Fan gear box vibration fault monitoring and diagnosing system
Through deep learning models and generation of diverse vibration signal samples (GANs), the problem of difficult to extract early fault features in fan gearbox fault diagnosis is solved, and higher fault identification accuracy and data diversity are achieved.
Patent Information
- Application Number
- CN202411704075.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-26
- Publication Date
- 2025-03-14
AI Technical Summary
In the fault diagnosis of fan gearboxes, the vibration signal energy in the early fault state is small and easily covered by noise. It is difficult for the existing technology to effectively extract early fault characteristics, resulting in difficulty in diagnosis and scarce failure data samples.
Deep learning model combined with data augmentation technology is used to generate diverse vibration signal samples by generating adversarial networks (GANs), enhance the training set of the model to make up for the insufficient fault data samples.
It improves the fault identification ability of the model under complex operating conditions, enhances data diversity and model adaptability, and improves the accuracy and efficiency of fault diagnosis.
Smart Images
Figure CN119622411A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of mechanical equipment fault diagnosis, and in particular to a fan gearbox vibration fault monitoring and diagnosis system, which is mainly used for monitoring and analyzing the vibration signal of the fan gearbox to detect and diagnose possible faults. Background Art
[0002] The main problems that exist at present are: the gearbox is an important component of the wind turbine, and its main function is to accelerate the rotation of the rotor shaft to drive the generator to generate electricity. Since the impeller speed is low, far lower than the speed required for the generator to generate electricity, the gearbox is needed to increase the speed. At the same time, due to the use conditions and environmental restrictions, the gearbox failure of the wind turbine speed increaser is frequent; the maintenance of large components of the wind turbine generator set requires a lot of manpower and material resources, and the mechanical structure failure requires a longer maintenance time, resulting in reduced power generation efficiency and increased maintenance costs. Therefore, fault monitoring and diagnosis of equipment plays a key role in ensuring the stable operation of the whole machine.
[0003] In gearbox fault diagnosis, obtaining sufficient data samples and fault status samples is the key to accurate diagnosis. However, in actual operation, due to the limitations of field conditions, especially the complex working conditions and high maintenance costs of large equipment such as wind turbine gearboxes, sample acquisition usually faces many challenges.
[0004] In the current existing technology,
[0005] 1. Sensor layout requires precise point selection, but due to complex working conditions, the signal often contains a lot of noise and interference. In addition, the monitoring range is limited, and it is difficult to fully cover all potential fault points, especially when the early fault characteristics are weak, the signal-to-noise ratio is low, causing the fault characteristics to be easily masked by noise.
[0006] 2. It takes a long time to accumulate enough data samples, and when fault samples are scarce, it is impossible to ensure that there is sufficient data for each fault type. In addition, early fault signals are usually hidden in normal data, and the signal performance under each working condition is different, resulting in poor generalization of the data.
[0007] In summary, the vibration signal energy of the early fault state is very small and can be easily submerged by noise. Conventional signal processing methods cannot effectively extract early fault features, resulting in diagnosis difficulties. Fault state samples are difficult to obtain in large quantities in a short period of time, especially in the actual working conditions of large equipment such as fan gearboxes. The frequency of faults is low, resulting in a scarcity of fault data samples. Summary of the invention
[0008] The purpose of the present invention is to generate more representative and diverse samples and improve the fault identification ability of the model under complex working conditions. A fan gearbox vibration fault monitoring and diagnosis system is proposed.
[0009] A fan gearbox vibration fault monitoring and diagnosis system, comprising a vibration data acquisition module, a signal preprocessing and noise reduction module, a fault feature extraction module and a fault diagnosis and prediction module; characterized in that:
[0010] The fault diagnosis and prediction module uses a deep learning model to classify and identify the extracted fault features. The deep learning model also combines data enhancement technology to generate training samples and uses generative adversarial networks (GANs) to generate diverse vibration signal samples to enhance the model's training set and make up for the lack of fault data samples.
