CN-122014530-A - Fan bearing detection method and device for wind power generation equipment

CN122014530ACN 122014530 ACN122014530 ACN 122014530ACN-122014530-A

Abstract

The application discloses a fan bearing detection method and device of wind power generation equipment, which can realize continuous adaptation and stable diagnosis capability improvement of a fan bearing diagnosis model under variable working conditions through physical prompt-guided pre-training and incremental update based on new working condition data.

Inventors

WU SHIDONG
ZHU XIAOYI
HAN JUNFEI
GUO SHUAINAN
WANG QIAN
SU YIBO

Assignees

中国长江三峡集团有限公司

Dates

Publication Date: 20260512
Application Date: 20260210

Claims (10)

1. The method for detecting the fan bearing of the wind power generation equipment is characterized by comprising the following steps of: obtaining equivalent mass, damping coefficient and equivalent rigidity of the fan bearing along with time change; Reading structural parameters of the fan bearing, and calculating fault theoretical characteristic frequency of a target position of the fan bearing under a target working condition based on the structural parameters; Splicing the fault theoretical characteristic frequency, the equivalent mass, the damping coefficient and the equivalent stiffness to generate a physical prompt vector; collecting original vibration signals under different working conditions; Determining a target loss function based on the physical prompt vector and the original vibration signal, and pre-training a physical information diffusion model with cross-working condition basic generation and diagnosis capability based on the target loss function; receiving a new working condition vibration signal in real time, and determining the new working condition vibration signal as online incremental data; Performing incremental training on the physical information diffusion model based on the online incremental data to generate optimal incremental model parameters, and generating an incremental diagnostic model based on the optimal incremental model parameters; And when receiving the current vibration signal of the target fan bearing, controlling the incremental diagnosis model to generate a detection result aiming at the target fan bearing.
2. The method of claim 1, further comprising, prior to the step of determining a target loss function based on the physical cue vector and the raw vibration signal: And performing wavelet threshold denoising operation and normalization processing operation on the original vibration signal to generate a preprocessed original vibration signal.
3. The method of claim 2, wherein the step of determining a target loss function based on the physical cue vector and the raw vibration signal comprises: Constructing a deep learning framework comprising a feature extraction module formed by a convolutional neural network, a diffusion back diffusion module formed based on a diffusion probability model framework and a classification module; inputting the preprocessed original vibration signal into the feature extraction module to generate a high-dimensional feature vector; Fusing the high-dimensional feature vector and the physical prompt vector, and executing forward diffusion noise adding and reverse diffusion denoising operations through the diffusion and reverse diffusion module to obtain a reconstructed original feature vector; Establishing a physical correlation model between the fault impact load and the final vibration displacement based on a vibration response physical equation by adopting the equivalent mass, the damping coefficient and the equivalent rigidity; substituting the original feature vector into the physical association model, calculating an error between the original feature vector and the physical prompt vector, and generating physical constraint loss; calculating the mean square error between the original feature vector and the high-dimensional feature vector to obtain the inverse diffusion reconstruction loss; Determining the prediction probability of the fault type of the original feature vector through a classification module, and marking a real label corresponding to the original vibration signal adopting the original feature vector; Inputting the original feature vector into the classification module, and calculating cross entropy loss between the prediction probability and the real label to obtain classification loss; And constructing a total loss function by combining the physical constraint loss, the inverse diffusion reconstruction loss and the classification loss.
4. A method according to claim 3, wherein the step of pre-training a physical information diffusion model with cross-regime basis generation and diagnostic capabilities based on the objective loss function comprises: Performing a pre-training operation on the deep learning frame by adopting the total loss function to obtain pre-training model parameters; And solidifying the pre-training model parameters to generate a pre-training physical information diffusion model.
