Bearing fault diagnosis method and system based on Grubrum angle field and transfer learning

By converting the timing data of rolling bearings into Gram angle field images and combining transfer learning methods, the high demand for training samples in the existing technology is solved, and the accuracy and applicability of fault diagnosis on small sample data sets are improved. , realizes early fault detection and equipment safety guarantee.

CN119048833APending Publication Date: 2024-11-29XIAN UNIV OF POSTS & TELECOMM

Patent Information

Application Number
CN202411181568.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-27
Publication Date
2024-11-29

AI Technical Summary

Technical Problem

The existing network model of rolling bearing fault diagnosis methods is relatively complex and requires high number of training samples. The model is difficult to effectively train in the absence of sufficient training samples, and the generalization performance is poor.

Method used

Using a method based on Gram Angle Field and transfer learning, the timing data of rolling bearings is converted into Gram Angle Field images, and the fine-tuning of the convolutional neural network model is enhanced by geometric transform data and pre-trained convolutional neural network model to improve the generalization ability and application of the model. sex.

Benefits of technology

The accuracy and applicability of fault diagnosis are improved on the small sample data set, and the bearing failures can be detected in the early stage, extend the service life of the equipment, and improve asset utilization efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119048833A_ABST
    Figure CN119048833A_ABST
Patent Text Reader

Abstract

The invention belongs to the technical field of rolling bearing fault diagnosis, and discloses a rolling bearing fault diagnosis method based on a Grubrum angle field and transfer learning, which comprises the following steps of: preprocessing and cutting collected rolling bearing vibration signal data, and coding a data sample into an image in a polar coordinate system by using a Grubrum matrix; according to the method, the running time sequence data of the rolling bearing is converted into the Grubrum angle field image for representation, so that important features and modes hidden in the data can be observed more visually, the spatial information and correlation of the data can be better captured, and different health states of the rolling bearing can be effectively represented; a multi-transfer learning architecture is adopted, the generalization performance and applicability of the established transfer learning model on different data sets are further improved, the effectiveness of the model in practical application is further improved, and when the number of available labeled samples of a target domain is small, the effect of using a secondary transfer learning fault diagnosis model is better.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

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

[0002] Rolling bearings are widely used in industrial equipment, and fault monitoring and diagnosis of rolling bearings are important for ensuring the reliable operation of equipment. Current fault diagnosis methods based on deep learning often require a large number of training data sets, which is difficult to meet in actual industrial environments. To this end, scholars have introduced transfer learning methods to train models on source domain data sets and transfer relevant knowledge to target domain data, thereby alleviating the problem of insufficient fault samples. On the other hand, the representation method of input data has an important impact on the performance of the prediction model. Industrial equipment operation data is mainly multidimensional time series data. Converting time series data into images is an effective way to represent data, which can more intuitively observe the important features and patterns hidden in the data, and help to better capture the spatial information and correlation of the data. CN202211293256.7 discloses a bearing fault diagnosis method based on recursive graph and deep adaptive Transformer, which collects bearing vibration signal data and performs enhanced preprocessing, then uses recursive graph coding technology to generate a two-dimensional texture image from a one-dimensional vibration signal to form a feature map, and divides it into three data sets: training set, verification set, and test set, which are input into the established deep adaptive Transformer-based bearing fault diagnosis device for model training, parameter optimization, and model testing to complete bearing fault diagnosis, solving the problem of low bearing fault diagnosis accuracy in the prior art. However, the established network model is relatively complex, and has a high demand on the number of training samples. In the absence of sufficient training samples, the model is difficult to train effectively, and the generalization performance is poor.

[0003] Through the above analysis, the problems and defects of the prior art are as follows:

[0004] The established network model is relatively complex and has a high demand on the number of training samples. In the absence of sufficient training samples, the model is difficult to train effectively and the generalization performance is poor. Summary of the invention

[0005] In view of the problems existing in the prior art, the present invention provides a rolling bearing fault diagnosis method based on Gram angle field and transfer learning.

[0006] The present invention is implemented in this way: a rolling bearing fault diagnosis method based on Gram angle field and transfer learning comprises:

[0007] Step 1: Image conversion;

[0008] The vibration signals of rolling bearings under different health conditions are collected and segmented as initial input data. The one-dimensional original signal is mapped to the polar coordinate system using the Gramian matrix and then converted into a two-dimensional Gramian angle difference field image as the input of the fault diagnosis model.

