A Rolling Bearing Fault Diagnosis Method Based on Unsupervised Transfer Learning

Abstract

The present invention discloses a rolling bearing fault diagnosis method based on unsupervised transfer learning. The diagnosis method is as follows: Step 1: Collect the data set collected by the machine under fault conditions, and process the data set. The processed data set is stored in the fault server; Step 2: Collect the data of the rolling bearing to be detected, and process the collected data; Step 3: The processed data is diagnosed through a fault diagnosis algorithm based on unsupervised transfer learning; Step 4: Compare the diagnosed data with the data set stored in the fault server. The beneficial effects of the present invention are: The diagnosed data is compared with the data set stored in the fault server and exported to the display terminal for viewing, which increases the accuracy of rolling bearing fault diagnosis; reduces the interference during fault diagnosis; when extracting the pulse signal, a band-pass filter is applied to the original signal to obtain a more pulsed signal for envelope spectrum analysis.

Description

Technical Field

[0001] The present invention belongs to the technical field of rolling bearing fault diagnosis, and particularly relates to a rolling bearing fault diagnosis method based on unsupervised transfer learning. Background Art

[0002] Rotating machinery is the focus of equipment condition monitoring and fault diagnosis work, and a considerable proportion of the faults of rotating machinery are related to rolling bearings; rolling bearings are one of the vulnerable parts of machines. According to incomplete statistics, about 30% of the faults of rotating machinery are caused by rolling bearings.

[0003] During the operation of rolling bearings, damage may be caused by various reasons, such as improper assembly, poor lubrication, intrusion of moisture and foreign objects, corrosion, and overload, etc., which may all lead to premature bearing damage; even when the installation, lubrication, and use and maintenance are normal, after a period of operation, the bearing will also have problems of fatigue spalling and wear and cannot work properly.

[0004] A rolling bearing fault diagnosis method and system based on unsupervised transfer learning with the application number CN202211230446.4. This patent discloses the following steps: a data set with fault labels under a certain working condition is called the source domain, and a data set without fault labels under another working condition is called the target domain. During the transfer learning training process of the network, the sample batch input matrices of the two are input into the network simultaneously for feature extraction, and the features extracted from the two are subjected to domain adaptation; for the fault category output batch matrix G obtained by training the target domain sample batch input matrix through the network, the nuclear norm ‖G‖* of the fault category output batch matrix G is maximized; a deep convolutional neural network model based on the maximization of the fast batch nuclear norm of transfer learning is constructed, and the source data of the rolling bearing to be tested for faults is input into the deep convolutional neural network model to obtain a diagnosis result.

[0005] When diagnosing rolling bearing faults by existing unsupervised transfer learning, in the case of strong shielding signals from other machine components, it is easy to cause interference and affect the accuracy of diagnosis. Summary of the Invention

[0006] The purpose of the present invention is to provide a rolling bearing fault diagnosis method based on unsupervised transfer learning, which reduces interference and improves the accuracy of diagnosis.

[0007] To achieve the above purpose, the present invention provides the following technical solution: a rolling bearing fault diagnosis method based on unsupervised transfer learning, and the diagnosis method is as follows:

[0008] Step 1: Collect the data set collected by the machine under fault conditions, and process the data set. The processed data set is stored in the fault server;

[0009] Step 2: Collect the data of the rolling bearing to be detected and process the collected data;

[0010] Step 3: Diagnose the processed data through a fault diagnosis algorithm based on unsupervised transfer learning;

[0011] Step 4: Compare the diagnosed data with the dataset stored in the fault server and export it to a display terminal for viewing.

[0012] As a preferred technical solution of the present invention, the dataset includes acceleration signals, sampling rates, shaft speeds, load weights, and critical frequencies representing different fault locations.

[0013] As a preferred technical solution of the present invention, the critical frequencies include the ball passing frequency outer ring, the ball passing frequency inner ring, the basic training frequency, and the ball spin frequency.

