Intelligent diagnosis system for planetary gear set of gear box of wind turbine generator system

Abstract

The present application belongs to the field of wind power generation, and relates to an intelligent diagnosis system for planetary gear set of gear box of wind turbine generator set. The present application proposes an intelligent fault diagnosis system for planetary gear set of gear box of wind turbine generator, which comprises four modules of interference suppression, algorithm optimization, feature enhancement and intelligent diagnosis. Under the action of the system, the vibration signal can remove the interference of working condition change and high frequency vibration of coupled components through the interference suppression module, select the diagnosis position and optimize the structure and parameters of the algorithm through the algorithm optimization module, highlight the fault features in the signal through the feature enhancement module, and diagnose through the intelligent diagnosis module. After the diagnosis is completed, the system will return to the algorithm optimization module to diagnose the remaining parts and output the conclusion. Compared with the traditional gear box fault diagnosis method, the present application overcomes the interference of working condition change and high frequency vibration of coupled components of wind turbine generator, forms a perfect intelligent diagnosis system, and effectively improves the accuracy of fault diagnosis.

Description

Technical Field

[0001] This invention belongs to the field of wind power generation and relates to an intelligent diagnostic system for planetary gear sets in wind turbine generator gearboxes. Background Technology

[0002] Wind turbines are classified into direct-drive, semi-direct-drive, and doubly-fed types. Among them, semi-direct-drive and doubly-fed wind turbines contain a speed-increasing gearbox structure.

[0003] Planetary gear sets are characterized by high load capacity and large transmission ratio. They are commonly used in the gearboxes of wind turbines and are connected to the main shaft. They consist of an internal gear ring, several planetary gears, and a sun gear, and are more complex than parallel gear transmission systems.

[0004] During long-term high-load operation, planetary gear sets often suffer from pitting, indentation and other faults on the tooth surface. After the faults deteriorate, they may cause the gear teeth to break, and in severe cases, the gears may fail, affecting the normal power generation operation of the wind turbine.

[0005] Therefore, fault diagnosis of planetary gear sets is significant in preventing major failures and reducing wind turbine operation and maintenance costs.

[0006] Gear fault diagnosis in wind turbine gearboxes typically employs vibration analysis: sensors are installed at the target location to collect vibration signals, and signal processing methods are used for fault diagnosis.

[0007] The signal analysis method based on Fast Fourier Transform is widely used in the parallel stage gears of wind turbine gearboxes. This method uses Fast Fourier Transform to obtain the spectrum of vibration signals and identifies the characteristic frequencies in the spectrum to diagnose faults.

[0008] In actual vibration analysis, the fault characteristics of the planetary gear set of the wind turbine are not obvious, and the above methods are poor in fault diagnosis, fault location identification and fault level judgment.

[0009] The reasons why the fault characteristics are not obvious are as follows:

[0010] 1. The sensor's installation location is limited, making it impossible to collect high-quality vibration signals;

[0011] 2. The large size of planetary gear sets, the small mass of the planetary gears themselves, and the complex vibration transmission path result in weak vibration signals collected by the sensors;

[0012] 3. Changes in wind turbine operating conditions result in indistinct fault characteristic frequencies in the frequency spectrum;

[0013] 4. Vibration signals are affected by related coupling components, and fault characteristics are overwhelmed by interference signals. Summary of the Invention

[0014] This invention proposes an intelligent diagnostic system for planetary gear sets in wind turbine generator gearboxes, which involves four modules: interference suppression, algorithm optimization, feature enhancement, and intelligent diagnosis.

[0015] The original vibration signal is first passed through the interference suppression module to remove interference from changes in operating conditions and high-frequency vibrations of coupled components. Then, the algorithm optimization module selects the diagnostic location of the intelligent diagnostic module and optimizes the structure and parameters of the feature extraction algorithm. Next, the feature enhancement module highlights the fault features in the signal. Finally, the intelligent diagnostic module diagnoses the location.

[0016] After the diagnosis is completed, the system will return to the algorithm optimization module to re-optimize the structure and parameters of the algorithm, allowing the vibration signal to pass through the fault enhancement module and the intelligent diagnosis module again to diagnose the remaining parts of the planetary gear set. Once all parts have been diagnosed, the diagnosis results will be output.