[0009] Step 2: Data enhancement;

[0010] The data enhancement method using geometric transformation increases the diversity and richness of image data by using image rotation, flipping, and scaling transformations, improves the generalization ability of the model, and reduces the risk of overfitting. The image data sets are then converted into RGB images before being input into the pre-trained model.

[0011] Step 3: Fine-tune the pre-trained model;

[0012] Use the convolutional neural network model trained with the ImageNet dataset as the pre-trained model, and use the rolling bearing fault dataset to fine-tune the parameters;

[0013] Step 4: Fault diagnosis model application;

[0014] According to the migration strategy in the above steps, the model with parameter fine-tuning can be applied to the fault diagnosis task of a new test data set in the target domain.

[0015] Furthermore, based on the analysis of the running data of rolling bearings, a data representation method is proposed to convert the time series data into images, using the Gram matrix data to encode the Gram angular field image in the polar coordinate system;

[0016] Using the Gram angular field image in combination with transfer learning, a pre-trained model of a convolutional neural network (CNN) is fine-tuned to achieve fault diagnosis;

[0017] The conversion of the time series data into image features includes: cutting the collected rolling bearing vibration signal data samples, then using the Gram matrix to perform linear correlation calculation on the data samples, and encoding the data samples into a Gram angular field image in a polar coordinate system;

[0018] The image features are adjusted, including: assigning the calculated Gram matrix to obtain a color image, and converting all color images into RGB images to unify the pixel size of the final image.

[0019] Furthermore, the time series data is converted into a data representation method of an image, the Gram matrix is ​​calculated, and the collected time series data is fitted into a Gram angular field image.

[0020] Further, the classification model training is performed using the Gram angle field image to obtain the fault diagnosis result of each sample in the data set, including:

[0021] Assign index values ​​to normal and abnormal types as labels for each type, divide the source domain data into training set and test set to form a modeling dataset; use the convolutional neural network model trained with the ImageNet dataset as the pre-training model, and use the rolling bearing fault dataset to fine-tune the parameters; choose a more suitable migration strategy based on the number of labeled samples available in the target domain for model fine-tuning; when the number of available labeled samples is large, adopt a one-time migration strategy, that is, directly migrate from the original pre-trained model to the target domain bearing fault dataset; when the number of available labeled samples is small, adopt a two-time migration strategy, that is, use the bearing fault dataset under different working conditions as an intermediate transfer, first complete a migration of the original pre-trained model under the source domain dataset, and then perform a second migration of the target domain bearing fault dataset.

[0022] Further, the method of using the Gram angle field image to train the fault diagnosis model to obtain the fault diagnosis result of each sample in the data set also includes:

[0023] The samples in the source domain and target domain are expanded using data enhancement techniques such as cropping, rotation, and scaling. Part of the target domain data is used to perform a convolutional layer parameter freezing test on the convolutional neural network pre-training model. The freezing method with the least impact on the health monitoring accuracy of the target domain is selected to freeze the initial model parameters, and the remaining network parameters are fine-tuned.

[0024] Furthermore, the Gram angular field image obtained according to the Gram matrix also includes:

[0025] The rolling bearing operation data is preprocessed, and the trend items are removed after the effective vibration direction data are screened.

[0026] Another object of the present invention is to provide a rolling bearing fault diagnosis system based on Gram angle field and transfer learning, comprising:

[0027] The image conversion module is used to collect the vibration signals of rolling bearings in different health states, intercept them in sections as initial input data, and use the Gramian matrix to map the one-dimensional original signal to the polar coordinate system and then convert it into a two-dimensional Gramian angle difference field image as the input of the fault diagnosis model;

[0028] The data enhancement module is used to use the data enhancement method of geometric transformation, which increases the diversity and richness of image data by using the image rotation, flipping, and scaling transformation methods, improves the generalization ability of the model, and reduces the risk of overfitting; then the image data sets are converted into RGB images before being input into the pre-trained model;

[0029] A pre-trained model fine-tuning module is used to use a convolutional neural network model trained with the ImageNet dataset as a pre-trained model and fine-tune parameters using a rolling bearing fault dataset;

[0030] The fault diagnosis model application module is used to complete the model with parameter fine-tuning according to the migration strategy in the above steps, so that it can be applied to the fault diagnosis task of the new test data set in the target domain.

