] (5)

[0155] In the formula: These are the lower and upper boundaries of the optimal foraging area; The globally optimal position;

[0156] Based on the principle of reproductive location update, the following formula for dung beetle foraging location update is obtained:

[0157] (6)

[0158] In the formula: The updated location of the foraging dung beetle; These are random numbers that follow a normal distribution.

[0159] (e) theft

[0160] Within a dung beetle colony, some dung beetles steal the dung balls of other dung beetles; these are called thieving dung beetles. As can be seen from equation (5), It is the optimal food source, therefore, assuming The area around the dung beetle is the best location for competing for food, and the dung beetle's location information is updated according to the following formula:

[0161] (7)

[0162] In the formula: The updated location of the dung beetle; It is a constant; It is a random vector that follows a normal distribution;

[0163] Selection of fitness function: The envelope entropy of the modes after VMD decomposition is selected as the fitness function of DBO. Its calculation formula is as follows:

[0164] (8)

[0165] In the formula: for Normalized representation; The envelope signal after demodulation by Hilbert transform; Let be the envelope entropy.

[0166] Through iteration of the DBO algorithm, the optimal parameters for 10 types of data VMD decomposition can be obtained. , And the minimum value of the average envelope entropy.

[0167] Step 3: VMD decomposition and signal reconstruction stage;

[0168] Step 3 first involves running VMD decomposition to obtain the corresponding modes and mode spectra; then, fault feature extraction is performed to obtain the reconstructed signal. Step 3 is implemented specifically according to the following steps:

[0169] Step 3.1: Use the results of the DBO algorithm as the initialization parameters for VMD decomposition;

[0170] Step 3.2: Run VMD decomposition;

[0171] Step 3.3: Obtain the time-domain plot and spectrum of each mode;

[0172] Step 3.4: Calculate the kurtosis of each mode;

[0173] Step 3.5: Select the mode corresponding to the maximum kurtosis for envelope spectrum analysis;

[0174] Step 3.6: Select the mode corresponding to the maximum kurtosis as the reconstructed signal;

[0175] Step 3.7: Fault feature extraction complete.

[0176] The VMD decomposition stage in step 3 is as follows:

[0177] This stage includes using the results calculated by the DBO algorithm to perform VMD decomposition on the input data. The principle of VMD decomposition is: first, the process of establishing the variational model:

[0178] Signal Decomposed into several modal component signals As shown in equation (9), each mode is demodulated to obtain a single-sided spectrum.

[0179] (9)

[0180] In the formula: Unit impact function;

[0181] By introducing a center frequency The spectrum of each demodulated signal is modulated onto the corresponding baseband, as shown in equation (10):

[0182] (10)

[0183] The square norm of the gradient in equation (10) is calculated and the bandwidth corresponding to each modal component is estimated, resulting in the following constrained variational problem model.

[0184] (11)

[0185] In the formula: After decomposition One modal component; for The center frequency of each modal component;

[0186] Next, the variational model addresses the problem. To solve the constraint problem established by equation (11), a quadratic penalty factor is introduced. and Lagrange multiplication operators This transforms the problem into an unconstrained variational problem as follows:

[0187] (12)

[0188] In the formula: For Lagrange multiplication operators; To represent the inner product operation;

[0189] In equation (12), through continuous updating , , The optimal solution to the variational problem can be obtained by finding the values ​​of the three terms. , , The update formulas are respectively equations (13) to (15):

[0190] (13)

[0191] (14)

[0192] (15)

[0193] In the formula: for Fourier transform; for Fourier transform; for Fourier transform;

[0194] By following the steps above, the signal's mode, spectrum, and center frequency can be determined.

[0195] The signal reconstruction stage in step 3 is as follows:

[0196] The kurtosis of the modes in the DBO-VMD decomposition is calculated using the following formula:

[0197] (16)

[0198] The mode with the maximum kurtosis was selected as the extracted bearing fault feature.