[0014] As a preferred technical solution of the present invention, amplitude modulation is used at the ball passing frequency outer ring to extract pulse signals or improve the signal-to-noise ratio.

[0015] As a preferred technical solution of the present invention, a band-pass filter is applied to the original signal when extracting the pulse signal.

[0016] As a preferred technical solution of the present invention, spectral detection is required during diagnosis, and the models to be used include a hybrid model, a spatial sub-model, and a statistical model.

[0017] As a preferred technical solution of the present invention, target pixels and background pixels are selected during spectral detection.

[0018] As a preferred technical solution of the present invention, the method for processing the collected data: set the collection rate and collection time; establish a data queue and set a predetermined value according to the set collection rate and collection time; read in the data and count the read-in data; compare the number of read-in data with the predetermined value, if they are equal, pack the data and move it into the data queue, and determine whether to process the data.

[0019] As a preferred technical solution of the present invention, the display terminal includes a display, a mobile phone, and a computer.

[0020] Compared with the prior art, the beneficial effects of the present invention are:

[0021] Compare the diagnosed data with the dataset stored in the fault server and export it to a display terminal for viewing, which increases the accuracy of rolling bearing fault diagnosis;

[0022] Set the acquisition rate and acquisition time; establish a data queue and set a predetermined value according to the set acquisition rate and acquisition time; read in the data and count the read-in data; compare the number of read-in data with the predetermined value, if they are equal, then pack the data and move it into the data queue, and determine whether to process the data, reducing interference during fault diagnosis;

[0023] At the outer ring of the ball passing frequency, amplitude modulation is used to extract the pulse signal or improve the signal-to-noise ratio. The envelope signal generated by amplitude demodulation conveys more diagnostic information, which cannot be obtained by the spectral analysis of the original signal;

[0024] When extracting the pulse signal, apply a band-pass filter to the original signal to obtain a more pulsed signal for envelope spectrum analysis. Brief Description of the Drawings

[0025] Figure 1 It is a flowchart of the rolling bearing fault diagnosis method of the present invention. Detailed Embodiments

[0026] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0027] Embodiment 1

[0028] Please refer to Figure 1 , which is the first embodiment of the present invention. This embodiment provides an unsupervised transfer learning-based rolling bearing fault diagnosis method, including the following steps:

[0029] Step 1: Collect the data set collected by the machine under fault conditions and process the data set. The processed data set is stored in the fault server;

[0030] Step 2: Collect the data of the rolling bearing to be detected and process the collected data;

[0031] Step 3: Diagnose the processed data through a fault diagnosis algorithm based on unsupervised transfer learning;

[0032] Step 4: Compare the diagnosed data with the data set stored in the fault server and export it to the display terminal for viewing.

[0033] The fault diagnosis algorithm of unsupervised transfer learning generates extreme target and background pixels from robust outlier detection, providing inputs for target samples and background samples in transfer learning; target pixels and background pixels are selected during spectral detection; pixels are calculated based on the root points in the segmentation method, aiming to retain the maximum distribution characteristics of the background after dimension reduction; a sparse constraint is imposed during the transfer learning process, with which a simpler and more concentrated subspace can be constructed; multivariate outlier analysis is used to select target pixels and background pixels and use them as training samples; a segmentation method is adopted to obtain the most representative and informative unlabeled samples, so that the rich continuous spatial features in the hyperspectral image can be fully considered; by training labeled samples and unlabeled samples, a subspace construction method based on transfer learning is developed, and pairwise discriminant analysis is used to enhance the separability of target background pixels.

[0034] In this embodiment, preferably, the dataset includes acceleration signals, sampling rates, shaft speeds, load weights, and critical frequencies representing different fault locations.

[0035] In this embodiment, preferably, the critical frequencies include the outer raceway frequency, the inner raceway frequency, the fundamental train frequency, and the ball spin frequency.

[0036] In this embodiment, preferably, at the outer raceway frequency, amplitude modulation is used to extract pulse signals or improve the signal-to-noise ratio. The envelope signal generated by amplitude demodulation conveys more diagnostic information, which cannot be obtained by the spectral analysis of the original signal.