[0031] Another object of the present invention is to provide a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the rolling bearing fault diagnosis method based on Gram angle field and transfer learning.

[0032] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to perform the steps of the rolling bearing fault diagnosis method based on Gram angle field and transfer learning.

[0033] Another object of the present invention is to provide an information data processing terminal, which is used to implement the rolling bearing fault diagnosis system based on Gram angle field and transfer learning.

[0034] In combination with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows:

[0035] First, the Gram matrix is ​​used to convert the timing operation signal of the rolling bearing into a Gram angular field image, which can more intuitively display the signal characteristics of the rolling bearing under different health conditions, and improve the model classification performance with the help of machine vision processing methods.

[0036] By adopting the pre-trained and fine-tuned transfer learning architecture, the network was pre-trained using the ImageNet dataset, and then fine-tuned on a small sample of the target domain to achieve recognition of different health states of rolling bearings, which can improve the generalization performance and applicability on small sample datasets.

[0037] Second, as auxiliary evidence of the inventiveness of the claims of the present invention, it is also reflected in the following important aspects:

[0038] (1) The expected benefits and commercial value of the technical solution of the present invention after transformation are:

[0039] The present invention can be applied to new working conditions where there are not enough samples accumulated, that is, scenarios where the target domain has a small number of labeled samples. By selecting a suitable migration strategy to implement fault diagnosis, the anomaly can be discovered in the early failure stage of the rolling bearing, and the bearing failure problem can be discovered and diagnosed in time to ensure the safety of the production process, avoid equipment downtime caused by bearing damage, extend equipment service life, and improve asset utilization efficiency and enterprise economic benefits.

[0040] (2) The technical solution of the present invention fills the technical gap in the industry at home and abroad: The present invention proposes a bearing fault diagnosis method and system based on Gram angle field and transfer learning, which makes up for the shortcomings of the existing method when the number of test condition samples is scarce, and improves the accuracy and reliability of bearing fault diagnosis in new conditions.

[0041] Third, the embodiment of the present invention provides a rolling bearing fault diagnosis method based on Gram angular field and transfer learning. The method first converts the collected vibration signals of rolling bearings in different health states into two-dimensional Gram angular difference field images through image conversion technology as input of the fault diagnosis model. This step effectively utilizes the Gramian matrix to map the one-dimensional original signal to the polar coordinate system, and then converts it into a two-dimensional image form that is easier to process, which facilitates subsequent diagnosis.

[0042] Secondly, the present invention adopts a data enhancement strategy to increase the diversity and richness of image data through geometric transformations such as image rotation, flipping, and scaling. This step not only helps to improve the generalization ability of the model, but also effectively reduces the risk of overfitting. At the same time, the image dataset is converted into RGB images before being input into the pre-trained model, which further improves the compatibility and efficiency of data processing.

[0043] In terms of fine-tuning the pre-trained model, the present invention uses a convolutional neural network model trained with the ImageNet dataset as a pre-trained model, and fine-tunes parameters using a rolling bearing fault dataset. This step fully utilizes the advantages of transfer learning, allowing the model to adapt to new task domains more quickly, thereby improving the accuracy and efficiency of fault diagnosis.

[0044] Finally, after the migration strategy and parameter fine-tuning in the above steps, the method of the present invention can be applied to the fault diagnosis task of the new test data set in the target domain. This innovative method not only solves the problems existing in the prior art, but also achieves significant technical progress and provides a new and more effective solution for rolling bearing fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a flow chart of a rolling bearing fault diagnosis method based on Gram angle field and transfer learning provided in an embodiment of the present invention.

[0046] Figure 2 It is a structural block diagram of a rolling bearing fault diagnosis system based on Gram angle field and transfer learning provided by an embodiment of the present invention.

[0047] Figure 3 This is a comparison effect diagram of the rolling bearing fault diagnosis method based on Gram angle field and transfer learning provided by an embodiment of the present invention and other methods.

[0048] Figure 4 It is a comparison diagram of the effects of different migration strategies of the rolling bearing fault diagnosis method based on Gram angle field and transfer learning provided by an embodiment of the present invention.

