Distributed bearing fault diagnosis method based on point symmetry mode conversion data enhancement
Through the point-symmetric mode conversion data enhancement method, one-dimensional time series information is converted into two-dimensional images, a multi-modal data set is constructed, and a federated learning framework collaborative training model is used to solve the problem of data modality caused by the singleness of data in bearing fault diagnosis. Performance reduction issues improve diagnostic accuracy and robustness, and protect data privacy.
Patent Information
- Application Number
- CN202411323893.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-23
- Publication Date
- 2025-01-10
AI Technical Summary
In bearing fault diagnosis, the diagnostic performance of federated learning is reduced due to the singularity of data modes, resulting in insufficient model accuracy and robustness, especially in the data island phenomenon.
Using point symmetric mode conversion data enhancement method, convert one-dimensional time series information into two-dimensional images, build a multi-modal data set, and collaborate the model between central servers and participating users through the federated learning framework, and update the global model to improve diagnosis performance.
It effectively improves the accuracy and robustness of the model, improves bearing fault diagnosis performance, and protects data privacy, overcomes the limitations of data modality.
Smart Images
Figure CN119295808A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of bearing fault diagnosis, and in particular to a distributed bearing fault diagnosis method based on point symmetric pattern conversion data enhancement. Background Art
[0002] Rolling bearings are key components of mechanical equipment, and their health directly affects the normal operation of the entire equipment. Once a bearing fails, it may not only cause huge economic losses, but also threaten personnel safety. Therefore, in-depth research on bearing fault diagnosis technology is extremely important to ensure the stable and efficient operation of rotating mechanical equipment.
[0003] In recent years, many scholars have devoted themselves to the research in the field of fault diagnosis. Deep learning methods have significantly improved the accuracy of fault diagnosis due to their excellent feature extraction capabilities. However, the high accuracy of data-driven diagnostic methods depends on a large amount of supervised data to train the model. In reality, due to the restrictions of economic costs and data privacy laws, the "data island" phenomenon has emerged. In this case, due to the lack of sufficient training data sets, the rolling bearing intelligent diagnosis method based on deep learning faces significant limitations.
[0004] As a distributed training framework, federated learning can coordinate distributed users to jointly train shared models without infringing data privacy. Applying federated learning to rolling bearing fault diagnosis can effectively solve the problems of data silos and data privacy. However, a major problem facing current technology is that when the data of each client is not independent and identically distributed, the diagnostic performance of federated learning will be significantly reduced. Especially in the field of fault diagnosis, the singleness of data modality has become a key factor leading to the decline in diagnostic performance. Therefore, how to overcome the challenges brought by the singleness of data modality has become the key to improving the performance of federated learning in the field of bearing fault diagnosis. Summary of the invention
[0005] The purpose of the present invention is to provide a distributed bearing fault diagnosis method based on point symmetric mode conversion data enhancement to solve the problems caused by the singleness of data mode.
[0006] To achieve the above object, the present invention provides a distributed bearing fault diagnosis method based on point symmetric pattern conversion data enhancement, comprising the following steps:
[0007] S1. Participating users collect and annotate local bearing fault samples, use the point symmetry mode, i.e., SDP conversion method, to convert the one-dimensional time series information into a two-dimensional image, and use the data of the two modalities of one-dimensional data and two-dimensional data to construct a local data set;
[0008] S2, the central server initializes and distributes the global model to participating users;
[0009] S3, participating users perform local training based on local data sets;
[0010] S4, the participating users upload the local model parameters to the central server;
[0011] S5. The central server aggregates the model parameters uploaded by all participating users and updates the global model.
[0012] S6. The central server sends the updated global model to participating users;
[0013] S7. Repeat steps S3 to S6 until a preset maximum communication round or a predetermined model performance indicator is reached.
[0014] Preferably, in step S1, the steps of converting the one-dimensional time series information into a two-dimensional image using the SDP conversion method are as follows:
[0015] S11, normalizing the original signal;
[0016] S12. Calculate the angle of each point in polar coordinates. For each data point x(i), assign two angle positions symmetrically on both sides of the central axis:
[0017] S13. Each data point is mapped to a polar coordinate system, where r(i) is the radius, and θ(i) and φ(i) are the angle coordinates, to achieve the conversion of one-dimensional signal to two-dimensional image.
[0018] Preferably, the formula for normalization processing in S11 is as follows:
[0019]
[0020] Among them, x(i) is the i-th point in the time series signal, x min With x max are the minimum and maximum values in the time series signal, and r(i) is the normalized value.
[0021] Preferably, the two angle positions in S12 are as follows:
[0022]
[0023] Among them, θ(i) and φ(i) are the two angular positions of the data points, is the base angle of the lth segment, N is the original signal length, ξ is the scaling factor of the angle distribution, and a is the index offset, which is used to define the relationship between adjacent data points.
[0024] Preferably, S2 includes a central server and several participating users.