[0011] The beneficial effects of the present invention are:
[0012] Compared with the existing technology, the system can achieve fine-tuning of transfer learning based on the synthetic samples generated by GANs. During the initial migration, the deep learning models (such as CNN, LSTM) that have been pre-trained on similar working conditions or equipment are migrated to the current wind turbine gearbox fault diagnosis task. By introducing high-quality samples generated by GANs, the data diversity of the model is enhanced, and it gradually adapts to the new task environment, thereby improving the overall performance. BRIEF DESCRIPTION OF THE DRAWINGS
[0013] Figure 1 This is the overall system architecture diagram;
[0014] Figure 2 This is a detailed diagram of the vibration data acquisition module;
[0015] Figure 3 This is the flow chart of the signal preprocessing and noise reduction module;
[0016] Figure 4 This is the working principle diagram of the fault feature extraction module;
[0017] Figure 5 This is a diagram showing the relationship between deep learning models and GANs;
[0018] Figure 6 Flowchart for iterative optimization of transfer learning loop. DETAILED DESCRIPTION
[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0020] like Figure 1As shown, a wind turbine gearbox vibration fault monitoring and diagnosis system includes a vibration data acquisition module, a signal preprocessing and noise reduction module, a fault feature extraction module and a fault diagnosis and prediction module;
[0021] like Figure 2 As shown in the figure, in the vibration data acquisition module, various types of sensors, such as acceleration sensors and speed sensors, are installed at key parts of the gearbox to collect the vibration data of the gearbox in real time. The collected data include vibration signals, rotation speed, temperature, etc. The sensor network can perform multi-point synchronous monitoring to ensure a comprehensive perception of the overall operating status of the equipment. It is equipped with high-sensitivity sensors and performs multi-point data synchronous collection for vibration signals of different frequency bands to ensure the integrity and accuracy of the signal.
[0022] like Figure 3 As shown in the figure, in the signal preprocessing and noise reduction module, the collected original vibration signal is preprocessed to remove noise and interference. The vibration signal is decomposed into several intrinsic mode functions (IMFs) by using empirical mode decomposition (EMD) and adaptive noise separation technology. By screening effective IMF components, noise is removed and fault-related characteristic signals are retained. EMD has strong adaptability and can process complex nonlinear and non-stationary signals. At the same time, the adaptive noise separation technology can significantly improve the signal-to-noise ratio and ensure the accuracy of subsequent analysis.
[0023] like Figure 4 As shown in the figure, there are a large number of vibration features in the gearbox vibration data, such as frequency, amplitude, phase, etc. These features can reflect the operating status and existing faults of the gearbox, and the vibration data needs to be feature extracted. The fault feature extraction module performs multi-resolution analysis on the denoised vibration signal through wavelet packet decomposition (WPD), extracts key time domain and frequency domain features, and extracts wavelet packet energy features. The above extracted features are used to obtain the energy proportion of each frequency band iterated to the third layer, thereby separating the signal components of different frequency bands and characterizing the local information of the signal in the time domain and frequency domain at the same time. The wavelet packet decomposition is used to refine the signal analysis capabilities in the time domain and frequency domain. Especially when processing early fault signals, the wavelet packet decomposition can effectively capture the weak changes in the signal and improve the accuracy of feature extraction.
[0024] like Figure 5As shown, as a preferred embodiment, in the fault diagnosis and prediction module, a deep learning model is used to classify and identify the extracted fault features, and data enhancement technology is combined to make up for the problem of insufficient field data samples. The model automatically completes fault classification and diagnosis by learning the characteristic patterns under different working conditions. The system can also combine historical data to predict the development trend of gearbox faults, increase the diversity of samples through data enhancement technology, and improve the accuracy of fault classification by the deep learning model. At the same time, the self-learning and adaptive capabilities of the deep learning model are used to optimize the diagnostic algorithm in real time, thereby improving the fault detection effect of the system under complex working conditions.
[0025] Generative adversarial networks (GANs) are used to generate diverse vibration signal samples in fault diagnosis to enhance the model's training set and improve the accuracy of fault classification. GANs consist of a generator and a discriminator. The generator attempts to generate realistic samples, while the discriminator is responsible for distinguishing between generated samples and real samples. GANs generates realistic synthetic samples to make up for the problem of insufficient field data samples. In the combination of data enhancement and transfer learning, GANs are used to generate synthetic samples to make up for the lack of actual data, and transfer learning is combined to fine-tune the model. GANs are used to enhance data diversity, improve the adaptability of transfer learning, and perform efficient migration between different working conditions or equipment to ensure model accuracy in the case of very little data.