[0049] Figure 5 It is a detailed flow chart of a rolling bearing fault diagnosis method based on Gram angle field and transfer learning provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0051] like Figure 1 As shown, a rolling bearing fault diagnosis method based on Gram angle field and transfer learning provided by an embodiment of the present invention includes the following steps:

[0052] S101: image conversion;

[0053] The vibration signals of rolling bearings under different health conditions are collected and segmented as initial input data. The one-dimensional original signal is mapped to the polar coordinate system using the Gramian matrix and then converted into a two-dimensional Gramian angle difference field image as the input of the fault diagnosis model.

[0054] S102: Data enhancement;

[0055] The data enhancement method using geometric transformation increases the diversity and richness of image data by using image rotation, flipping, and scaling transformations, improves the generalization ability of the model, and reduces the risk of overfitting. The image data sets are then converted into RGB images before being input into the pre-trained model.

[0056] S103: Pre-training model fine-tuning;

[0057] Use the convolutional neural network model trained with the ImageNet dataset as the pre-trained model, and use the rolling bearing fault dataset to fine-tune the parameters;

[0058] S104: Fault diagnosis model application;

[0059] According to the migration strategy in the above steps, the model with parameter fine-tuning can be applied to the fault diagnosis task of a new test data set in the target domain.

[0060] The embodiment of the present invention provides a data representation method for converting time series data into an image based on the analysis of the running data of a rolling bearing, and uses the Gram matrix data to encode the Gram angular field image in a polar coordinate system;

[0061] Using the Gram angular field image in combination with transfer learning, a pre-trained model of a convolutional neural network (CNN) is fine-tuned to achieve fault diagnosis;

[0062] The conversion of the time series data into image features includes: cutting the collected rolling bearing vibration signal data samples, then using the Gram matrix to perform linear correlation calculation on the data samples, and encoding the data samples into a Gram angular field image in a polar coordinate system;

[0063] The image features are adjusted, including: assigning the calculated Gram matrix to obtain a color image, and converting all color images into RGB images to unify the pixel size of the final image.

[0064] The data representation method for converting time series data into an image provided by an embodiment of the present invention calculates a Gram matrix and fits the collected time series data into a Gram angular field image.

[0065] The method of using the Gram angle field image to perform classification model training to obtain a fault diagnosis result for each sample in the data set provided by the embodiment of the present invention includes:

[0066] Assign index values ​​to normal and abnormal types as labels for each type, divide the source domain data into training set and test set to form a modeling dataset; use the convolutional neural network model trained with the ImageNet dataset as the pre-training model, and use the rolling bearing fault dataset to fine-tune the parameters; choose a more suitable migration strategy based on the number of labeled samples available in the target domain for model fine-tuning; when the number of available labeled samples is large, adopt a one-time migration strategy, that is, directly migrate from the original pre-trained model to the target domain bearing fault dataset; when the number of available labeled samples is small, adopt a two-time migration strategy, that is, use the bearing fault dataset under different working conditions as an intermediate transfer, first complete a migration of the original pre-trained model under the source domain dataset, and then perform a second migration of the target domain bearing fault dataset.

[0067] The method provided by the embodiment of the present invention uses the Gram angle field image to train a fault diagnosis model, and before obtaining the fault diagnosis result of each sample in the data set, it also includes:

[0068] The samples in the source domain and target domain are expanded using data enhancement techniques such as cropping, rotation, and scaling. Part of the target domain data is used to perform a convolutional layer parameter freezing test on the convolutional neural network pre-training model. The freezing method with the least impact on the health monitoring accuracy of the target domain is selected to freeze the initial model parameters, and the remaining network parameters are fine-tuned.

[0069] The Gram angular field image obtained according to the Gram matrix provided in the embodiment of the present invention also includes:

[0070] The rolling bearing operation data is preprocessed, and the trend items are removed after the effective vibration direction data are screened.

[0071] like Figure 2 As shown, a rolling bearing fault diagnosis system based on Gram angle field and transfer learning provided by an embodiment of the present invention includes:

[0072] The image conversion module is used to collect the vibration signals of rolling bearings in different health states, intercept them in sections as initial input data, and use the Gramian matrix to map the one-dimensional original signal to the polar coordinate system and then convert it into a two-dimensional Gramian angle difference field image as the input of the fault diagnosis model;

[0073] The data enhancement module is used to use the data enhancement method of geometric transformation, which increases the diversity and richness of image data by using the image rotation, flipping, and scaling transformation methods, improves the generalization ability of the model, and reduces the risk of overfitting; then the image data sets are converted into RGB images before being input into the pre-trained model;

[0074] A pre-trained model fine-tuning module is used to use a convolutional neural network model trained with the ImageNet dataset as a pre-trained model and fine-tune parameters using a rolling bearing fault dataset;

[0075] The fault diagnosis model application module is used to complete the model with parameter fine-tuning according to the migration strategy in the above steps, so that it can be applied to the fault diagnosis task of the new test data set in the target domain.