[0037] In this embodiment, preferably, when extracting pulse signals, a band-pass filter is applied to the original signal to obtain a more pulsed signal for envelope spectral analysis.

[0038] In this embodiment, preferably, the display terminal is a display.

[0039] Embodiment 2

[0040] Please refer to Figure 1 , which is the second embodiment of the present invention. This embodiment provides a rolling bearing fault diagnosis method based on unsupervised transfer learning, including the following steps:

[0041] Step 1: Collect the dataset collected by the machine under fault conditions and process the dataset. The processed dataset is stored in the fault server.

[0042] Step 2: Collect the data of the rolling bearing to be detected and process the collected data.

[0043] Step 3: The processed data is diagnosed by the fault diagnosis algorithm based on unsupervised transfer learning.

[0044] Step 4: Compare the diagnosed data with the data set stored in the faulty server and export it to the display terminal for viewing.

[0045] The fault diagnosis algorithm of unsupervised transfer learning generates extreme targets and background pixels from robust outlier detection, providing inputs for the target samples and background samples in transfer learning; target pixels and background pixels are selected during spectral detection; the pixels are calculated based on the root points in the segmentation method, aiming to retain the maximum distribution characteristics of the background after dimension reduction; a sparse constraint is imposed during the transfer learning process, with this constraint, a simpler and more concentrated subspace can be constructed; multivariate outlier analysis is used to select target pixels and background pixels and use them as training samples; a segmentation method is adopted to obtain the most representative and informative unlabeled samples, which can fully consider the rich continuous space features in the hyperspectral image; by training the labeled samples and unlabeled samples, a subspace construction method based on transfer learning is developed, and pairwise discriminant analysis is used to enhance the separability of target background pixels.

[0046] In this embodiment, preferably, the data set includes acceleration signals, sampling rates, shaft speeds, load weights, and critical frequencies representing different fault locations.

[0047] In this embodiment, preferably, the critical frequencies include the outer raceway frequency, the inner raceway frequency, the fundamental train frequency, and the ball spin frequency.

[0048] In this embodiment, preferably, when extracting the pulse signal, a band-pass filter is applied to the original signal to obtain a more pulsed signal for envelope spectrum analysis.

[0049] In this embodiment, preferably, spectral detection is required during diagnosis, and the models to be used include a hybrid model, a spatial sub-model, and a statistical model.

[0050] In this embodiment, preferably, the method for processing the collected data: set the acquisition rate and acquisition time; establish a data queue and set a predetermined value according to the set acquisition rate and acquisition time; read in the data and count the read-in data; compare the number of read-in data with the predetermined value, if they are equal, then pack the data and move it into the data queue, and determine whether to process the data.

[0051] In this embodiment, preferably, the display terminal is a combination of a monitor and a mobile phone.

[0052] Embodiment 3

[0053] Please refer to Figure 1 , which is the third embodiment of the present invention. This embodiment provides a rolling bearing fault diagnosis method based on unsupervised transfer learning, including the following steps:

[0054] Step 1: Collect the data set collected by the machine under fault conditions and process the data set. The processed data set is stored in the fault server;

[0055] Step 2: Collect the rolling bearing data to be detected and process the collected data;

[0056] Step 3: The processed data is diagnosed by a fault diagnosis algorithm based on unsupervised transfer learning;

[0057] Step 4: Compare the diagnosed data with the data set stored in the fault server and export it to the display terminal for viewing.

[0058] The fault diagnosis algorithm of unsupervised transfer learning generates extreme targets and background pixels from robust outlier detection, providing inputs for target samples and background samples in transfer learning; target pixels and background pixels are selected during spectral detection; pixels are calculated based on the root points in the segmentation method, aiming to retain the maximum distribution characteristics of the background after dimension reduction; a sparse constraint is imposed during the transfer learning process, with this constraint, a simpler and more concentrated subspace can be constructed; multivariate outlier analysis is used to select target pixels and background pixels and use them as training samples; a segmentation method is adopted to obtain the most representative and informative unlabeled samples, which can fully consider the rich continuous space features in the hyperspectral image; by training labeled samples and unlabeled samples, a subspace construction method based on transfer learning is developed, and pairwise discriminant analysis is used to enhance the separability of target background pixels.

