Bearing fault diagnosis model training method, diagnosis method and system

Abstract

The invention discloses a bearing fault diagnosis model training method, a bearing fault diagnosis method and a bearing fault diagnosis system, and the training method comprises the steps: firstly introducing a Stochastic Depth mechanism into a SENet model, improving the SENet model, and employing an ImageNet natural image data set as a source domain to carry out the pre-training of the improved SENet model; dividing the pre-trained SENet model into a plurality of sub-networks, performing quantitative analysis on feature similarity between adjacent sub-networks by adopting mean square error and structural similarity determination, and determining the adjacent sub-networks as freezing layers according to the similarity to perform transfer learning training; taking a real data sample of the bearing in each state as a target domain, and carrying out transfer learning training on a frozen layer to obtain a rolling bearing fault diagnosis model based on natural data set original domain transfer; the model can effectively improve the bearing fault diagnosis accuracy under the condition that fault data samples are insufficient, and an effective tool is provided for rolling bearing fault diagnosis under the actual engineering background.

Description

Technical Field

[0001] The present invention relates to the technical field of bearing fault diagnosis, and in particular to a bearing fault diagnosis model training method, a diagnosis method and a system. Background Art

[0002] As the core supporting parts of rotating machinery, the reliability of rolling bearings is crucial to the service performance of mechanical equipment. Once a rolling bearing fails, it will directly lead to the loss of function and damage of the equipment, and even cause huge economic losses and casualties in serious cases. In view of this, the research on rolling bearing fault diagnosis technology has important economic benefits and practical value.

[0003] With the development of science and technology, rolling bearing fault diagnosis methods based on deep learning are becoming increasingly mature. However, this type of method has a potential drawback, which is that it requires large-capacity, balanced distribution of normal and faulty bearing data samples for model training. In actual engineering environments, most rolling bearings are in normal operation, and only a few rolling bearings are faulty. Therefore, the actual amount of faulty bearing data samples collected is too small, making it difficult to establish a large and balanced distribution of normal and faulty bearing data sample library. The above-mentioned unbalanced data sample problem will lead to insufficient accuracy of rolling bearing fault diagnosis methods based on deep learning when dealing with actual engineering problems. Summary of the invention

[0004] In view of the problems existing in the prior art, the present invention provides a bearing fault diagnosis model training method, a diagnosis method and a system, which improve the accuracy of the diagnosis result when the amount of faulty bearing sample data is insufficient.

[0005] The present invention is achieved through the following technical solutions: In a first aspect, the present application provides a bearing fault diagnosis model training method, comprising the following steps: The Stochastic Depth mechanism is introduced into the SENet model to improve the SENet model. The ImageNet natural image dataset is used as the source domain to pre-train the improved SENet model. The pre-trained SENet model is divided into several sub-networks, and the feature similarity between adjacent sub-networks is quantitatively analyzed using mean square error and structural similarity. Based on the similarity, adjacent sub-networks are determined as frozen layers for transfer learning training. The data samples of real bearings in various states are used as the target domain, and the frozen layer is trained by transfer learning to obtain a rolling bearing fault diagnosis model based on the original domain migration of natural data sets.

[0006] Preferably, a Stochastic Depth mechanism is introduced into the SENet model to randomly invalidate the residual block and improve the generalization ability of the SENet model. The Stochastic Depth mechanism is as follows:

[0007] in, for l Layer residual block input; for l Layer residual block output; is probability; is the l-layer residual block transformation function.

[0008] Preferably, the SENet model improvement also includes simplifying the SENet model by regularization to improve the objective function of the SENet model. Add norm penalty parameter , simplify the SENet model.

[0009] Preferably, the modified objective function The expression is as follows:

[0010] in, is the objective function, is the penalty function; is the norm penalty term Hyperparameters of .

[0011] Preferably, determining the adjacent sub-networks as frozen layers according to similarity includes: The feature similarity between the output features of adjacent sub-networks is calculated, the feature similarities are sorted, and the adjacent sub-networks whose feature similarity in the sorting is greater than the threshold are selected as the frozen layer.