[0199] Step 4: Dataset generation stage;

[0200] Step 4: First, perform overlap sampling and CQT transform on the reconstructed signal to obtain the time-frequency diagram. Then, divide the dataset. Step 4 is implemented in the following steps:

[0201] Step 4.1: Import the data after fault feature extraction;

[0202] Step 4.2: Calculate the number of overlapping samples;

[0203] Step 4.3: Perform CQT transformation on each sample;

[0204] Step 4.4: Create a bearing dataset with 10 states;

[0205] Step 4.5: Swap the dataset according to the 8:1:1 ratio;

[0206] Dataset generation stage: The extracted fault features are used to generate CQT time-frequency maps using the overlapping sampling method. Assuming the data length of the overlapping sampling is L, and CQT transformation is performed on 1024 points of length T, with a step size of S=200 for each movement, the number of samples N that can be generated from one data file is determined by the following formula.

[0207] (17)

[0208] The generated dataset is divided into training, validation, and test sets in an 8:1:1 ratio.

[0209] Step 5: Fault Diagnosis Stage.

[0210] Combination Figure 5 Step 5 is implemented in the following steps:

[0211] Step 5.1: Train the RepVGG network using the partitioned dataset and save the model weights;

[0212] Step 5.2: Load the trained RepVGG model weights and use the test set for prediction;

[0213] Step 5.3: Obtain the fault diagnosis results;

[0214] The training and validation sets are input into the RepVGG model. After 60 training iterations, the weights are saved. During testing, the weights are loaded to perform bearing fault diagnosis.

[0215] The results obtained by the DBO-VMD algorithm of this invention after testing are shown in Table 1:

[0216] Table 1. Calculation results of DBO-VMD

[0217]

[0218] The fault diagnosis results of this invention are shown in Table 2:

[0219] Table 2 Fault Diagnosis Results

[0220]

[0221] Example 1

[0222] In the DBO-VMD-based method for diagnosing wind turbine bearing faults under varying operating conditions, the data preprocessing stage is implemented according to the following steps:

[0223] Step 1.1: Select the driver-side data from the CWRU dataset;

[0224] Step 1.2: Create driver-side data tags, including 10 types of data, of which 9 are fault data: ba_7, ba_14, ba_21, ir_7, ir_14, ir_21, or_7, or_14, or_21, and 1 type of normal state data nc; ba_X is for rolling element fault, ir_X is for inner ring fault, and or_X is for outer ring fault.

[0225] Step 2: Optimize VMD parameters using the DBO algorithm. , ]stage;

[0226] Combination Figures 2-4 In step 2, firstly, bearing vibration data of a certain state is selected as the input data for the DBO algorithm; then, the fitness function of the DBO algorithm is constructed; the parameters of the DBO algorithm are initialized, and the parameters of the DBO algorithm are iterated; finally, the running results of the DBO algorithm are obtained and the results are substituted into the VMD algorithm.

[0227] Step 2 is implemented in the following steps:

[0228] Step 2.1: Extract bearing fault data segments;

[0229] Step 2.2: Construct the average envelope entropy function for the data segment;

[0230] Step 2.3: Establish the DBO algorithm model, initialize the parameters, and begin M algorithm iterations;

[0231] Step 2.4: In one iteration, start a loop with a loop size of N equal to the population size, and execute steps 2.5-2.8;

[0232] Step 2.5: Update the location information of the dung beetle;

[0233] Step 2.6: Update the location information of the incubation ball;

[0234] Step 2.7: Update the location information of adult dung beetles;

[0235] Step 2.8: Update the location information of the dung beetle thief;

[0236] Step 2.9: Increment the loop count by 1;

[0237] Step 2.10: After the current iteration ends, increment the main loop count by 1;

[0238] Step 2.11: Obtain the global optimal position and fitness function value.