[0025] Preferably, in step S3, the specific operations of the participating users training the local model based on the local data set are as follows:
[0026] S31, in communication round t, participating user k uses the designed local feature extractor to extract local features of the one-dimensional signal and the two-dimensional image, and concatenates the local features and inputs them into the local classifier, and calculates the classification loss according to the corresponding defect category label;
[0027] S32. Minimize the classification loss to update the parameters of the local feature extractor and the local classifier:
[0028]
[0029] Among them, α is the learning rate, is the differential operator, W t,k ', , which are the updated classifiers, the feature extractors corresponding to the one-dimensional signals, and the feature extractors corresponding to the two-dimensional signals.
[0030] Preferably, the calculation formula of the classification loss in S31 is as follows:
[0031]
[0032] in, is the local data set, n is the sample number, x n ,p n ,y n are the one-dimensional signal, two-dimensional image, and sample label corresponding to the sample number n, respectively. They are the feature extractors corresponding to one-dimensional signals and two-dimensional images, are the extracted features, concat(·,·) represents the connection operation of features, and W t,k is the classifier, It is the probability distribution vector output by connecting the features and sending them to the classifier The probability value corresponding to category c.
[0033] Preferably, in step S5, the specific operation of the central server aggregately calculating the next round of global model is as follows:
[0034]
[0035] Among them, W t+1,G , They are the weights of the t+1 round global classifier, the global one-dimensional signal corresponding feature extractor, and the global two-dimensional image corresponding feature extractor; K is the total number of participating users.
[0036] Therefore, the distributed bearing fault diagnosis method based on point symmetric mode conversion data enhancement using the above structure in the present invention has the following beneficial effects:
[0037] (1) The federated learning framework is used to jointly train the model using the data of multiple participants while ensuring data privacy, which greatly improves the accuracy of the model;
[0038] (2) By converting one-dimensional vibration signals into two-dimensional image information, a local dataset is constructed to improve the robustness and generalization ability of federated learning and the bearing fault diagnosis performance.
[0039] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0040] Figure 1 It is a flow chart of a distributed bearing fault diagnosis method based on point symmetric mode conversion data enhancement of the present invention;
[0041] Figure 2 It is a comparison chart of the experimental results of Example 1 of the present invention under the setting of a single mode and the method of the present invention. DETAILED DESCRIPTION
[0042] Example
[0043] The following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. 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.
[0044] like Figure 1 As shown, this embodiment provides a distributed bearing fault diagnosis method based on point symmetric pattern conversion data enhancement, comprising the following steps:
[0045] S1. Participating users collect and annotate local bearing fault samples, use the point symmetry pattern (SDP) conversion method to convert the one-dimensional time series information into a two-dimensional image, and use the data of the two modalities of one-dimensional data and two-dimensional data to construct a local data set. The specific process is as follows:
[0046] This embodiment uses the method framework of this embodiment to perform fault diagnosis on the data of the drive end and the fan end at different sampling frequencies. The Case Western Reserve University bearing fault data set used is a public data set, which contains normal data and data of different fault types and different fault diameters. SDP conversion can convert the original signal into a mirror-symmetrical snowflake image, so that the recognition and distinction between patterns are directly related to the amplitude and frequency differences between the signals, accurately expressing the changes in the signal.
[0047] The steps of converting one-dimensional time series information into two-dimensional images using the SDP conversion method are as follows:
[0048] First, normalize the original signal:
[0049]
[0050] Among them, x(i) is the i-th point in the time series signal, x min With x max are the minimum and maximum values in the time series signal, and r(i) is the normalized value;
[0051] Next, calculate the angle of each point in polar coordinates, and for each data point x(i), assign two angular positions symmetrically on both sides of the central axis:
[0052]
[0053] Among them, θ(i) and φ(i) are the two angular positions of the data points, is the base angle of the lth segment, N is the original signal length, ξ is the scaling factor of the angle distribution, and a is the index offset, which is used to define the relationship between adjacent data points;
[0054] Finally, each data point is mapped to a polar coordinate system, where r(i) is the radius, θ(i) and φ(i) are the angle coordinates, thus realizing the conversion of one-dimensional signal to two-dimensional image.
[0055] S2. The central server initializes and distributes the global model to participating users.
[0056] S3. Participating users conduct local training based on local data sets. The specific operations are as follows:
[0057] In communication round t, participating user k uses the designed local feature extractor to extract local features of one-dimensional signals and two-dimensional images, and concatenates the local features and inputs them into the local classifier. The classification loss is calculated according to the corresponding defect category label. The calculation formula of the classification loss is as follows:
[0058]
[0059] in, is the local data set, n is the sample number, x n ,p n ,y n are the one-dimensional signal, two-dimensional image, and sample label corresponding to the sample number n, respectively. They are the feature extractors corresponding to one-dimensional signals and two-dimensional images, are the extracted features, concat(·,·) represents the connection operation of features, and W t,k is the classifier, It is the probability distribution vector output by connecting the features and sending them to the classifier The probability value corresponding to category c;
[0060] Finally, the classification loss is minimized to update the parameters of the local feature extractor and local classifier:
[0061]
[0062] Among them, α is the learning rate, is the differential operator, W t,k ', , which are the updated classifiers, the feature extractors corresponding to the one-dimensional signals, and the feature extractors corresponding to the two-dimensional signals.