[0012] Preferably, the improved SENet model is pre-trained using the ImageNet natural image dataset as a source domain, including: uniformly compressing the images in the source domain to make them into samples of the same size and format, and dividing the sample set into a training set, a validation set, and a test set; The training set is used to pre-train the improved SENet model, the validation set is used to verify the pre-trained SENet model, and the test set is used to test the verified SENet model.

[0013] Preferably, the data samples of the real bearing in various states include: Obtain data samples of the bearing in normal or faulty state; The fault state includes data samples under the fault state of the outer ring of the bearing, data samples under the fault state of the inner ring of the bearing, and data samples under the fault state of the rolling element of the bearing; Convert the sample data into a two-dimensional time-frequency graph.

[0014] In a second aspect, the present application provides a bearing fault diagnosis model training system, characterized in that it includes: The pre-training module is used to introduce the Stochastic Depth mechanism into the SENet model, improve the SENet model, and use the ImageNet natural image dataset as the source domain to pre-train the improved SENet model; The similarity analysis module is used to divide the pre-trained SENet model into several sub-networks, use mean square error and structural similarity to determine the quantitative analysis of the feature similarity between adjacent sub-networks, and determine the adjacent sub-networks as frozen layers for transfer learning training based on the similarity; The transfer training module is used to use data samples of real bearings in various states as the target domain, perform transfer learning training on the frozen layer, and obtain a rolling bearing fault diagnosis model based on the original domain migration of the natural data set.

[0015] In a third aspect, the present application provides a bearing fault diagnosis method, comprising: S10, obtaining a vibration acceleration signal of the bearing; S20, inputting the vibration acceleration signal of the bearing into the rolling bearing fault diagnosis model based on the original domain migration of the natural data set to obtain a diagnosis result of the bearing.

[0016] In a fourth aspect, the present application provides a bearing fault diagnosis system, comprising: An acquisition module is used to obtain the vibration acceleration signal of the bearing; The diagnosis module is used to input the vibration acceleration signal of the bearing into the rolling bearing fault diagnosis model based on the original domain migration of the natural data set to obtain the diagnosis result of the bearing.

[0017] Compared with the prior art, the present invention has the following beneficial technical effects: The present application provides a bearing fault diagnosis model training method, in which the ImageNet natural image dataset is used as a source domain pre-trained network model, thereby effectively reducing the demand for faulty bearing samples. The Stochastic Depth mechanism is introduced into the SENe model to improve the generalization ability of the pre-trained model, so that the subsequent transfer learning of a small number of target domain data samples is more accurate. At the same time, before transfer learning, the mean square error (MSE) and structural similarity (SSIM) are introduced to improve the quantitative analysis of the feature similarity between adjacent sub-networks of the SENet model, and based on this, a transfer training unfreezing layer selection strategy is formulated to effectively improve the feature capture capability and improve the model classification accuracy. Finally, real bearing fault data samples are used as the target domain for transfer learning, which effectively eliminates the distortion of model feature parameters and ensures the diagnostic accuracy of the model in the actual engineering environment.

[0018] The ImageNet natural image dataset used as the source domain for pre-model training in this application has the characteristics of sufficient sample capacity and balanced distribution. It has been verified by a large number of experiments that it can effectively eliminate the problem of insufficient training accuracy caused by unbalanced data samples. At the same time, in the process of transfer learning, real fault bearing data is used as the target domain sample, which makes the model have better robustness and adaptability.

[0019] The present application also proposes a bearing fault diagnosis model training system, an electronic device and a computer storage medium, which have all the advantages of the above-mentioned bearing fault diagnosis model training method. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0021] Figure 1 It is a flow chart of the rolling bearing fault diagnosis method based on the original domain migration of natural data sets of the present invention; Figure 2 It is a training flow chart of the SENet network of the present invention; Figure 3 This is a curve diagram of the accuracy and loss rate of SENet-1 of the present invention; Figure 4 This is a curve diagram of the accuracy and loss rate of SENet-2 of the present invention. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations.

[0023] Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application for which protection is sought, but merely represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0024] In order to solve the problem of sample imbalance in bearing fault diagnosis, this application establishes a rolling bearing fault diagnosis model based on the ImageNet natural image dataset to realize transfer learning of a small amount of data samples, thereby improving the accuracy of bearing fault diagnosis in actual engineering environments.