[0239] Example 2

[0240] In this invention, the DBO algorithm optimizes VMD parameters in a variable-condition wind turbine bearing fault diagnosis method based on DBO-VMD. , [Stage, combined] Figures 2-4 In step 2, firstly, bearing vibration data of a certain state is selected as the input data for the DBO algorithm; then, the fitness function of the DBO algorithm is constructed; the parameters of the DBO algorithm are initialized, and the parameters of the DBO algorithm are iterated; finally, the running results of the DBO algorithm are obtained and the results are substituted into the VMD algorithm.

[0241] Step 2 is implemented in the following steps:

[0242] Step 2.1: Extract bearing fault data segments;

[0243] Step 2.2: Construct the average envelope entropy function for the data segment;

[0244] Step 2.3: Establish the DBO algorithm model, initialize the parameters, and begin M algorithm iterations;

[0245] Step 2.4: In one iteration, start a loop with a loop size of N equal to the population size, and execute steps 2.5-2.8;

[0246] Step 2.5: Update the location information of the dung beetle;

[0247] Step 2.6: Update the location information of the incubation ball;

[0248] Step 2.7: Update the location information of adult dung beetles;

[0249] Step 2.8: Update the location information of the dung beetle thief;

[0250] Step 2.9: Increment the loop count by 1;

[0251] Step 2.10: After the current iteration ends, increment the main loop count by 1;

[0252] Step 2.11: Obtain the global optimal position and fitness function value.

[0253] In step 2,

[0254] Model establishment of the DBO algorithm: The DBO algorithm mainly simulates the dung beetle's behaviors in nature, including rolling, dancing, foraging, stealing, and reproduction. The specific behaviors of the simulated dung beetle include the following:

[0255] (a) Rolling ball

[0256] To simulate the behavior of the rolling ball, the dung beetle is assumed to move along a given direction throughout the search space, using sunlight as its navigation. It is also assumed that the intensity of the light source affects the dung beetle's path. The formula for calculating the dung beetle's position update during the rolling process is:

[0257] (1)

[0258] In the formula: The updated position of the rolling dung beetle; This represents the current iteration number; Location information for the dung beetle; is a constant representing the deflection coefficient, with a value range of (0, 0.2]. The value range is (0,1); The natural coefficient; This is the worst position globally. For changes in light intensity;

[0259] In equation (1), the parameter and The appropriate selection is crucial; generally speaking, The value is 0.1. The value is 0.3. Simulate the natural factors that cause dung beetles to deviate from their original direction, when A value of 1 indicates that the direction has not deviated. A value of -1 indicates that the direction deviates from the original direction. A higher value indicates a weaker light source. By controlling the size of the light source and expanding the search range, the DBO algorithm can search the entire problem space as much as possible during the optimization process, while reducing the possibility of getting trapped in local optima.

[0260] (b) Dancing

[0261] When a dung beetle encounters an obstacle and cannot move forward, it will dance to reorient itself and obtain a new route. To simulate the dancing behavior, the tangent function is used to obtain the new rolling direction. After obtaining the new direction, the dung beetle will roll the ball in this direction. The position update formula is as shown in equation (2):

[0262] (2)

[0263] In the formula: The updated location of the Dancing Dung Beetle; The deflection angle is [0, ...]. ];

[0264] In the above formula, The value is 0. / 2、 If the position is not updated, it is assumed that the position has not been updated; if other values ​​are taken, it is assumed that the position of the dung beetle has been updated.

[0265] (c) Reproduction

[0266] Choosing a suitable oviposition site is crucial for dung beetles to provide a safe environment for their offspring. To simulate the oviposition area of ​​female dung beetles, a boundary selection strategy is proposed, defined as shown in equation (3):

[0267] (3)

[0268] In the formula: These are the lower and upper boundaries of the spawning area; This is the current local optimum position; To optimize the lower and upper bounds of the problem; , Indicates the maximum number of iterations;

[0269] Once the oviposition area is determined, the female dung beetle will choose to lay its eggs in this area. In the DBO algorithm, it is assumed that each female dung beetle lays only one egg in each iteration. According to equation (3), the boundary range of the oviposition area changes dynamically with the value. Therefore, the position of the egg also changes dynamically during the iteration process. Its update formula is as follows:

[0270] (4)

[0271] In the formula: The updated location of the oocyte; For the first Position information for the next iteration; They are two independent random vectors;

[0272] (d) Foraging

[0273] Based on the foraging process of adult dung beetles, an optimal foraging area model is established, as shown in equation (5):

[0274] (5)

[0275] In the formula: These are the lower and upper boundaries of the optimal foraging area; The globally optimal position;

[0276] Based on the principle of reproductive location update, the following formula for dung beetle foraging location update is obtained:

[0277] (6)

[0278] In the formula: The updated location of the foraging dung beetle; These are random numbers that follow a normal distribution.

[0279] (e) theft

[0280] Within a dung beetle colony, some dung beetles steal the dung balls of other dung beetles; these are called thieving dung beetles. As can be seen from equation (5), It is the optimal food source, therefore, assuming The area around the dung beetle is the best location for competing for food, and the dung beetle's location information is updated according to the following formula:

[0281] (7)

[0282] In the formula: The updated location of the dung beetle; It is a constant; It is a random vector that follows a normal distribution;

[0283] Selection of fitness function: The envelope entropy of the modes after VMD decomposition is selected as the fitness function of DBO. Its calculation formula is as follows:

[0284] (8)

[0285] In the formula: for Normalized representation; The envelope signal after demodulation by Hilbert transform; Let be the envelope entropy.

[0286] Through iteration of the DBO algorithm, the optimal parameters for 10 types of data VMD decomposition can be obtained. , And the minimum value of the average envelope entropy.

[0287] Example 3

[0288] The VMD decomposition and signal reconstruction stages in the variable operating condition wind turbine bearing fault diagnosis method based on DBO-VMD of this invention are as follows: First, VMD decomposition is performed to obtain the corresponding modes and mode spectra; then, fault feature extraction is performed to obtain the reconstructed signal. Step 3 is implemented according to the following steps:

[0289] Step 3.1: Use the results of the DBO algorithm as the initialization parameters for VMD decomposition;

[0290] Step 3.2: Run VMD decomposition;

[0291] Step 3.3: Obtain the time-domain plot and spectrum of each mode;

[0292] Step 3.4: Calculate the kurtosis of each mode;

[0293] Step 3.5: Select the mode corresponding to the maximum kurtosis for envelope spectrum analysis;

[0294] Step 3.6: Select the mode corresponding to the maximum kurtosis as the reconstructed signal;

[0295] Step 3.7: Fault feature extraction complete.

Claims

1. A method for fault diagnosis of wind turbine bearings under variable operating conditions based on DBO-VMD, characterized in that, The specific steps are as follows: Step 1: Data preprocessing stage; Step 1 is implemented in the following steps: Step 1.1: Select the driver-side data from the CWRU dataset; Step 1.2: Create driver-side data tags, including 10 types of data, of which 9 are fault data: ba_7, ba_14, ba_21, ir_7, ir_14, ir_21, or_7, or_14, or_21, and 1 type of normal data nc; ba_X indicates rolling element fault, ir_X indicates inner ring fault, and or_X indicates outer ring fault; Step 2: Optimize VMD parameters using the DBO algorithm. , ]stage; In step 2, firstly, bearing vibration data of a certain state is selected as the input data for the DBO algorithm; then, the fitness function of the DBO algorithm is constructed; the parameters of the DBO algorithm are initialized, and the parameters of the DBO algorithm are iterated; finally, the running results of the DBO algorithm are obtained and the results are substituted into the VMD algorithm. Step 2 is implemented in the following steps: Step 2.1: Extract bearing fault data segments; Step 2.2: Construct the average envelope entropy function for the data segment; Step 2.3: Establish the DBO algorithm model, initialize the parameters, and begin M algorithm iterations; Step 2.4: In one iteration, start a loop with a loop size of N equal to the population size, and execute steps 2.5-2.8; Step 2.5: Update the location information of the dung beetle; Step 2.6: Update the location information of the incubation ball; Step 2.7: Update the location information of adult dung beetles; Step 2.8: Update the location information of the dung beetle thief; Step 2.9: Increment the loop count by 1; Step 2.10: After the current iteration ends, increment the main loop count by 1; Step 2.11: Obtain the global optimal position and fitness function value; In step 2, Model establishment of DBO algorithm: The simulated dung beetle behaviors include the following: (a) Rolling ball Assuming the dung beetle moves along a given direction throughout the search space, using sunlight as its navigation, and assuming that light intensity also affects the dung beetle's path, the formula for calculating the dung beetle's position update during rolling is: (1) In the formula: The updated position of the rolling dung beetle; This represents the current iteration number; Location information for the dung beetle; is a constant representing the deflection coefficient, with a value range of (0, 0.2]. The value range is (0,1); The natural coefficient; This is the worst position globally. For changes in light intensity; In equation (1), the parameter and The appropriate selection is crucial; generally speaking, The value is 0.