[0026] Furthermore, in GANs, there are two networks, the generator and the discriminator. The generator is responsible for generating realistic fault samples, while the discriminator is responsible for determining whether these samples are real or generated by the generator. The two networks are continuously iterated and optimized through game theory. The goal of the generator is to generate samples that are as realistic as possible so that the discriminator cannot distinguish between true and false. The goal of the discriminator is to improve its ability to distinguish between real data and generated data. Through multiple rounds of iteration, the generator continuously optimizes the quality of sample generation, so that the generated data is close to the distribution of real fault samples, thereby improving data diversity.
[0027] The samples generated by the loop iteration are evaluated and screened to select high-quality synthetic data:
[0028] Error evaluation of the discriminator: In each round of GAN iteration, the generator continuously adjusts its parameters based on the feedback from the discriminator. After multiple rounds of training, the generated samples become more realistic and the errors decrease. The samples with lower errors are retained and used as training data for subsequent transfer learning.
[0029] Sample diversity evaluation: In order to ensure the diversity of generated samples, some diversity measurement methods are used, such as using the distribution difference in the feature space to evaluate the similarity between samples, and screen out samples with high diversity but high quality for subsequent training.
[0030] like Figure 6 As shown, the transfer learning model is trained through loop iteration. After generating enough realistic samples, the transfer learning model is further fine-tuned through loop iteration:
[0031] 1) Initial migration: Migrate the deep learning model (such as CNN, LSTM) that has been pre-trained on similar working conditions or equipment to the current wind turbine gearbox fault diagnosis task. At this time, due to insufficient data, the performance of the model may be poor.
[0032] 2) Synthetic sample enhancement training: Add high-quality and realistic samples generated by GANs to the training set to enhance the data diversity of the migration model. At this time, the model can gradually adapt to the new task environment by learning more diverse sample distributions.
[0033] 3) Iterative optimization: After each round of training, the feedback of the transfer learning model is used to evaluate the performance of the current model in the fault classification task, and the generation strategy of GANs is gradually adjusted. The generator continues to generate more complex or more representative fault samples, so that the transfer learning model gradually optimizes its performance.
[0034] 4) Close the loop: GANs generate new realistic samples → transfer learning models are trained using synthetic data → feedback from the transfer learning model guides GANs to optimize the generation strategy → iterate the cycle to improve the adaptability of the final model and the accuracy of fault diagnosis.
[0035] In the technical solution of this application,
[0036] 1. In the fault diagnosis process, GANs are used to generate diverse vibration signal samples to enhance the model's training set. The generator continuously tries to generate realistic samples, while the discriminator is responsible for distinguishing between generated samples and real samples. The two are continuously optimized through game theory. This process effectively compensates for the problem of insufficient field data samples and ensures that the model can still maintain high accuracy when facing different fault types. Based on the synthetic samples generated by GANs, the system can achieve fine-tuning of transfer learning. During the initial migration, the deep learning models (such as CNN and LSTM) that have been pre-trained on similar working conditions or equipment are migrated to the current wind turbine gearbox fault diagnosis task. By introducing high-quality samples generated by GANs, the data diversity of the model is enhanced, and it gradually adapts to the new task environment, thereby improving the overall performance.
[0037] 2. After each round of training, the feedback of the transfer learning model is used to evaluate the performance of the current model in the fault classification task. The system will continuously adjust the generation strategy of GANs to generate more complex or more representative fault samples. Through cyclic iteration, the generator continuously optimizes the quality of sample generation, making the generated data closer to the distribution of real fault samples, thereby improving data diversity and model adaptability.
[0038] 3. Generate new samples through GANs, train the transfer learning model with synthetic data, and use feedback to guide GANs optimization strategy and iterate in cycles to improve the adaptability and accuracy of the final model. The introduction of synthetic data significantly expands the diversity and scale of the training set, making the model more robust when facing different fault modes. Secondly, the high-quality samples generated by GANs can make up for the problem of insufficient field data, especially in the scenario of few samples, the model training is more sufficient, thereby improving the accuracy of fault detection and diagnosis. In addition, transfer learning can quickly adapt to new working conditions or equipment environments, reduce model training time and computing costs, and improve the model's ability to handle new problems. The performance of the model can be continuously evaluated and adjusted to ensure that the generator and discriminator are developing in the optimal direction in each round of iteration, thereby improving the model's fault identification ability under complex working conditions. In the process of iterative cycles, the generator gradually learns to generate more representative and diverse samples, further improving the generalization ability of the model and ensuring stable performance under various operating conditions. Closed-loop optimization not only enhances the intelligence of the fault diagnosis system, but also provides more accurate and efficient technical support for real-time monitoring and early warning.