[0025] A bearing fault diagnosis model training method comprises the following steps: Step 1: Introduce the Stochastic Depth mechanism into the SENet model to improve the SENet model, and use the ImageNet natural image dataset as the source domain to pre-train the improved SENet model; SENet (Squeeze-and-Excitation Networks) is a convolutional neural network architecture that enhances the representation power of the model by explicitly modeling dependencies between channels.

[0026] Stochastic Depth is a regularization technique that simulates network structures of different depths by randomly discarding certain layers in the network during training, thereby improving the generalization ability of the model.

[0027] In this step, the Stochastic Depth mechanism is introduced into the SENet model to enhance the learning ability and robustness of the model. This improvement enables the model to learn feature representations of different depths during training, which is helpful for subsequent transfer learning tasks.

[0028] ImageNet is a large natural image dataset that contains a large number of well-annotated image samples. It is a commonly used pre-training dataset in the field of computer vision. The ImageNet dataset is used as the source domain to pre-train the improved SENet model. The purpose of pre-training is to allow the model to learn a common feature representation on a large amount of natural image data. Through pre-training, the model can capture the basic features in the image (such as edges, textures, shapes, etc.).

[0029] Step 2: Divide the pre-trained SENet model into several sub-networks, use mean square error and structural similarity to determine the quantitative analysis of the feature similarity between adjacent sub-networks, and determine the adjacent sub-networks as frozen layers for transfer learning training based on the similarity; In order to perform quantitative analysis of feature similarity, the pre-trained SENet model is first divided into several sub-networks, each of which contains a part of the model's layers. The mean square error and structural similarity are used to perform quantitative analysis of feature similarity, and certain sub-networks are determined as frozen layers.

[0030] The mean square error (MSE) measures the average error between two signals and can reflect the numerical similarity of feature representations.

[0031] Structural similarity (SSIM) is a measure of the similarity between two images, taking into account the similarity in brightness, contrast, and structure.

[0032] In this step, the MSE and SSIM indicators are used to quantify the feature similarity between adjacent subnetworks. By calculating the MSE and SSIM values ​​between the output features of adjacent subnetworks, their numerical and visual similarity can be evaluated. If the feature similarity between adjacent subnetworks is high (MSE value is small and SSIM value is close to 1), it can be considered that the feature representations of these layers are similar between the source domain and the target domain. Therefore, in the process of transfer learning, these similar adjacent subnetworks can be used as frozen layers, keeping their parameters unchanged, and only updating the parameters of the remaining layers. Doing so can effectively utilize the knowledge of the source domain pre-trained model while avoiding problems such as overfitting and negative transfer.

[0033] Step 3: Use the data samples of real bearings in various states as the target domain, perform transfer learning training on the frozen layer, and obtain a rolling bearing fault diagnosis model based on the original domain migration of the natural data set.

[0034] Collect data samples of real bearings in various states, such as normal state, vibration signals under different fault types, etc. These data samples will be used as target domain data for transfer learning training of the frozen layer.

[0035] In the process of transfer learning, the parameters of the frozen layer are kept unchanged, and only the parameters of the remaining layers are updated. The model is trained using the target domain data (i.e., real bearing data samples) so that the model can adapt to the tasks in the target domain (i.e., rolling bearing fault diagnosis). After transfer learning training, a rolling bearing fault diagnosis model based on the original domain migration of the natural data set is obtained. This model combines the knowledge of the pre-trained model in the source domain (ImageNet natural image data set) and the unique feature representation of the target domain (real bearing data samples), and has strong generalization ability and accuracy. Therefore, the model trained by transfer learning can learn the unique feature representation of the target domain, and the model can be used to diagnose faults in new bearing data, thereby improving the efficiency and accuracy of diagnosis.

[0036] Example 1 See also Figure 1-4 , a bearing fault diagnosis model training method, comprising the following steps: S1: Obtain the vibration acceleration signals of the bearing in normal and faulty states, use the vibration acceleration signals as data samples of the bearing, classify the vibration signals according to different fault states, and label the data sample types using one-hot encoding.