1. The value is 0.

3. Simulate the natural factors that cause dung beetles to deviate from their original direction, when A value of 1 indicates that the direction has not deviated. A value of -1 indicates that the direction deviates from the original direction. A higher value indicates a weaker light source. Control the size of the light source to expand the search range; (b) Dancing The new rolling direction is obtained using the tangent function. After obtaining the new direction, the dung beetle will roll the ball in this direction. The position update formula is as shown in equation (2): (2) In the formula: The updated location of the Dancing Dung Beetle; The deflection angle is [0, ...]. ]; In the above formula, The value is 0. / 2、 If the position is not updated, it is assumed that the position has not been updated; if other values ​​are taken, it is assumed that the position of the dung beetle has been updated. (c) Reproduction A boundary selection strategy is proposed, and the concept of the boundary selection strategy is defined as shown in equation (3): (3) In the formula: These are the lower and upper boundaries of the spawning area; This is the current local optimum position; To optimize the lower and upper bounds of the problem; , Indicates the maximum number of iterations; Once the oviposition area is determined, the female dung beetle will choose to lay its eggs in this area. In the DBO algorithm, it is assumed that each female dung beetle lays only one egg in each iteration. According to equation (3), the boundary range of the oviposition area changes dynamically with the value. Therefore, the position of the egg also changes dynamically during the iteration process. Its update formula is as follows: (4) In the formula: The updated location of the oocyte; For the first Position information for the next iteration; They are two independent random vectors; (d) Foraging Based on the foraging process of adult dung beetles, an optimal foraging area model is established, as shown in equation (5): (5) In the formula: These are the lower and upper boundaries of the optimal foraging area; The globally optimal position; Based on the principle of reproductive location update, the following formula for dung beetle foraging location update is obtained: (6) In the formula: The updated location of the foraging dung beetle; These are random numbers that follow a normal distribution. (e) theft Within a dung beetle colony, some dung beetles steal the dung balls of other dung beetles; these are called thieving dung beetles. As can be seen from equation (5), It is the optimal food source, therefore, assuming The area around the dung beetle is the best location for competing for food, and the dung beetle's location information is updated according to the following formula: (7) In the formula: The updated location of the dung beetle; It is a constant; It is a random vector that follows a normal distribution; Selection of fitness function: The envelope entropy of the modes after VMD decomposition is selected as the fitness function of DBO. Its calculation formula is as follows: (8) In the formula: for Normalized representation; The envelope signal after demodulation by Hilbert transform; Envelope entropy; Through iteration of the DBO algorithm, the optimal parameters for 10 types of data VMD decomposition can be obtained. , And the minimum value of the average envelope entropy; Step 3: VMD decomposition and signal reconstruction stage; Step 4: Dataset generation stage; Step 5: Fault Diagnosis Stage.