[0076] Another object of the present invention is to provide a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the rolling bearing fault diagnosis method based on Gram angle field and transfer learning.

[0077] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to perform the steps of the rolling bearing fault diagnosis method based on Gram angle field and transfer learning.

[0078] Another object of the present invention is to provide an information data processing terminal, which is used to implement the rolling bearing fault diagnosis system based on Gram angle field and transfer learning.

[0079] like Figure 5 As shown, the rolling bearing fault diagnosis method based on Gram angle field and transfer learning of the present invention comprises the following steps:

[0080] Step 1, data collection.

[0081] Obtain the operating data of the rolling bearing under different working conditions and different health states, specifically the rolling bearing operating data collected by the acceleration sensor under each state.

[0082] Step 2: Data preprocessing and image conversion.

[0083] (1) Clean the collected rolling bearing operation data, filter out the effective direction data and remove the trend data;

[0084] (2) According to the Gram matrix, the running data is encoded in the polar coordinate system and converted into an RGB image. The specific steps are: when encoding the time series L(i) in the polar coordinate system, the GASF / GADF encoding diagram is generated by the trigonometric function sum / difference operation. The calculation formula is as follows:

[0085] φ=arccos(L(i)),-1≤L(i)≤1,L(i)∈L

[0086]

[0087] Where φ represents the time series value in polar coordinates. i and are the timestamp and constant stabilization factor in the polar coordinate space, respectively.

[0088] The time series L(i) is scaled so that all values ​​fall between the interval [-1, 1] or [0, 1]; then the numerical value is encoded as the cosine of the angle and the time label i is encoded as the radius r, thereby representing the rescaled time series L(i) in polar coordinates; finally, as time increases, the corresponding values ​​will bend between different corners on the circle.

[0089] (3) The Gram matrix obtained from the running data of the rolling bearing is normalized to obtain a color image, and all color images are converted into RGB images to unify the pixel size of the final image.

[0090] Step 3: Data enhancement.

[0091] The data enhancement method using geometric transformation utilizes image rotation, flipping, scaling and other transformations to increase the diversity and richness of image data, improve the generalization ability of the model, and reduce the risk of overfitting.

[0092] Step 4: Fine-tune the pre-trained model

[0093] (1) The rolling bearing dataset under different working conditions is divided into source domain and target domain. The obtained source domain images are expanded by data enhancement technology and then input into the pre-trained model, where the pre-trained model uses the network parameters trained by the Imagenet dataset;

[0094] (2) The strategy of freezing some parameters for model fine-tuning is adopted. The weight parameters of the convolutional layer and pooling layer that extract feature structures in the pre-trained model are transferred to the rolling bearing dataset. At the same time, the number of categories in the output layer of the original pre-trained network is changed so that the number of neurons is the same as the number of categories in the bearing fault dataset to build a new model. The bearing fault dataset processed in the source domain is divided into a training set and a test set. The training set is used to complete the fine-tuning of network parameters. After the training is completed, the final model is tested using the test set data to verify the effectiveness of the model. Experiments are conducted by freezing different convolutional layers of the pre-trained model to determine the optimal fine-tuning strategy and obtain a transfer learning fault diagnosis model.

[0095] (3) According to the number of labeled samples in the target domain that can be used for model fine-tuning during transfer learning, a more suitable transfer strategy can be selected. When the number of available labeled samples is large, a one-time transfer strategy can be adopted, that is, directly migrating from the original pre-trained model to the target domain bearing fault dataset. When the number of available labeled samples in the target domain is small, the one-time transfer learning fault diagnosis model obtained by fine-tuning the network parameters using the source domain data is migrated to the target domain. After the target domain data is expanded through data augmentation operations, a small amount of target domain labeled data is used to perform a convolutional layer parameter freezing test on the one-time transfer learning fault diagnosis model to determine the optimal fine-tuning strategy, and a secondary transfer learning fault diagnosis model is obtained.