[0037] The bearing data samples in this embodiment include data samples under the normal state of the bearing, data samples under the outer ring fault state of the bearing, data samples under the inner ring fault state of the bearing, and data samples under the rolling element fault state of the bearing, totaling samples under 4 states. Table 1 is the mechanical bearing state label, using 0-3 as the state label.

[0038] Table 1 Mechanical bearing status labels

[0039] S2: The continuous wavelet transform (cwt) method is used to convert the vibration acceleration signal of the bearing under various conditions into a two-dimensional time-frequency diagram. The continuous wavelet transform (cwt) method is as follows: (1) S3: Stochastic Depth is introduced into the SENet model to improve the SENet model, realize random failure of residual blocks, and enhance the generalization ability of the SENet network.

[0040] The Stochastic Depth is shown in formula (2): (2) in, for l Layer residual block input; for lLayer residual block output; is the probability, Bernoulli random variables; is the transformation function of the l-layer residual block; characteristic parameter. The probability corresponding to Stochastic Depth for: (3) in, L is the total number of network layers; is the minimum retention probability.

[0041] Step 4: Use regularization to simplify the improved SENet model to obtain a simplified SENet model.

[0042] Towards the objective function Add norm penalty parameter , and then the modified objective function is used to simplify the model.

[0043] (4) in, is the objective function, is the penalty function; is the norm penalty term Hyperparameters of .

[0044] S5: Use the ImageNet natural image dataset as the source domain to pre-train the simplified SENet model. After the training is completed, the pre-trained SENet model is obtained.

[0045] The ImageNet natural image dataset is used as the source domain, and the images in the source domain are uniformly compressed to ensure the same size and format. The data in the source domain are divided into four categories according to different categories, and each category is further divided according to the ratio of training set: validation set: test set = 0.5:0.25:0.25. The training set is input into the simplified SENet model for pre-training, and the learning rate is set between 0.001-0.003, and the number of learning times is set between 80-100. After the training is completed, the pre-trained SENet model is obtained.

[0046] S6: The pre-trained SENet model is divided into several sub-networks. The mean square error (MSE) and structural similarity (SSIM) are introduced using formula (5) to quantitatively analyze the feature similarity between adjacent sub-networks. Based on the similarity, the adjacent sub-networks are determined as frozen layers for transfer learning training.

[0047] The pre-trained SENet model is divided into several sub-networks. The mean square error (MSE) and structural similarity (SSIM) are introduced using formula (5) to quantitatively analyze the feature similarity between adjacent sub-networks and determine the similarity of adjacent sub-networks.

[0048] The MSE threshold is set to =1400 and = 2000, the SSIM threshold is set to =0.5 and =0.7.

[0049] (5) S7: Use the two-dimensional time-frequency diagram of the bearing under various states obtained in step 2 as the target domain, perform transfer learning training on the frozen layer, obtain a rolling bearing fault diagnosis model based on the original domain migration of the natural data set, and perform bearing fault diagnosis based on the rolling bearing fault diagnosis model.

[0050] In this embodiment, adjacent subnetworks corresponding to high similarity and medium similarity are selected as frozen layers for transfer learning training, denoted as SENet-1 and SENet-2. The time-frequency domain diagram of the faulty bearing vibration signal obtained in step 2 is input into the SENet-1 and SENet-2 models as the target domain sample, and the final unbalanced sample rolling bearing fault diagnosis model based on the original domain migration of the natural data set is obtained. The final diagnosis results of the SENet-1 and SENet-2 models are shown in Figure 2. Figure 3 and Figure 4 shown.

