CN-121977842-A - Graphic neural network rolling bearing fault diagnosis method based on physical guidance

CN121977842ACN 121977842 ACN121977842 ACN 121977842ACN-121977842-A

Abstract

The application relates to the technical field of machine manufacturing, in particular to a graphic neural network rolling bearing fault diagnosis method based on physical guidance and a middleware. The method comprises the steps of obtaining bearing vibration signals of the rolling bearing to be tested, sampling the bearing vibration signals to obtain a plurality of sampling signals, obtaining time domain features and envelope spectrum features of the sampling signals to obtain physical priori features of the rolling bearing to be tested, constructing a diagnosis map, wherein the diagnosis map comprises a plurality of map nodes, the map nodes correspond to the sampling signals one by one, and processing the diagnosis map based on a neural network model which is trained in advance to obtain fault diagnosis results of the rolling bearing to be tested. The method and the middleware for diagnosing the fault of the rolling bearing of the graphic neural network based on the physical guidance can achieve the technical effects of improving the diagnosis efficiency of the rolling bearing of the graphic neural network based on the physical guidance and improving the accuracy of the diagnosis result.

Inventors

DING YUFENG
ZHANG ZIYU
LIU HUIYONG
YUE FAN
Tong quan

Assignees

武汉理工大学
湖北电梯厂有限公司

Dates

Publication Date: 20260505
Application Date: 20251230

Claims (6)

1. The fault diagnosis method for the rolling bearing of the graph neural network based on the physical guidance is characterized by comprising the following steps of: Acquiring a bearing vibration signal of a rolling bearing to be tested, and sampling the bearing vibration signal to obtain a plurality of sampling signals; For each sampling signal, carrying out feature extraction on the sampling signal to obtain a time domain feature and an envelope spectrum feature of the sampling signal, and combining the geometric parameter, the motion parameter and the sampling signal of the rolling bearing to be detected to obtain a physical priori feature of the rolling bearing to be detected; constructing a diagnosis map based on the time domain features, the envelope spectrum features and the physical prior features, wherein the diagnosis map comprises a plurality of map nodes, and the map nodes are in one-to-one correspondence with the sampling signals; And processing the diagnosis map based on the neural network model which is pre-trained to obtain a fault diagnosis result of the rolling bearing to be tested.
2. The method for diagnosing a rolling bearing fault in a physical guidance-based graph neural network of claim 1, wherein the neural network model comprises a first graph rolling network, a second graph rolling network, a first attention path, a second attention path, a fusion network and a classifier network; The first graph convolution network and the second graph convolution network are used for performing dimension mapping on the time domain features, the envelope spectrum features and the physical priori features of the sampling signals to obtain convolution network output; The first attention path is used for carrying out nonlinear transformation and normalization processing on the convolution network output to obtain a first attention weight vector, and multiplying the first attention weight vector with the convolution network output element by element to obtain a first node characteristic representation; The second attention path is used for carrying out nonlinear transformation and normalization processing on the physical prior characteristics of each sampling signal to obtain a second attention weight vector, and multiplying the second attention weight vector with the convolution network output element by element to obtain a second node characteristic representation; the fusion network is used for fusing the first node characteristic representation and the second node characteristic representation to obtain fusion characteristics; And the classifier network is used for outputting the fault probability distribution of the rolling bearing to be tested according to the fusion characteristic as the fault diagnosis result.
3. The method for diagnosing a rolling bearing failure based on a physical guidance according to claim 1, wherein said constructing a diagnostic map based on the time domain features, the envelope spectrum features, and the physical prior features comprises: constructing a plurality of map nodes, wherein the map nodes are in one-to-one correspondence with the sampling signals; calculating to obtain the signal similarity between any two sampling signals according to the time domain features, the envelope spectrum features and the physical priori features corresponding to the sampling signals; determining neighboring nodes of each map node according to the signal similarity, and constructing map edges connecting the map nodes and the neighboring nodes to form the diagnostic map; And for any map edge, calculating the physical similarity corresponding to the map edge, wherein the physical similarity is the similarity of the physical prior features corresponding to two map nodes connected with the map edge, and weighting and calculating the signal similarity and the material similarity to obtain the edge weight of the map edge.
4. The method for diagnosing a rolling bearing fault based on a physical guidance according to claim 1, wherein the step of obtaining the physical priori characteristics of the rolling bearing to be tested by combining the geometric parameters, the motion parameters and the sampling signals of the rolling bearing to be tested comprises the steps of: Obtaining a plurality of fault characteristic frequency theoretical formulas, wherein each fault characteristic frequency theoretical formula corresponds to a rolling bearing fault type; Calculating the geometric parameters and the motion parameters according to the theoretical formulas of the fault characteristic frequencies to obtain theoretical fault characteristic frequencies corresponding to the fault types of the rolling bearings; acquiring an envelope spectrum signal of the sampling signal; determining a plurality of target frequency ranges according to a preset narrowband range and each theoretical fault characteristic frequency; Calculating frequency domain energy corresponding to each rolling bearing fault type according to each target frequency range and each envelope spectrum signal; Normalizing all the frequency domain energy to obtain energy duty ratios corresponding to the frequency domain energy; And constructing the characteristic vector according to the energy duty ratio.
5. The method for diagnosing a rolling bearing failure in a graph neural network based on physical guidance according to claim 4, wherein said acquiring an envelope spectrum signal of the sampled signal includes: performing Hilbert transform on the sampling signal to obtain an amplitude envelope signal; performing fast Fourier transform on the amplitude envelope signal to obtain the envelope spectrum signal; the geometric parameter comprises the number of rolling bodies of the rolling bearing to be tested Rolling element size Size of pitch circle Angle of contact angle The motion parameters comprise the bearing rotating speed of the rolling bearing to be tested ; The rolling bearing fault type comprises an inner ring fault, an outer ring fault, a rolling body fault and a retainer fault; the failure characteristic frequency theoretical formula comprises an inner ring failure characteristic formula, an outer ring failure characteristic formula, a rolling body failure characteristic formula and a retainer failure characteristic formula; The calculating the geometric parameter and the motion parameter according to the theoretical formula of the fault characteristic frequency to obtain the theoretical fault characteristic frequency corresponding to the fault type of each rolling bearing comprises the following steps: According to the bearing rotation speed Obtaining the bearing rotating frequency of the rolling bearing to be tested ; Calculating and obtaining the characteristic frequency of the inner ring faults corresponding to the inner ring faults according to the characteristic formula of the inner ring faults ; Calculating according to the outer ring fault characteristic formula to obtain the outer ring fault characteristic frequency corresponding to the outer ring fault ; Calculating according to the rolling body fault characteristic formula to obtain the rolling body fault characteristic frequency corresponding to the rolling body fault ; Calculating and obtaining the retainer fault characteristic frequency corresponding to the retainer fault according to the retainer fault characteristic formula 。
6. A graphic neural network rolling bearing fault diagnosis middleware based on physical guidance, characterized by comprising: the signal acquisition module is used for acquiring bearing vibration signals of the rolling bearing to be detected, and sampling the bearing vibration signals to obtain a plurality of sampling signals; The feature extraction module is used for extracting features of the sampling signals to obtain time domain features and envelope spectrum features of the sampling signals, and combining geometric parameters, motion parameters and the sampling signals of the rolling bearing to be detected to obtain physical priori features of the rolling bearing to be detected; The spectrum construction module is used for constructing a diagnosis spectrum based on the time domain features, the envelope spectrum features and the physical prior features, and the diagnosis spectrum comprises a plurality of spectrum nodes which are in one-to-one correspondence with the sampling signals; And the fault diagnosis module is used for processing the diagnosis map based on the neural network model which is pre-trained to obtain a fault diagnosis result of the rolling bearing to be tested.