[0096] Step 4: Model testing.

[0097] According to different migration strategies, the parameters are fine-tuned, and the trained model can be applied to the fault diagnosis task of the new test data set in the target domain.

[0098] In order to prove the creativity and technical value of the technical solution of the present invention, this section provides application examples of the claimed technical solution on specific products or related technologies.

[0099] Using the gearbox rolling bearing dataset published by Southeast University, the experiment was conducted on a powertrain dynamic simulation platform, which consists of six main parts: motor controller, motor, planetary gearbox, reduction gearbox, load, and load controller.

[0100] Step 1, obtaining the operating data of the rolling bearing.

[0101] Five types of bearing health status data were collected under two working conditions: working condition 1, speed 20Hz (1200rpm), no load 0V (0Nm), working condition 2, speed 30Hz (1800rpm), load 2V (7.32Nm), and the sampling frequency was 5120Hz.

[0102] Step 2: data preprocessing.

[0103] (1) The vibration signal of the acceleration sensor on the planetary gearbox in the Z-axis direction was selected for analysis and modeling, and data cleaning was performed to remove trend items. Finally, the data was segmented and the length of a single sample data was unified to 512 sampling points;

[0104] (2) Use the Gram matrix to calculate the linear correlation of the one-dimensional time series data, encode it in the polar coordinate system and convert it into an RGB image;

[0105] Step 3: Model training.

[0106] (1) In this embodiment, the AlexNet network model trained with the Imagenet dataset is selected as the pre-trained model, and the weight parameters of the convolutional pooling layer that extracts the feature structure in the AlexNet pre-trained model are migrated to the rolling bearing dataset to construct a new fault diagnosis model;

[0107] (2) The rolling bearing dataset under different working conditions is divided into a source domain and a target domain. The images of the source domain dataset are expanded through data enhancement techniques such as cropping, rotation, and scaling, and then input into the new fault diagnosis model;

[0108] (3) The strategy of freezing some parameters for model fine-tuning is adopted. The weight parameters of the convolutional layer and pooling layer that extract the feature structure in the pre-trained model are migrated to the rolling bearing dataset. At the same time, the number of categories in the output layer of the original pre-trained network is changed so that the number of its neurons is the same as the number of categories in the bearing fault dataset to construct a new model. Finally, the prediction results are output to complete the construction of a one-time transfer learning fault diagnosis model for the source domain dataset.

[0109] (4) When the number of labeled samples available in the target domain is set to a small sample data set, the first transfer learning fault diagnosis model obtained by fine-tuning the network parameters using the source domain data is migrated to the target domain. After the target domain data is expanded through operations such as cropping, rotation, and scaling, the convolutional layer parameter freezing test of the first transfer learning fault diagnosis model is performed. After determining the optimal fine-tuning strategy, the target domain bearing fault data set is used for a second migration to obtain a secondary transfer learning fault diagnosis model.

[0110] Step 4: Model testing.

[0111] Four different groups of experiments were set up for multi-faceted verification in order to verify the necessity of image conversion, the effects of different model fine-tuning strategies, the superiority of the proposed method compared with a variety of other transfer learning fault diagnosis methods, and the performance of the proposed model on small sample data sets.

[0112] In order to verify the necessity of image conversion, the original one-dimensional time series data and the converted two-dimensional image are respectively input into similar networks for experiments, and the results are shown in Table 1. It can be seen that the input method of converting one-dimensional time series data into images proposed in the present invention is more effective, and the accuracy rate is improved by 5.6% compared with the traditional fault diagnosis model based on one-dimensional time series data, which can realize accurate identification of the operating status of the bearing.

[0113] Table 1 Classification results of different data input methods

[0114]

[0115]

[0116] In order to compare and analyze the effects of different model fine-tuning strategies, the obtained image dataset is used to update the parameters of the AlexNet model pre-trained on the ImageNet dataset, and the experimental results of different model fine-tuning strategies are compared, as shown in Table 2. It can be seen that as the number of frozen layers decreases, the accuracy of the model is gradually improving, and the strategy of not freezing the model parameters finally achieves the highest accuracy. Therefore, when there are sufficient training samples, the migration model can be left unfrozen, but the parameters of the entire pre-trained model can be re-learned and updated.