2. The method for fault diagnosis of wind turbine bearings under variable operating conditions based on DBO-VMD according to claim 1, characterized in that, Step 3 involves first performing VMD decomposition to obtain the corresponding modes and mode spectra; then, fault feature extraction is performed to obtain the reconstructed signal. Step 3 is specifically implemented according to the following steps: Step 3.1: Use the results of the DBO algorithm as the initialization parameters for VMD decomposition; Step 3.2: Run VMD decomposition; Step 3.3: Obtain the time-domain plot and spectrum of each mode; Step 3.4: Calculate the kurtosis of each mode; Step 3.5: Select the mode corresponding to the maximum kurtosis for envelope spectrum analysis; Step 3.6: Select the mode corresponding to the maximum kurtosis as the reconstructed signal; Step 3.7: Fault feature extraction complete.

3. The method for fault diagnosis of wind turbine bearings under variable operating conditions based on DBO-VMD according to claim 2, characterized in that, The VMD decomposition stage in step 3 is as follows: This stage includes using the results calculated by the DBO algorithm to perform VMD decomposition on the input data. The principle of VMD decomposition is: first, the process of establishing the variational model: Signal Decomposed into several modal component signals As shown in equation (9), each mode is demodulated to obtain a single-sided spectrum. (9) In the formula: Unit impact function; By introducing a center frequency The spectrum of each demodulated signal is modulated onto the corresponding baseband, as shown in equation (10): (10) The square norm of the gradient in equation (10) is calculated and the bandwidth corresponding to each modal component is estimated, resulting in the following constrained variational problem model. (11) In the formula: After decomposition One modal component; for The center frequency of each modal component; Next, the variational model addresses the problem. To solve the constraint problem established by equation (11), a quadratic penalty factor is introduced. and Lagrange multiplication operators This transforms the problem into an unconstrained variational problem as follows: (12) In the formula: For Lagrange multiplication operators; To represent the inner product operation; In equation (12), through continuous updating , , The optimal solution to the variational problem can be obtained by finding the values ​​of the three terms. , , The update formulas are respectively equations (13) to (15): (13) (14) (15) In the formula: for Fourier transform; for Fourier transform; for Fourier transform; By following the steps above, the signal's mode, spectrum, and center frequency can be determined.

4. The method for fault diagnosis of wind turbine bearings under variable operating conditions based on DBO-VMD according to claim 3, characterized in that, The signal reconstruction stage in step 3 is as follows: The kurtosis of the modes in the DBO-VMD decomposition is calculated using the following formula: (16) The mode with the maximum kurtosis was selected as the extracted bearing fault feature.

5. The method for fault diagnosis of wind turbine bearings under variable operating conditions based on DBO-VMD according to claim 4, characterized in that, In step 4, the reconstructed signal is first subjected to overlap sampling and CQT transform to obtain a time-frequency diagram. Then, the dataset is divided. Step 4 is implemented in the following steps: Step 4.1: Import the data after fault feature extraction; Step 4.2: Calculate the number of overlapping samples; Step 4.3: Perform CQT transformation on each sample; Step 4.4: Create a bearing dataset with 10 states; Step 4.5: Swap the dataset according to the 8:1:1 ratio; Dataset generation stage: Overlap sampling is used to generate CQT time-frequency maps from the extracted fault features. Assuming the overlap sampling data length is L, and CQT transformation is performed on points of length T=1024, with a step size S=200 for each movement, the number of samples N that can be generated from one data file is determined by the following formula: (17) The generated dataset is divided into training, validation, and test sets in an 8:1:1 ratio.

6. The method for fault diagnosis of wind turbine bearings under variable operating conditions based on DBO-VMD according to claim 5, characterized in that, Step 5 is implemented in the following steps: Step 5.1: Train the RepVGG network using the partitioned dataset and save the model weights; Step 5.2: Load the trained RepVGG model weights and use the test set for prediction; Step 5.3: Obtain the fault diagnosis results; The training and validation sets are input into the RepVGG model. After 60 training iterations, the weights are saved. During testing, the weights are loaded to perform bearing fault diagnosis.

Images