CN-122014528-A - Fan bearing detection method and device for wind power generation equipment

CN122014528ACN 122014528 ACN122014528 ACN 122014528ACN-122014528-A

Abstract

The application discloses a fan bearing detection method and device for wind power generation equipment, which drive dynamic parameters in real time through working condition data to adaptively update, and carry out closed-loop constraint on neural network output by combining a physical mechanism, so that the problem of poor adaptability of a variable working condition caused by fixed physical parameters in the traditional diagnosis method is effectively solved, and the physical interpretability of a diagnosis conclusion and the reliability under a complex operation environment are obviously improved.

Inventors

WU SHIDONG
HU JINGPENG
SHI JINGSHU
GUO SHUAINAN
WANG QIAN
SU YIBO
LI YUXIN
ZHANG XIAOLONG
HAN JUNFEI
ZHU XIAOYI
JIA DONGQIANG

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: collecting original vibration signals of the fan bearing and working condition data reflecting the running state of the wind power generation equipment; Constructing an input sample reflecting the mapping relation between the original vibration signal and the working condition data; generating a state estimation subnet predicted value reflecting vibration displacement of the fan bearing generated by the original vibration signal based on the input sample; generating real-time physical parameters including equivalent dynamic stiffness and equivalent damping parameters of the fan bearing based on the working condition data; determining a target loss function through the state estimation subnet predicted value and the real-time physical parameter; training and generating a target model based on the target loss function; When target data comprising a target vibration signal and target working condition data are obtained, a detection result for a target fan bearing is output by adopting the target model based on the target data.
2. The method of claim 1, further comprising, prior to the step of generating a state estimation subnet prediction value reflecting a vibrational displacement of the fan bearing due to the original vibrational signal based on the input samples: building a double-flow network, wherein the double-flow network comprises a state estimation sub-network and a parameter self-adaptive sub-network.
3. The method of claim 2, wherein the step of generating a state estimation subnet prediction value reflecting a vibration displacement of the fan bearing due to the original vibration signal based on the input sample comprises: And inputting the input sample into the state estimation sub-network, and controlling the state estimation sub-network to output a state estimation sub-network predicted value reflecting the vibration displacement of the fan bearing generated by the original vibration signal.
4. A method according to claim 3, wherein the operating condition data comprises at least instantaneous rotational speed data and instantaneous power data, and the step of generating real-time physical parameters including equivalent dynamic stiffness and equivalent damping parameters of the fan bearing based on the operating condition data comprises: And inputting the working condition data into the parameter self-adaptive sub-network, and controlling the parameter self-adaptive sub-network to output real-time physical parameters including equivalent stiffness parameters and equivalent damping parameters under the current working condition according to the instant rotating speed data and the instant power data.
5. The method of claim 4, wherein the step of determining an objective loss function from the state estimation subnet predicted value and the real-time physical parameter comprises: establishing a two-degree-of-freedom dynamics equation based on the geometric structure of the fan bearing; Calculating an external load excitation item representing external impact force born by the fan bearing in real time by utilizing the instant power data; calculating the derivative of the state estimation subnet predicted value, substituting the derivative, the real-time physical parameter and the external load excitation item into the two-degree-of-freedom dynamics equation together, and determining physical consistency loss by calculating residual errors at two ends of the two-degree-of-freedom dynamics equation; Calculating the mean square error between the state estimation subnet predicted value and the original vibration signal to determine data fitting loss; Based on the overflow degree of the residual error relative to a preset physical rule violation tolerance zone, combining a learnable relaxation variable to construct soft reward loss for representing the degree of deviation of a model from the physical rule; And establishing a composite total loss function consisting of the data fitting loss, the physical consistency loss and the soft reward loss.
6. The method of claim 5, wherein the step of generating a target model based on the target loss function training comprises: Setting initial weighting coefficients of the data fitting loss, the physical consistency loss and the soft reward and punishment loss; Acquiring initial gradient norms of each loss item to network parameters in the data fitting loss, the physical consistency loss and the soft reward and punishment loss; The values of the corresponding physical consistency loss and soft reward loss in the initial weighting coefficients are adjusted to obtain first-stage target weighting coefficients; preheating iteration is carried out on the double-flow network by utilizing the first-stage target weighting coefficient, so that preheating network weights which primarily capture the original vibration signal distribution rule are obtained; calculating the gradient energy duty ratio of each loss item in the current iteration round in real time by taking the initial gradient norm as a reference value through a gradient norm balance algorithm to obtain a second-stage real-time weighting coefficient dynamically corrected along with time; Performing joint optimization iteration on the composite total loss function by utilizing the second-stage real-time weighting coefficient and the preheating network weight to obtain candidate network parameters of which the physical residual stability falls into a preset physical rule violation tolerance band; and when the composite total loss value corresponding to the candidate network parameter is judged to reach a preset convergence threshold, freezing the candidate network parameter, and generating a target model.