[0017] The technical solution process of this invention is as follows: Figure 1 As shown, the functions of the four modules involved in this system are as follows:

[0018] The interference suppression module can suppress the influence of operating condition changes in vibration signals, which is beneficial for signal processing and analysis by other modules. Simultaneously, this module can suppress the influence of high-frequency and high-amplitude vibrations in the vibration signal, allowing the signal processing and analysis of other modules to focus on the fault characteristic frequency band, thus facilitating fault feature extraction. The flowchart of this module is as follows: Figure 2 As shown.

[0019] The algorithm optimization module is divided into two parts: algorithm structure optimization and algorithm parameter optimization. Its purpose is to optimize the algorithm structure and parameters in the feature enhancement module.

[0020] In planetary gear sets, the internal ring gear, planetary gears, and sun gear all exhibit different fault characteristics. Therefore, the input parameters in the feature enhancement module also differ. The algorithm optimization module provides different algorithm structures and parameter optimizations for the diagnosis of different parts. The module flow is as follows: Figure 3 As shown.

[0021] The feature enhancement module, by inputting parameters from the algorithm optimization module, can highlight the covered periodic impacts in the vibration signal. These impacts are characteristic of local faults in the planetary gear set. Based on the algorithm output selection in the algorithm optimization module, the target content can be output. The process is as follows: Figure 4 As shown.

[0022] The intelligent diagnosis module uses machine learning methods to train a model on the dataset. Successfully trained models are used for fault diagnosis. For different fault characteristic frequencies, this module repeats the steps of the algorithm optimization module, feature enhancement module, and intelligent diagnosis module, selecting different parameters, and finally outputs the overall diagnostic results for the planetary gear set. The module's process is as follows: Figure 5 As shown. Attached Figure Description

[0023] Figure 1 Invention technical solution flowchart

[0024] Figure 2 Interference Suppression Module Flowchart

[0025] Figure 3 Algorithm optimization module flowchart

[0026] Figure 4 Feature Enhancement Module Flowchart

[0027] Figure 5 Intelligent Diagnostic Module Flowchart

[0028] Figure 6 Parameter optimization flowchart Detailed Implementation

[0029] The processing steps for vibration signals in this invention are as follows:

[0030] Step 1: Call the interference suppression module and use the order analysis method to sample the original signal at equal angles to obtain the amplitude-angle signal;

[0031] The sampling model for the method of suppressing the influence of operating condition changes in vibration signals is as follows:

[0032]

[0033] In the formula

[0034] Indicates the rotation of the first Zhou, after the first Instantaneous rotational speed at each sampling point;

[0035] Indicates the rotation of the first Zhou, the The sampling time of each sampling point;

[0036] Indicates the rotation angle parameter;

[0037] Indicates the rotation of the first Angular acceleration of the circle;

[0038] This indicates the number of sampling points at equal angles per week.

[0039] Based on the isoangular sampling time The original vibration signal is interpolated to obtain the angular domain resampled signal.

[0040] Step 2: Invoke the interference suppression module and select the following model to perform low-pass filtering on the above signal:

[0041]

[0042] In the formula

[0043] Represents frequency The amplitude;

[0044] Indicates the cutoff frequency;

[0045] Indicates the meshing frequency band parameters;

[0046] Indicates characteristic frequency band parameters;

[0047] Indicates order;

[0048] Indicates the filter order;

[0049] Represents the number of teeth. These represent the number of teeth on the internal gear ring, planet gears, and sun gear, respectively.

[0050] Step 3: Call the algorithm optimization module to optimize the algorithm structure.

[0051] The algorithm structure is affected by the number of iterations, the shift factor, and the algorithm output. Its optimization strategy is as follows:

[0052] The number of iterations and the displacement factor in this invention are determined experimentally based on a large amount of actual vibration data, wherein:

[0053] Number of iterations The experiment determined that the number of iterations was 60, which affects the algorithm's performance and computation time.

[0054] Displacement factor With periodic parameters The number of shifts that jointly influence convolution is determined experimentally. The value is 1.

[0055] The algorithm output can be selected to include the enhanced feature signal, envelope spectrum, optimal filter, and fault feature index change curve. These four items are output during the diagnostic process, while the first two items are output during the training of the intelligent diagnostic model.