[0117] Table 2 Prediction results of different fine-tuning strategies

[0118]

[0119] In order to fully illustrate the effectiveness of the deep transfer learning fault diagnosis model based on pre-training fine-tuning proposed in the present invention, the method proposed in this paper is compared with a variety of other transfer learning fault diagnosis methods. For the method proposed in the present invention, a fine-tuning strategy without freezing the convolutional layer is adopted to perform secondary migration between the two working condition data sets. During the second migration, the source domain model is parameter-tuned again using the target domain training set samples. The results are shown in Table 3. It can be seen that compared with other methods, the method proposed in the present invention has significant performance advantages and has achieved the highest accuracy in both migration tasks.

[0120] Table 3 Prediction results of different transfer learning models

[0121]

[0122] In order to explore the performance of the proposed model on a small sample dataset, a small sample training set with different sample numbers was set to verify the model in all aspects. The number of samples in the target domain that can be used for model fine-tuning was changed to 30, 50, and 100, respectively, and the remaining samples in the target domain were used for model testing. The fine-tuning strategy without freezing the convolutional layer of the model was used for secondary migration, and compared with the method of directly migrating parameters from the AlexNet pre-trained model, that is, the one-time migration method. The final results are shown in Table 4. Through horizontal comparison, it can be seen that compared with the one-time migration learning method, the secondary migration learning method using bearing fault datasets under different working conditions for transfer has achieved higher accuracy, indicating that in the case of small samples, using only a small number of samples in the target domain is not enough to fully update the parameters of the pre-trained model, especially when the images of the two datasets are very different. By using a more similar image dataset as a transition, it will be able to lay a better foundation for small sample learning, thereby improving the performance of the model. From a vertical comparison, with the increase in the number of labeled samples that can be used for model fine-tuning, the accuracy of both the secondary transfer method and the primary transfer method is basically increasing, indicating that when more samples are used for model fine-tuning, features that are more suitable for the target domain fault diagnosis problem can be learned, thereby improving the model diagnosis accuracy.

[0123] Table 4 Model prediction results

[0124]

[0125] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. It can be understood by a person of ordinary skill in the art that the above-mentioned devices and methods can be implemented using computer executable instructions and / or contained in a processor control code, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. Such code is provided on the carrier medium. The device and its modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, and can also be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.

[0126] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principle of the present invention should be covered by the protection scope of the present invention.

Claims

1. A rolling bearing fault diagnosis method based on Gram angle field and transfer learning, It is characterized in that The rolling bearing fault diagnosis method based on Gram angle field and transfer learning comprises the following steps: Step 1: Image conversion; The vibration signals of rolling bearings under different health conditions are collected and segmented as initial input data. The one-dimensional original signal is mapped to the polar coordinate system using the Gramian matrix and then converted into a two-dimensional Gramian angle difference field image as the input of the fault diagnosis model. Step 2: Data enhancement; The data enhancement method using geometric transformation increases the diversity and richness of image data by using image rotation, flipping, and scaling transformations, improves the generalization ability of the model, and reduces the risk of overfitting. The image data sets are then converted into RGB images before being input into the pre-trained model. Step 3: Fine-tune the pre-trained model; Use the convolutional neural network model trained with the ImageNet dataset as the pre-trained model, and use the rolling bearing fault dataset to fine-tune the parameters; Step 4: Fault diagnosis model application; According to the migration strategy in the above steps, the model with parameter fine-tuning can be applied to the fault diagnosis task of a new test data set in the target domain.

2. The rolling bearing fault diagnosis method based on Gram angle field and transfer learning as claimed in claim 1, It is characterized in that Based on the analysis of rolling bearing operation data, a data representation method is proposed to convert time series data into images. The Gram matrix data is encoded into Gram angular field images in the polar coordinate system. Using the Gram angular field image in combination with transfer learning, a pre-trained model of a convolutional neural network (CNN) is fine-tuned to achieve fault diagnosis; The conversion of the time series data into image features includes: cutting the collected rolling bearing vibration signal data samples, then using the Gram matrix to perform linear correlation calculation on the data samples, and encoding the data samples into a Gram angular field image in a polar coordinate system; The image features are adjusted, including: assigning the calculated Gram matrix to obtain a color image, and converting all color images into RGB images to unify the pixel size of the final image.