5. The method of claim 4, wherein the step of performing incremental training on the physical information diffusion model based on the online incremental data to generate optimal incremental model parameters comprises: Performing wavelet threshold denoising operation and normalization processing operation on the online incremental data to generate a preprocessed online incremental training set and an online verification set; Freezing bottom layer parameters of a feature extraction module in the pre-training physical information diffusion model, and adjusting top layer parameters of the feature extraction module, the diffusion back-diffusion module and the classification module to obtain a fine tuning parameter set; fusing the online incremental training set and the physical prompt vector, executing incremental learning operation through the fine tuning parameter set, and calculating back diffusion reconstruction loss, frequency deviation loss, dynamic deviation loss and classification loss of the online incremental data; constructing an incremental loss function in combination with the inverse diffusion reconstruction loss, the frequency deviation loss, the kinetic deviation loss, and the classification loss of the incremental data; inputting random noise based on a reverse diffusion process of the pre-training physical information diffusion model, and generating a virtual history sample consistent with the history fault type characteristics of the original vibration signal; Mixing the virtual history sample with the online incremental training set according to a preset proportion to form a mixed training set; Performing incremental training operation on the mixed training set by adopting the incremental loss function to obtain incremental updating model parameters; Calculating classification precision and loss value of the incremental update model parameters by combining the online verification set; And stopping executing the incremental training operation on the mixed training set to generate optimal incremental model parameters when the classification precision and the loss value are judged to meet preset conditions.
6. The method of claim 5, wherein the step of controlling the incremental diagnostic model to generate a test result for the target fan bearing comprises: Performing wavelet threshold denoising operation and normalization processing operation on the current vibration signal to generate a preprocessed current vibration signal; inputting the preprocessed current vibration signal into a feature extraction module of the incremental diagnostic model to generate a current high-dimensional feature vector; fusing the current high-dimensional feature vector and the physical prompt vector, and executing reconstruction operation through a diffusion inverse diffusion module of the incremental diagnostic model to obtain a reconstructed current feature vector; Inputting the reconstructed current feature vector into a classification module of the incremental diagnostic model, and outputting a fault type prediction result of the target fan bearing and a confidence level aiming at the fault type prediction result; if the confidence coefficient is lower than a preset threshold value, performing secondary optimization on the reconstructed current feature vector through a back diffusion process of the incremental diagnostic model to obtain a refined feature vector, and calculating physical constraint errors of the refined feature vector and the physical association model; And when the physical constraint error is smaller than a preset threshold value, generating a corrected fault type prediction result by adopting the refined feature vector.
7. The method as recited in claim 5, further comprising: And when the physical constraint error is not smaller than a preset threshold value, generating an early warning signal, and storing the fault type prediction result in association with real-time working condition data to generate a fault working condition association database.
8. The utility model provides a fan bearing detection device of wind power generation equipment which characterized in that includes: the bearing rotating part parameter acquisition module is used for acquiring equivalent mass, damping coefficient and equivalent rigidity changing along with time of the fan bearing; the fault theoretical characteristic frequency calculation module is used for reading structural parameters of the fan bearing and calculating the fault theoretical characteristic frequency of the target position of the fan bearing under the target working condition based on the structural parameters; The physical prompt vector generation module is used for splicing the fault theoretical characteristic frequency, the equivalent mass, the damping coefficient and the equivalent stiffness to generate a physical prompt vector; the original vibration signal acquisition module is used for acquiring original vibration signals under different working conditions; the physical information diffusion model training module is used for determining a target loss function based on the physical prompt vector and the original vibration signal, and pre-training a physical information diffusion model with cross-working condition basic generation and diagnosis capability based on the target loss function; The online incremental data acquisition module is used for receiving the new working condition vibration signal in real time and determining the new working condition vibration signal as online incremental data; the incremental diagnostic model generation module is used for executing incremental training on the physical information diffusion model based on the online incremental data, generating optimal incremental model parameters and generating an incremental diagnostic model based on the optimal incremental model parameters; And the detection result generation module is used for controlling the incremental diagnosis model to generate a detection result aiming at the target fan bearing when receiving the current vibration signal of the target fan bearing.
9. An electronic device comprising a processor, a memory and a program or instruction stored on the memory and executable on the processor, which program or instruction when executed by the processor implements the method of claims 1-7.
10. A readable storage medium, characterized in that it stores thereon a program or instructions, which when executed by a processor, implements the method according to claims 1-7.