[0056] Step 4: Call the algorithm optimization module to calculate the periodic parameters The calculation method is as follows:

[0057]

[0058] In the formula

[0059] Represents the sampling rate;

[0060] The frequency doubling factor represents the fault characteristics;

[0061] Represents order;

[0062] Represents the number of teeth. These represent the number of teeth on the internal gear ring, planet gears, and sun gear, respectively.

[0063] The optimal filter parameters are calculated using a one-dimensional particle swarm optimization algorithm, and the filter is initialized. The optimization process is as follows: Figure 6 As shown.

[0064] Assume a certain space exists A particle swarm consisting of n particles, where the nth particle is... The position of each particle is Its speed is Its local optimal solution is The global optimal solution of the particle swarm is .

[0065] During the iteration, each particle updates its position and velocity according to the following formula.

[0066]

[0067] In the formula

[0068] Indicates the iteration number;

[0069] Indicates the first learning factor;

[0070] Indicates the second learning factor;

[0071] Indicates inertia weight;

[0072] It is a random number in [0,1].

[0073] This invention introduces envelope entropy as the objective function of the particle swarm optimization algorithm, and the processed signal Envelope entropy The calculation method is as follows:

[0074]

[0075] In the formula

[0076] Indicates signal The signal after envelope demodulation.

[0077] When periodic shocks occur, Smaller.

[0078] When the iteration ends, the global optimal solution and particle positions are output, and the rounded values ​​are used as the optimal filter parameters.

[0079] Step 5: Invoke the feature enhancement module and input the parameters into it. The model that highlights the fault features is as follows:

[0080]

[0081] In the formula

[0082] Represents the original vibration signal;

[0083] This represents the filtered signal;

[0084] Represents the filter coefficients;

[0085] This represents the number of iterations, with an upper limit of 1. ;

[0086] Indicates the first Replace the original signal weighting coefficients;

[0087] Indicates the first Alternate filtering signal weighting coefficients;

[0088] Indicates the periodic parameter;

[0089] Indicates the displacement factor;

[0090] Indicates filter parameters;

[0091] Indicates the number of sampling points;

[0092] , , This indicates that an intermediate value is being calculated.

[0093] Based on the above signal processing results, the fault diagnosis strategy of the present invention is as follows:

[0094] Step 1: A dataset is established using a large amount of measured vibration data from faulty gears and normal gears, and corresponding labels are added. The vibration data in the dataset is then processed through an interference suppression module, an algorithm optimization module, and a feature enhancement module, and then trained using a WDCNN network to obtain an intelligent diagnostic model. The specific steps for model training are as follows:

[0095] Step 1.1: Obtain the data from a wind turbine with a faulty planetary gear set in its gearbox. The vibration data points were marked as fault sets, obtained from the unit after the same model gearbox was replaced. The vibration data points are labeled as a normal set, and the two together are called the dataset;

[0096] Step 1.2: Select 70% of the data from the fault set and the normal set respectively as training data, and use the remaining 30% as validation data;

[0097] Step 1.3: After training the training data through the WDCNN neural network, it can correspond to two types of outputs: normal and faulty, thus completing the model training;

[0098] Step 1.4: Verify the data. After the model diagnosis is performed, compare the diagnosis results with its original labels. If the accuracy is above 99%, the training model can be output; otherwise, the training will fail.

[0099] Step 2: Call the intelligent diagnostic module to process the new vibration signal using the above signal processing method and use it as input to the diagnostic model;

[0100] Step 3: After the output results are evaluated by the algorithm, if the diagnosis of all parts is not completed, return to the parameter optimization module, select new period parameters and filter parameters, and perform feature enhancement and fault diagnosis again.

[0101] Step 4: Organize the above diagnostic results and output the status of the internal gear ring, planet gears, and sun gear of the planetary gear set.