[0059] In this embodiment, preferably, the data set includes acceleration signals, sampling rates, shaft speeds, load weights, and critical frequencies representing different fault positions.

[0060] In this embodiment, preferably, the critical frequencies include the ball passing frequency outer ring, the ball passing frequency inner ring, the basic training frequency, and the ball spin frequency.

[0061] In this embodiment, preferably, at the ball passing frequency outer ring, amplitude modulation is used to extract pulse signals or improve the signal-to-noise ratio. The envelope signal generated by amplitude demodulation conveys more diagnostic information, which cannot be obtained by the spectral analysis of the original signal.

[0062] In this embodiment, preferably, when extracting pulse signals, a band-pass filter is applied to the original signal to obtain a more pulsed signal for envelope spectrum analysis.

[0063] In this embodiment, preferably, spectral detection is required during diagnosis, and the models to be used include a hybrid model, a spatial submodel, and a statistical model.

[0064] In this embodiment, preferably, the method for processing the collected data is as follows: set the collection rate and collection time; establish a data queue and set a predetermined value according to the set collection rate and collection time; read in the data and count the read-in data; compare the number of read-in data with the predetermined value, if they are equal, then pack the data and move it into the data queue, and determine whether to process the data.

[0065] In this embodiment, preferably, the display terminal is a combination of a display, a mobile phone, and a computer.

[0066] Although the embodiments of the present invention have been shown and described, see the above detailed description. For those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A rolling bearing fault diagnosis method based on unsupervised transfer learning, characterized in that: The diagnosis method is as follows: Step 1: Collect the data set collected by the machine under fault conditions, and process the data set. The processed data set is stored in the fault server; the data set includes acceleration signals, sampling rates, shaft speeds, load weights, and critical frequencies representing different fault positions; the critical frequencies include ball passing frequency outer race, ball passing frequency inner race, fundamental train frequency, and ball spin frequency; Step 2: Collect the data of the rolling bearing to be detected, and process the collected data; Step 3: The processed data is diagnosed by a fault diagnosis algorithm based on unsupervised transfer learning; spectral detection is required during diagnosis, and the models to be used include a hybrid model, a spatial sub-model, and a statistical model; target pixels and background pixels are selected during spectral detection; During the transfer learning process, a sparse constraint is imposed, and multivariate outlier analysis is used to select target pixels and background pixels and use them as training samples; A segmentation method is adopted to obtain the most representative and informative unlabeled samples; by training the labeled samples and unlabeled samples, a subspace construction method based on transfer learning is formulated, and pairwise discriminant analysis is used to enhance the separability of target background pixels; Step 4: Compare the diagnosed data with the data set stored in the fault server, and export it to the display terminal for viewing.

2. A rolling bearing fault diagnosis method based on unsupervised transfer learning according to claim 1, characterized in that: At the ball passing frequency outer race, amplitude modulation is used to extract pulse signals or improve the signal-to-noise ratio.

3. A rolling bearing fault diagnosis method based on unsupervised transfer learning according to claim 2, characterized in that: When extracting the pulse signal, a band-pass filter is applied to the original signal.

4. A rolling bearing fault diagnosis method based on unsupervised transfer learning according to claim 1, characterized in that: The method for processing the collected data: Set the collection rate and collection time; According to the set collection rate and collection time, establish a data queue and set a predetermined value; Read in the data and count the read-in data; Compare the number of read-in data with the predetermined value. If they are equal, pack the data and move it into the data queue, and judge whether to process the data.

5. A rolling bearing fault diagnosis method based on unsupervised transfer learning according to claim 1, characterized in that: The display terminal includes a display, a mobile phone, and a computer.

Images