Description

Fan bearing detection method and device for wind power generation equipment Technical Field The invention relates to the technical field of wind turbine bearing detection of wind power generation equipment, in particular to a wind turbine bearing detection method of wind power generation equipment, a wind turbine bearing detection device of wind power generation equipment, electronic equipment and a readable storage medium. Background The wind generating set is used as an important carrier of clean energy, and a bearing assembly of the wind generating set plays a key role in the reliability and the power generation efficiency of the whole machine. The fan bearing is in a complex and changeable running environment for a long time and is continuously influenced by factors such as wind speed, load, temperature, humidity and the like, so that a bearing vibration signal presents obvious multi-working condition characteristics including non-stationarity, distribution drift and multi-mode characteristics. The traditional fan bearing fault diagnosis methods are mainly based on modeling training of data under a laboratory fixed working condition or a limited stable loading condition, and are difficult to effectively adapt to data distribution change caused by frequent working condition switching in actual wind field application, so that the generalization performance of a diagnosis model is obviously reduced, and the fault recognition accuracy is reduced. Disclosure of Invention Embodiments of the present invention provide a method, an apparatus, an electronic device, and a readable storage medium for detecting a fan bearing of a wind power generation device, so as to overcome or at least partially solve the above-mentioned problems. In order to solve the technical problems, the application is realized as follows: in a first aspect, an embodiment of the present application provides a method for detecting a fan bearing of a wind power generation device, including: obtaining equivalent mass, damping coefficient and equivalent rigidity of the fan bearing along with time change; Reading structural parameters of the fan bearing, and calculating fault theoretical characteristic frequency of a target position of the fan bearing under a target working condition based on the structural parameters; Splicing the fault theoretical characteristic frequency, the equivalent mass, the damping coefficient and the equivalent stiffness to generate a physical prompt vector; collecting original vibration signals under different working conditions; Determining a target loss function based on the physical prompt vector and the original vibration signal, and pre-training a physical information diffusion model with cross-working condition basic generation and diagnosis capability based on the target loss function; receiving a new working condition vibration signal in real time, and determining the new working condition vibration signal as online incremental data; Performing incremental training on the physical information diffusion model based on the online incremental data to generate optimal incremental model parameters, and generating an incremental diagnostic model based on the optimal incremental model parameters; And when receiving the current vibration signal of the target fan bearing, controlling the incremental diagnosis model to generate a detection result aiming at the target fan bearing. Optionally, before the step of determining a target loss function based on the physical cue vector and the raw vibration signal, the method further comprises: And performing wavelet threshold denoising operation and normalization processing operation on the original vibration signal to generate a preprocessed original vibration signal. Optionally, the step of determining a target loss function based on the physical cue vector and the raw vibration signal comprises: Constructing a deep learning framework comprising a feature extraction module formed by a convolutional neural network, a diffusion back diffusion module formed based on a diffusion probability model framework and a classification module; inputting the preprocessed original vibration signal into the feature extraction module to generate a high-dimensional feature vector; Fusing the high-dimensional feature vector and the physical prompt vector, and executing forward diffusion noise adding and reverse diffusion denoising operations through the diffusion and reverse diffusion module to obtain a reconstructed original feature vector; Establishing a physical correlation model between the fault impact load and the final vibration displacement based on a vibration response physical equation by adopting the equivalent mass, the damping coefficient and the equivalent rigidity; substituting the original feature vector into the physical association model, calculating an error between the original feature vector and the physical prompt vector, and generating physical constraint loss; calculating