Claims

1. An intelligent diagnostic system for planetary gear sets in wind turbine generator gearboxes, characterized in that: The original vibration signal first passes through the interference suppression module to remove interference from changes in operating conditions and high-frequency vibrations of coupled components. Then, the algorithm optimization module selects the diagnostic location of the intelligent diagnostic module and optimizes the structure and parameters of the feature extraction algorithm. Next, the feature enhancement module highlights the fault features in the signal. Finally, the intelligent diagnostic module diagnoses the location. After the location is diagnosed, the system returns to the algorithm optimization module to select other diagnostic locations and re-optimize the algorithm structure and parameters. The vibration signal then passes through the fault enhancement module and the intelligent diagnostic module again to diagnose the remaining parts of the planetary gear set. Once all parts are diagnosed, the diagnostic results are output. The algorithm optimization module of the system uses the following model for optimization: In the formula Indicates the iteration number; Indicates the first learning factor; Indicates the second learning factor; Indicates inertia weight; A random number in the range [0,1]; Let represent the envelope entropy, which is the objective function for parameter optimization. Indicates signal The signal after envelope demodulation; Indicates the first The particle in the first The position of the generation; Indicates the first The particle in the first The speed of generation; Indicates a local optimal solution; Indicates the globally optimal solution; The feature enhancement module of the system uses the following model to perform feature enhancement calculations: In the formula Represents the original vibration signal; This represents the filtered signal; Represents the filter coefficients; This represents the number of iterations, with an upper limit of 1. ; Indicates the first Replace the original signal weighting coefficients; Indicates the first Alternate filtering signal weighting coefficients; Indicates the periodic parameter; Indicates the displacement factor; Indicates filter parameters; Indicates the number of sampling points; , , This indicates that an intermediate value is being calculated.

2. The intelligent diagnostic system for planetary gear sets in a wind turbine generator gearbox according to claim 1, characterized in that: The method for suppressing high-frequency vibration interference of related coupled components in the interference suppression module is modeled as follows: In the formula: Represents frequency The amplitude; Indicates the cutoff frequency; Indicates the meshing frequency band parameters; Indicates characteristic frequency band parameters; Indicates order; Indicates the filter order; Indicates the number of teeth. These represent the number of teeth on the internal gear ring, planet gears, and sun gear, respectively.

3. The intelligent diagnostic system for planetary gear sets in a wind turbine generator gearbox according to claim 1, characterized by its algorithm. The optimal number of iterations and displacement factor in the optimization module were obtained through training experiments using a large amount of measured data with an intelligent diagnostic model. The optimization steps are as follows: Step 1: Fix the value of the displacement factor, select different iteration numbers with a step size of 10, and use the vibration data of the faulty planetary gear set and related parameters as input to train the intelligent diagnostic model. Step 2: For the intelligent diagnostic models trained with different number of iterations, first select the models that pass the diagnostic accuracy test; Step 3: For the models that pass the diagnostic accuracy test, select the model with the shortest training time and use its number of iterations as the optimal number of iterations; Step 4: Fix the optimal number of iterations, select the displacement factor with a step size of 1, and select the optimal displacement factor based on the principle of the above steps.

4. The intelligent diagnostic system for planetary gear sets in a wind turbine generator gearbox according to claim 1, characterized by its algorithm. The calculation method for the periodic parameter in the optimization module is as follows: In the formula Represents the sampling rate; The frequency doubling factor represents the fault characteristics; Represents order; Indicates the number of teeth. These represent the number of teeth on the internal gear ring, planet gears, and sun gear, respectively.

5. The intelligent diagnostic system for planetary gear sets in a wind turbine generator gearbox according to claim 1, characterized in that the intelligent diagnostic model is trained using a large amount of measured vibration data and based on a WDCNN network, and the model training steps are as follows: Step 1: Obtain from a wind turbine with a faulty planetary gear set in its gearbox. The vibration data points were marked as fault sets, obtained from the unit after the same model gearbox was replaced. The vibration data points are labeled as a normal set, and the two together are called the dataset; Step 2: Select 70% of the data from the fault set and the normal set as training data, and the remaining 30% as validation data; Step 3: After training the training data through the WDCNN neural network, it can correspond to two types of outputs: normal and faulty, thus completing the model training; Step 4: Verify the data. After the model diagnosis is performed, compare the diagnosis results with its original labels. If the accuracy is above 99%, the training model can be output; otherwise, the training will fail.

6. The intelligent diagnostic system for planetary gear sets in a wind turbine generator gearbox according to claim 1, characterized in that: After the vibration signal is judged by the algorithm effect of the intelligent diagnostic model, it can be returned to the algorithm optimization module to re-optimize the algorithm structure. When optimizing the parameters, the parameters of other parts of the planetary gear set are recalculated. Then, the diagnosis is completed by the feature enhancement module and the intelligent diagnostic module, and finally the diagnostic conclusion is output.

Images