[0063] S4. Participating users upload local model parameters to the central server.
[0064] S5. The central server aggregates the model parameters uploaded by all participating users and updates the global model. The specific operations of the central server to aggregate and calculate the next round of global model are as follows:
[0065]
[0066] Among them, W t+1,G , They are the weights of the t+1 round global classifier, the global one-dimensional signal corresponding feature extractor, and the global two-dimensional image corresponding feature extractor; K is the total number of participating users.
[0067] S6. The central server sends the updated global model to participating users.
[0068] S7. Repeat steps S3 to S6 until a preset maximum communication round or a predetermined model performance indicator is reached.
[0069] The experimental results of fault diagnosis are compared. Figure 2 As shown, it can be seen that under two different data modality settings, this embodiment achieves a higher classification accuracy.
[0070] Therefore, the present invention adopts the above-mentioned distributed bearing fault diagnosis method based on point-symmetric pattern conversion data enhancement, and obtains a local data set composed of one-dimensional and two-dimensional data by converting local one-dimensional data. The distributed bearing fault diagnosis method based on point-symmetric pattern conversion data enhancement is used to improve the performance of bearing fault diagnosis, while protecting privacy and effectively improving the adverse effects of single data modality.
[0071] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.
Claims
1. A distributed bearing fault diagnosis method based on point symmetric pattern conversion data enhancement, characterized in that: The following steps are involved: S1. Participating users collect and annotate local bearing fault samples, use the point symmetric mode SDP conversion method to convert the one-dimensional time series information into a two-dimensional image, and use the data of the two modalities of one-dimensional data and two-dimensional data to construct a local data set; S2, the central server initializes and distributes the global model to participating users; S3, participating users perform local training based on local data sets; S4, the participating users upload the local model parameters to the central server; S5. The central server aggregates the model parameters uploaded by all participating users and updates the global model. S6. The central server sends the updated global model to participating users; S7. Repeat steps S3 to S6 until a preset maximum communication round or a predetermined model performance indicator is reached.
2. The distributed bearing fault diagnosis method based on point symmetric pattern conversion data enhancement according to claim 1 is characterized in that: The steps of converting one-dimensional time series information into a two-dimensional image using the SDP conversion method in S1 are as follows: S11, normalizing the original signal; S12, calculating the angle of each point in polar coordinates, and for each data point x(i), assigning two angle positions symmetrically on both sides of the central axis; S13. Each data point is mapped to a polar coordinate system, where r(i) is the radius, and θ(i) and φ(i) are the angle coordinates, to achieve the conversion of one-dimensional signal to two-dimensional image.
3. The distributed bearing fault diagnosis method based on point symmetric pattern conversion data enhancement according to claim 2 is characterized in that: The formula for normalization processing in S11 is as follows: Among them, x(i) is the i-th point in the time series signal, x min With x max are the minimum and maximum values in the time series signal, and r(i) is the normalized value.
4. The distributed bearing fault diagnosis method based on point symmetric pattern conversion data enhancement according to claim 2 is characterized in that: The two angular positions in S12 are as follows: Among them, θ(i) and φ(i) are the two angular positions of the data points, is the base angle of the lth segment, N is the original signal length, ξ is the scaling factor of the angle distribution, and a is the index offset, which is used to define the relationship between adjacent data points.
5. The distributed bearing fault diagnosis method based on point symmetric pattern conversion data enhancement according to claim 1 is characterized in that: S2 includes a central server and several participating users.
6. The distributed bearing fault diagnosis method based on point symmetric pattern conversion data enhancement according to claim 1 is characterized in that: The specific operations of users in S3 training local models based on local data sets are as follows: S31, in communication round t, participating user k uses the designed local feature extractor to extract local features of the one-dimensional signal and the two-dimensional image, and concatenates the local features and inputs them into the local classifier, and calculates the classification loss according to the corresponding defect category label; S32. Minimize the classification loss to update the parameters of the local feature extractor and the local classifier: Among them, α is the learning rate, is the differential operator, W t,k ', , which are the updated classifiers, the feature extractors corresponding to the one-dimensional signals, and the feature extractors corresponding to the two-dimensional signals.
7. The distributed bearing fault diagnosis method based on point symmetric pattern conversion data enhancement according to claim 6 is characterized in that: The calculation formula for classification loss in S31 is as follows: in, is the local data set, n is the sample number, x n ,p n ,y n are the one-dimensional signal, two-dimensional image, and sample label corresponding to the sample number n, respectively. They are the feature extractors corresponding to one-dimensional signals and two-dimensional images, are the extracted features, concat(·,·) represents the connection operation of features, and W t,k is the classifier, It is the probability distribution vector output by connecting the features and sending them to the classifier The probability value corresponding to category c.
8. The distributed bearing fault diagnosis method based on point symmetric pattern conversion data enhancement according to claim 1 is characterized in that: The specific operations of the central server in S5 for aggregate calculation of the next round of global model are as follows: Among them, W t+1,G , They are the weights of the t+1 round global classifier, the global one-dimensional signal corresponding feature extractor, and the global two-dimensional image corresponding feature extractor; K is the total number of participating users.