7. The method of claim 5, wherein the step of outputting a detection result for a target fan bearing based on the target data using the target model comprises: extracting a target vibration sequence and target working condition characteristics which are matched with the dimension of an input layer of the target model in the target data to obtain an input tensor to be diagnosed; Inputting the input tensor to be diagnosed into a state estimation subnet in the target model to obtain a predicted vibration displacement sequence reflecting the real-time dynamic response of the target bearing fan; inputting the target working condition characteristics in the input tensor to be diagnosed into a parameter self-adaptive subnet in the target model to obtain a real-time rigidity predicted value and a real-time damping predicted value which correspond to the current working condition; calculating the derivative of the predicted vibration displacement sequence, substituting the derivative, the real-time stiffness predicted value, the real-time damping predicted value and an external load item calculated by the target working condition characteristics into a two-degree-of-freedom dynamics equation based on the geometric structure of a target fan bearing to obtain a target residual error distribution sequence; Obtaining a parameter degradation evaluation index reflecting the abnormal drift trend of the physical parameter by comparing the deviation degree of the real-time stiffness predicted value and the real-time damping predicted value relative to a reference value; Counting the average value and the peak value of the target residual error distribution sequence on a time axis, and carrying out quantitative evaluation by combining the physical rule violation tolerance zone to obtain a physical credibility score reflecting the degree of the model prediction behavior conforming to the physical rule; and integrating the parameter degradation evaluation index and the physical credibility score, executing fault logic judgment, and outputting a complete detection result comprising fault type, severity and diagnosis credibility.
8. The utility model provides a fan bearing detection device of wind power generation equipment which characterized in that includes: The training data acquisition module is used for acquiring an original vibration signal of the fan bearing and working condition data reflecting the running state of the wind power generation equipment; The input sample construction module is used for constructing an input sample reflecting the mapping relation between the original vibration signal and the working condition data; The state estimation subnet predicted value generation module is used for generating a state estimation subnet predicted value reflecting vibration displacement of the fan bearing generated by the original vibration signal based on the input sample; the real-time physical parameter generation module is used for generating real-time physical parameters comprising equivalent dynamic stiffness and equivalent damping parameters of the fan bearing based on the working condition data; The target loss function determining module is used for determining a target loss function through the state estimation subnet predicted value and the real-time physical parameter; the target model generation module is used for generating a target model based on the target loss function training; And the detection result generation module is used for outputting a detection result aiming at the target fan bearing based on the target data by adopting the target model when the target data comprising the target vibration signal and the target working condition data are acquired.
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 With the large-scale application of the wind power generation technology, the fan bearing is used as a core supporting component, and the running state of the fan bearing directly influences the safety and efficiency of the unit. At present, fan bearing fault diagnosis mainly faces the following technical challenges: Related art mechanism model diagnostic methods typically rely on fixed kinetic parameters (such as stiffness and damping). However, the wind generating set is in a random fluctuation working condition for a long time, and the equivalent dynamic characteristics of the bearing are nonlinear offset due to frequent changes of the rotating speed and the load. Because the traditional model cannot capture the dynamic evolution of the physical parameters in real time, the constraint accuracy of the physical equation under the complex working condition is insufficient, and false alarm is easy to generate. Although the related art deep learning diagnosis method can process complex vibration signals, the related art deep learning diagnosis method is often regarded as a 'black box' model, and the diagnosis result lacks clear physical basis. In a scene lacking massive annotation data, the pure data driving model is difficult to learn the dynamics rule behind vibration. When the operation condition exceeds the distribution range of the training set, the generalization capability of the model can be rapidly reduced, and the physical consistency of the diagnosis conclusion is difficult to ensure. 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: collecting original vibration signals of the fan bearing and working condition data reflecting the running state of the wind power generation equipment; Constructing an input sample reflecting the mapping relation between the original vibration signal and the working condition data; generating a state estimation subnet predicted value reflecting vibration displacement of the fan bearing generated by the original vibration signal based on the input sample; generating real-time physical parameters including equivalent dynamic stiffness and equivalent damping parameters of the fan bearing based on the working condition data; determining a target loss function through the state estimation subnet predicted value and the real-time physical parameter; training and generating a target model based on the target loss function; When target data comprising a target vibration signal and target working condition data are obtained, a detection result for a target fan bearing is output by adopting the target model based on the target data. Optionally, before the step of generating a state estimation subnet prediction value reflecting the vibration displacement of the fan bearing due to the original vibration signal based on the input sample, the method further comprises: building a double-flow network, wherein the double-flow network comprises a state estimation sub-network and a parameter self-adaptive sub-network. Optionally, the step of generating a state estimation subnet prediction value reflecting the vibration displacement of the fan bearing due to the original vibration signal based on the input sample includes: And inputting the input sample into the state estimation sub-network, and controlling the state estimation sub-network to output a state estimation sub-network predicted value reflecting the vibration displacement of the fan bearing generated by the original vibration signal. Optionally, the working condition data at least includes instantaneous rotational speed data and instantaneous power data, and the step of generating real-time physical parameters including equivalent dynamic stiffness and equivalent damping parameters of the fan bearing based on the working condition data includes: And inputting the working condition data into the parameter self-adaptive sub-network, and controlling the parameter self-adaptive sub-network to output real-time physical parameters including equivalent stiffness parameters and equivalent damping parameters under the current working condition ac