[0039] Finally, it should be noted that the above description is only an explanation of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail, those skilled in the art can still modify the technical solutions described above or replace some of the technical features with equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A wind turbine gearbox vibration fault monitoring and diagnosis system, comprising a vibration data acquisition module, a signal preprocessing and noise reduction module, a fault feature extraction module and a fault diagnosis and prediction module; characterized in that: The fault diagnosis and prediction module uses a deep learning model to classify and identify the extracted fault features. The deep learning model also combines data enhancement technology to generate training samples and uses generative adversarial networks (GANs) to generate diverse vibration signal samples to enhance the model's training set and make up for the lack of fault data samples.
2. According to claim 1, a fan gearbox vibration fault monitoring and diagnosis system is characterized in that: The vibration data acquisition module realizes real-time vibration data acquisition by installing various types of sensors at key locations of the wind turbine gearbox; the sensor types include acceleration sensors, velocity sensors, and temperature sensors, which can monitor vibration signals, rotation speed, and temperature.
3. According to claim 1, a fan gearbox vibration fault monitoring and diagnosis system is characterized in that: The signal preprocessing and noise reduction module preprocesses the collected original vibration signal, uses empirical mode decomposition (EMD) and adaptive noise separation technology to remove noise and interference and retain fault-related characteristic signals.
4. According to claim 1, a fan gearbox vibration fault monitoring and diagnosis system is characterized in that: The fault feature extraction module performs multi-resolution analysis on the denoised vibration signal through wavelet packet decomposition (WPD) to extract key features in the time domain and frequency domain. In the multi-resolution analysis, the signal components in different frequency bands are separated through wavelet packet decomposition, and the local information of the signal is characterized at the same time.
5. A fan gearbox vibration fault monitoring and diagnosis system according to claim 1, characterized in that: The fault diagnosis and prediction module automatically completes fault classification and diagnosis by learning the characteristic patterns under different working conditions; the fault development trend prediction predicts the fault development trend of the gearbox by combining historical data and optimizes the diagnosis algorithm.
6. A fan gearbox vibration fault monitoring and diagnosis system according to claim 1, characterized in that: When the fault diagnosis and prediction module is combined with data enhancement, data enhancement and transfer learning are combined; Generate diverse vibration signal samples using generative adversarial networks (GANs) to enhance the model’s training set and improve the robustness and accuracy of fault classification; Using the generator and the discriminator, the generator is responsible for generating realistic fault samples, and the discriminator is responsible for distinguishing whether the samples are real or not; Iterative optimization: through game iteration, the quality of generated samples is continuously optimized and data diversity is improved.
7. A fan gearbox vibration fault monitoring and diagnosis system according to claim 6, characterized in that: During the above-mentioned iterative optimization cycle, the generated samples are evaluated and screened, and high-quality synthetic data are selected for transfer learning training. The discriminator error is evaluated, and the generator parameters are adjusted according to the discriminator feedback. Samples with lower errors are screened for training. The diversity measurement method is used to ensure the diversity of generated samples and screen out high-quality samples.
8. A fan gearbox vibration fault monitoring and diagnosis system according to claim 7, characterized in that: The above transfer learning is combined with the generated realistic samples to fine-tune the transfer learning and improve the performance of the model on the task; Initial migration: Migrate the pre-trained deep learning model to the current task; Synthetic sample enhancement training: samples generated by GANs are added to the training set to increase data diversity; Iterative optimization: Based on the feedback from the transfer learning model, the generation strategy of GANs is continuously adjusted to optimize model performance.
9. A fan gearbox vibration fault monitoring and diagnosis system according to claim 1, characterized in that: The fault diagnosis and prediction module uses generative adversarial networks (GANs) to generate diverse vibration signal samples using a closed-loop optimization process; the closed-loop process: 1) GANs generate new samples, 2) the transfer learning model is trained using synthetic data, 3) feedback guides the GANs optimization strategy, and 4) the cycle iterates to improve the adaptability and accuracy of the model.