[0051] In order to solve the problem of unbalanced fault sample size in bearing fault diagnosis, the real rolling bearing fault data samples are first classified and numbered, and the two-dimensional time-frequency diagram of the fault vibration signal is obtained through wavelet transform; then the ImageNet‌ natural image dataset is used as the source domain to pre-train the SENet (Squeeze-and-Excitation Networks) model, and the ImageNet‌ natural image dataset with sufficient data is used to pre-train the SENet model to reduce the cost of fault data collection; secondly, the SENet model is improved, and the Stochastic Depth mechanism is added to enhance the model learning ability and improve the generalization ability of the model; finally, the mean square error (MSE) and structural similarity (SSIM) are used to analyze the number of frozen layers, and the real fault dataset is used as the training domain to import the improved SENet model for transfer learning, and a rolling bearing fault diagnosis model suitable for unbalanced datasets is established. Through transfer learning, the improved SENet model can perform high-accuracy diagnosis and classification of small sample areas after fine-tuning parameters through a small amount of actual mechanical fault signals. The present invention proposes a rolling bearing fault diagnosis method for unbalanced samples based on the original domain migration of natural data sets, which can effectively improve the bearing fault diagnosis accuracy when the fault data samples are insufficient, and provide an effective tool for rolling bearing fault diagnosis in actual engineering contexts.

[0052] Based on the above-mentioned bearing fault diagnosis model training method, the present application also proposes a bearing fault diagnosis model training system, which may include: The pre-training module is used to introduce the Stochastic Depth mechanism into the SENet model, improve the SENet model, and use the ImageNet natural image dataset as the source domain to pre-train the improved SENet model; The similarity analysis module is used to divide the pre-trained SENet model into several sub-networks, use mean square error and structural similarity to determine the quantitative analysis of the feature similarity between adjacent sub-networks, and determine the adjacent sub-networks as frozen layers for transfer learning training based on the similarity; The transfer training module is used to use data samples of real bearings in various states as the target domain, perform transfer learning training on the frozen layer, and obtain a rolling bearing fault diagnosis model based on the original domain migration of the natural data set.

[0053] The present application also proposes a bearing fault diagnosis method, which may include: S10, obtaining a vibration acceleration signal of the bearing; S20, inputting the vibration acceleration signal of the bearing into a rolling bearing fault diagnosis model based on the original domain migration of the natural data set to obtain a diagnosis result of the bearing.

[0054] Correspondingly, the present application also proposes a bearing fault diagnosis system, which may include: An acquisition module is used to obtain the vibration acceleration signal of the bearing; The diagnosis module is used to input the vibration acceleration signal of the bearing into the rolling bearing fault diagnosis model based on the original domain migration of the natural data set to obtain the diagnosis result of the bearing.

[0055] It should be noted that in the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of each module is only a logical function division. There may be other division methods in actual implementation. For example, multiple modules can be combined or integrated into another device, or some features can be ignored or not executed. The module described as a separate component may or may not be physically separated. The component displayed as a module may be a physical unit or multiple physical units, that is, it may be located in one place, or it may be distributed in multiple different places. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0056] In addition, each module in each embodiment of the present invention may be integrated into a processing unit, each module may exist physically separately, or two or more modules may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of a software functional unit.

[0057] An electronic device provided in an embodiment of the present application includes a memory and a processor. A computer program is stored in the memory. When the processor executes the computer program, the steps of the bearing fault diagnosis model training method described in any of the above embodiments are implemented.

[0058] Another electronic device provided in the embodiment of the present application may also include: an input port connected to the processor, used to transmit multimodal data collected by an external acquisition device to the processor; and a display unit connected to the processor, used to display the processing results of the processor to the outside world; a communication module connected to the processor, used to realize the communication between the electronic device and the outside world. The display unit can be a display panel, a laser scanning display, etc.; the communication mode adopted by the communication module includes but is not limited to mobile high-definition link technology (HML), universal serial bus (USB), high-definition multimedia interface (HDMI), wireless connection (including wireless fidelity technology (WiFi), Bluetooth communication technology, low-power Bluetooth communication technology, and communication technology based on IEEE802.11s).

[0059] An embodiment of the present application provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, the steps of the bearing fault diagnosis model training method described in any of the above embodiments are implemented.

[0060] For the description of the relevant parts of the bearing fault diagnosis model training system, electronic device, and computer-readable storage medium provided in the embodiments of the present application, please refer to the detailed description of the corresponding parts in the bearing fault diagnosis model training method provided in the embodiments of the present application, which will not be repeated here. In addition, the parts of the above-mentioned technical solutions provided in the embodiments of the present application that are consistent with the corresponding technical solutions in the prior art are not described in detail to avoid excessive elaboration.