Description

Graphic neural network rolling bearing fault diagnosis method based on physical guidance Technical Field The application relates to the technical field of machine manufacturing, in particular to a graphic neural network rolling bearing fault diagnosis method based on physical guidance and a middleware. Background Rotating machinery is a core component of modern industrial systems, where the health of the rolling bearings is directly related to the stability and safety of the whole device. The traditional rolling bearing fault diagnosis method relies on artificial feature extraction, and has poor adaptability under strong noise and variable working condition environments. In recent years, although a data driving method represented by a Graph Neural Network (GNN) can automatically learn features, the method still has limitations, and in addition, the existing domain self-adaptive method generally depends on complex countermeasure training in the feature distribution alignment process, has long training time and poor stability, and is not beneficial to industrial real-time application. The existing diagnostic methods based on the graph neural network mostly depend on pure data driving, and judgment is carried out by learning the statistical correlation among signals. In the early stage of failure, a weak failure signal is extremely easy to be submerged by noise, so that a model is difficult to effectively identify, when working conditions change, the statistical characteristics of the signal change along with the change, the generalization capability and the robustness of the model are reduced, meanwhile, a model driven by pure data lacks of physical interpretation, the diagnosis result of the model is difficult to be related with an actual physical failure mechanism, and the reliable application of the model in an industrial scene is limited. Therefore, how to improve the accuracy and the diagnosis efficiency of the diagnosis result of the model at the algorithm level becomes an important development direction of the current rolling bearing fault diagnosis technology. Disclosure of Invention In view of the foregoing, it is necessary to provide a method and an intermediate for diagnosing faults of a rolling bearing based on a physical guidance graph neural network, so as to achieve the technical effects of improving the efficiency of diagnosing faults of the rolling bearing and improving the accuracy of the diagnosis result. In order to solve the technical problem, in a first aspect, the present application provides a method for diagnosing a fault of a rolling bearing of a graph neural network based on physical guidance, including: Acquiring a bearing vibration signal of a rolling bearing to be tested, and sampling the bearing vibration signal to obtain a plurality of sampling signals; For each sampling signal, carrying out feature extraction on the sampling signal to obtain a time domain feature and an envelope spectrum feature of the sampling signal, and combining the geometric parameter, the motion parameter and the sampling signal of the rolling bearing to be detected to obtain a physical priori feature of the rolling bearing to be detected; constructing a diagnosis map based on the time domain features, the envelope spectrum features and the physical prior features, wherein the diagnosis map comprises a plurality of map nodes, and the map nodes are in one-to-one correspondence with the sampling signals; And processing the diagnosis map based on the neural network model which is pre-trained to obtain a fault diagnosis result of the rolling bearing to be tested. In one possible embodiment, the neural network model comprises a first graph roll-up network, a second graph roll-up network, a first attention path, a second attention path, a fusion network, and a classifier network; The first graph convolution network and the second graph convolution network are used for performing dimension mapping on the time domain features, the envelope spectrum features and the physical priori features of the sampling signals to obtain convolution network output; The first attention path is used for carrying out nonlinear transformation and normalization processing on the convolution network output to obtain a first attention weight vector, and multiplying the first attention weight vector with the convolution network output element by element to obtain a first node characteristic representation; The first attention path is used for carrying out nonlinear transformation and normalization processing on the physical prior characteristics of each sampling signal to obtain a second attention weight vector, and multiplying the second attention weight vector with the convolution network output element by element to obtain a second node characteristic representation; the fusion network is used for fusing the first node characteristic representation and the second node characteristic representation to obtain fusion characteristics;