3. The rolling bearing fault diagnosis method based on Gram angle field and transfer learning according to claim 2, It is characterized in that The data representation method converts time series data into images, calculates the Gram matrix, and fits the collected time series data into a Gram angular field image.

4. The rolling bearing fault diagnosis method based on Gram angle field and transfer learning according to claim 2, It is characterized in that The classification model training is performed using the Gram angle field image to obtain the fault diagnosis result of each sample in the data set, including: Assign index values ​​to normal and abnormal types as labels for each type, divide the source domain data into training set and test set to form a modeling dataset; use the convolutional neural network model trained with the ImageNet dataset as the pre-training model, and use the rolling bearing fault dataset to fine-tune the parameters; choose a more suitable migration strategy based on the number of labeled samples available in the target domain for model fine-tuning; when the number of available labeled samples is large, adopt a one-time migration strategy, that is, directly migrate from the original pre-trained model to the target domain bearing fault dataset; when the number of available labeled samples is small, adopt a two-time migration strategy, that is, use the bearing fault dataset under different working conditions as an intermediate transfer, first complete the migration of the original pre-trained model under the source domain dataset, and then perform the second migration of the target domain bearing fault dataset.

5. The rolling bearing fault diagnosis method based on Gram angle field and transfer learning according to claim 2, It is characterized in that The method of using the Gram angle field image to train the fault diagnosis model to obtain the fault diagnosis result of each sample in the data set also includes: The samples in the source domain and target domain are expanded using data enhancement techniques such as cropping, rotation, and scaling. Part of the target domain data is used to perform a convolutional layer parameter freezing test on the convolutional neural network pre-training model. The freezing method with the least impact on the health monitoring accuracy of the target domain is selected to freeze the initial model parameters, and the remaining network parameters are fine-tuned.

6. The rolling bearing fault diagnosis method based on Gram angle field and transfer learning according to claim 2, It is characterized in that The Gram angular field image obtained according to the Gram matrix also includes: The rolling bearing operation data is preprocessed, and the trend items are removed after the effective vibration direction data are screened.

7. A rolling bearing fault diagnosis system based on Gram's angle field and transfer learning, which implements the rolling bearing fault diagnosis method based on Gram's angle field and transfer learning as claimed in any one of claims 1 to 6, It is characterized in that The rolling bearing fault diagnosis system based on Gram angle field and transfer learning includes: The image conversion module is used to collect the vibration signals of rolling bearings in different health states, intercept them in sections as initial input data, and use the Gramian matrix to map the one-dimensional original signal to the polar coordinate system and then convert it into a two-dimensional Gramian angle difference field image as the input of the fault diagnosis model; The data enhancement module is used to use the data enhancement method of geometric transformation, which increases the diversity and richness of image data by using the image rotation, flipping, and scaling transformation methods, improves the generalization ability of the model, and reduces the risk of overfitting; then the image data sets are converted into RGB images before being input into the pre-trained model; A pre-trained model fine-tuning module is used to use a convolutional neural network model trained with the ImageNet dataset as a pre-trained model and fine-tune parameters using a rolling bearing fault dataset; The fault diagnosis model application module is used to complete the model with parameter fine-tuning according to the migration strategy in the above steps, so that it can be applied to the fault diagnosis task of the new test data set in the target domain.

8. A computer device, It is characterized in that The computer device includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the rolling bearing fault diagnosis method based on Gram angle field and transfer learning as described in any one of claims 1-6.

9. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the rolling bearing fault diagnosis method based on Gram angle field and transfer learning as described in any one of claims 1 to 6.

10. An information data processing terminal, It is characterized in that The information data processing terminal is used to implement the rolling bearing fault diagnosis system based on Gram angle field and transfer learning as described in claim 7.

Citation Information

Patent Citations

  • Bearing fault detection method and system based on recurrence plot and depth adaptive Transform

    CN115586005A

Cited By

  • Bearing fault diagnosis method, device and equipment and storage medium

    CN119290399A

  • Oil-immersed iron core reactor fault diagnosis method based on multi-physics field coupling and characteristic spectrum

    CN120163941A

  • Fault diagnosis method for oil-immersed core reactor based on multi-physical field coupling and feature map

    CN120163941B

  • Disturbance identification method based on Gramb angle difference field and transfer learning

    CN120431340A

  • A disturbance identification method based on gram angle difference field and transfer learning

    CN120431340B