[0061] The above contents are only for explaining the technical idea of ​​the present invention and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the claims of the present invention.

Claims

1. A bearing fault diagnosis model training method, characterized in that: The following steps are involved: The Stochastic Depth mechanism is introduced into the SENet model to improve the SENet model. The ImageNet natural image dataset is used as the source domain to pre-train the improved SENet model. The pre-trained SENet model is divided into several sub-networks, and the feature similarity between adjacent sub-networks is quantitatively analyzed using mean square error and structural similarity. Based on the similarity, adjacent sub-networks are determined as frozen layers for transfer learning training. The data samples of real bearings in various states are used as the target domain, and the frozen layer is trained by transfer learning to obtain a rolling bearing fault diagnosis model based on the original domain migration of natural data sets.

2. A bearing fault diagnosis model training method according to claim 1, characterized in that: The Stochastic Depth mechanism is introduced into the SENet model to make the residual block randomly invalid and improve the generalization ability of the SENet model. The Stochastic Depth mechanism is as follows: in, for l Layer residual block input; for l Layer residual block output; is probability; is the l-layer residual block transformation function.

3. A bearing fault diagnosis model training method according to claim 1, characterized in that: The SENet model improvement also includes simplifying the SENet model by regularization, and moving the objective function of the SENet model to Add norm penalty parameter , simplify the SENet model.

4. A bearing fault diagnosis model training method according to claim 3, characterized in that: Modified objective function The expression is as follows: in, is the objective function, is the penalty function; is the norm penalty term Hyperparameters of .

5. A bearing fault diagnosis model training method according to claim 1, characterized in that: The step of determining the adjacent sub-network as a frozen layer according to the similarity includes: The feature similarity between the output features of adjacent sub-networks is calculated, the feature similarities are sorted, and the adjacent sub-networks whose feature similarity in the sorting is greater than the threshold are selected as the frozen layer.

6. A bearing fault diagnosis model training method according to claim 1, characterized in that: The improved SENet model is pre-trained using the ImageNet natural image dataset as the source domain, including: uniformly compressing the images in the source domain to make them into samples of the same size and format, and dividing the sample set into training set, validation set and test set; The training set is used to pre-train the improved SENet model, the validation set is used to verify the pre-trained SENet model, and the test set is used to test the verified SENet model.

7. A bearing fault diagnosis model training method according to claim 1, characterized in that: The data samples of the real bearing in various states include: Obtain data samples of the bearing in normal or faulty state; The fault state includes data samples under the fault state of the outer ring of the bearing, data samples under the fault state of the inner ring of the bearing, and data samples under the fault state of the rolling element of the bearing; Convert the sample data into a two-dimensional time-frequency graph.

8. A bearing fault diagnosis model training system, characterized in that: include: The pre-training module is used to introduce the Stochastic Depth mechanism into the SENet model, improve the SENet model, and use the ImageNet natural image dataset as the source domain to pre-train the improved SENet model; The similarity analysis module is used to divide the pre-trained SENet model into several sub-networks, use mean square error and structural similarity to determine the quantitative analysis of the feature similarity between adjacent sub-networks, and determine the adjacent sub-networks as frozen layers for transfer learning training based on the similarity; The transfer training module is used to use data samples of real bearings in various states as the target domain, perform transfer learning training on the frozen layer, and obtain a rolling bearing fault diagnosis model based on the original domain migration of the natural data set.

9. A bearing fault diagnosis method, characterized in that: include: S10, obtaining a vibration acceleration signal of the bearing; S20. Input the vibration acceleration signal of the bearing into the rolling bearing fault diagnosis model based on the original domain migration of the natural data set trained by any training method of claims 1-7 to obtain the diagnosis result of the bearing.

10. A bearing fault diagnosis system, comprising: An acquisition module is used to obtain the vibration acceleration signal of the bearing; A diagnostic module is used to input the vibration acceleration signal of the bearing into the rolling bearing fault diagnosis model based on the original domain migration of the natural data set trained by any training method of claims 1-7 to obtain the diagnosis result of the bearing.

Images