Rolling bearing fault diagnosis method based on attention time-frequency feature extraction

Abstract

The invention discloses a rolling bearing fault diagnosis method based on attention time-frequency feature extraction, and belongs to the technical field of rolling bearing fault diagnosis, and the method comprises the steps: obtaining a vibration signal of a rolling bearing; preprocessing the vibration signal to convert the vibration signal into a first time frequency image; the first time-frequency image is input into a first convolutional neural network, the first convolutional neural network comprises a first branch, a second branch and a full connection layer, the first branch comprises CBAM, and the second branch comprises an efficient channel attention module ECANet; and obtaining a fault diagnosis result of the first time-frequency image output by the first convolutional neural network. The method can accurately perform fault diagnosis on the rolling bearing.

Description

Technical Field

[0001] The present disclosure relates to the technical field of rolling bearing fault diagnosis, and in particular to a rolling bearing fault diagnosis method based on attention time-frequency feature extraction. Background Art

[0002] The vibration signal of rolling bearings is significantly non-stationary and susceptible to interference, which is mainly attributed to the combined influence of the following aspects: First, the impact of the operating environment: The actual working environment of rolling bearings may be very complex, and is significantly affected by external conditions such as temperature changes, humidity fluctuations, dust particles, and mechanical vibrations. Temperature fluctuations may cause the expansion and contraction of materials, thereby affecting the stability of the vibration signal; humidity changes may affect the lubrication effect and increase friction; and the entry of dust particles may directly cause minor internal damage. These factors will affect the statistical characteristics of the vibration signal, making it show obvious non-stationarity in time.

[0003] Second, load fluctuation: During operation, the load borne by the bearing usually changes dynamically with the working conditions of the equipment, such as fluctuations in load size or changes in direction. Load fluctuations will cause changes in the contact force between the rolling element and the raceway, resulting in fluctuations in the amplitude and frequency of the vibration signal. This change is usually manifested as a dynamic adjustment of energy distribution in the frequency domain, increasing the nonlinear characteristics of the signal.

[0004] Third, damage and failure factors: rolling bearings are prone to local damage such as wear, fatigue peeling, and cracks during long-term operation. Such damage will introduce new frequency components or abnormal pulse signals, causing irregular disturbances in the vibration signal. As the damage or failure worsens, the abnormal components in the signal will increase significantly, further exacerbating the non-stationarity of the signal, and it is difficult to effectively describe it using traditional linear models.

[0005] Fourth, the influence of lubrication conditions: Lubrication is one of the key factors affecting the normal operation of bearings. Problems such as insufficient lubricating oil, oil aging, and uneven lubrication will increase bearing friction and generate additional vibration components. These nonlinear vibration signals are superimposed on the signals of normal bearing operation, forming a complex frequency aliasing phenomenon, which increases the susceptibility to interference and uncertainty of the signal.

[0006] Fifth, external mechanical interference: In practical applications, rolling bearings are often in complex systems with multiple devices coexisting. Mechanical interference such as vibration and impact of surrounding equipment will be transmitted to the bearings through the structure. These external interference signals are mixed with the bearing self-vibration signal, making the original vibration signal more complex, forming the characteristics of multi-frequency coexistence and multi-amplitude superposition, further increasing the difficulty of signal analysis and affecting accurate fault diagnosis.

[0007] It can be seen that the vibration signal of the rolling bearing shows significant non-stationarity and susceptibility to interference due to the interaction of multiple factors. Conventional fault diagnosis methods are difficult to accurately diagnose rolling bearing faults based on vibration signals. Summary of the invention

[0008] The present disclosure provides a rolling bearing fault diagnosis method based on attention time-frequency feature extraction, which can accurately diagnose rolling bearing faults. The technical solution at least includes the following solutions: In a first aspect, a rolling bearing fault diagnosis method based on attention time-frequency feature extraction is provided, comprising: obtaining a vibration signal of a rolling bearing; preprocessing the vibration signal to convert the vibration signal into a first time-frequency image; inputting the first time-frequency image into a first convolutional neural network, the first convolutional neural network comprising a first branch, a second branch, and a fully connected layer, the first branch comprising a convolutional attention module CBAM, the second branch comprising an efficient channel attention module ECANet, the first branch being used to extract a first fault feature from a second time-frequency image, the second branch being used to extract a second fault feature from the second time-frequency image, the output of the first branch and the output of the second branch being connected to the fully connected layer, the fully connected layer being used to splice the first fault feature and the second fault feature, and outputting a fault diagnosis result of the second time-frequency image, the second time-frequency image being any time-frequency image input into the first convolutional neural network; obtaining the fault diagnosis result of the first time-frequency image output by the first convolutional neural network.

[0009] Optionally, the CBAM includes a channel attention module and a spatial attention module; the channel attention module is used to process the first feature map to obtain a channel attention feature, and multiply the channel attention feature with the first feature map to obtain a second feature map, and the first feature map is the feature map input into the CBAM; the spatial attention module is used to process the second feature map to obtain a spatial attention feature, and multiply the spatial attention feature with the second feature map to obtain a first fault feature output by the CBAM.

[0010] Optionally, the ECANet includes a global average pooling GPA module, a channel feature adaptive convolution module and an activation function connected in sequence; the GPA module is used to perform global average pooling on the third feature map, and input the third feature map after the global average pooling process into the channel feature adaptive convolution module, and the third feature map is the feature map input into the ECANet; the channel feature adaptive convolution module is used to perform a convolution operation on the third feature map after the global average pooling process to obtain a first convolution result, and input the first convolution result into the activation function to obtain a fourth feature map, and the size of the convolution kernel in the convolution operation is adaptively determined based on the channel characteristics of the third feature map; in the ECANet, the fourth feature map is used to multiply the third feature map to obtain the second fault feature output by the ECANet.

[0011] Optionally, the preprocessing of the vibration signal to convert the vibration signal into a first time-frequency image includes: performing short-time Fourier transform on the vibration signal to obtain a third time-frequency image; and performing bilinear interpolation processing on the third time-frequency image to obtain the first time-frequency image.

[0012] Optionally, the first branch also includes a first convolutional layer and a second convolutional layer, the first convolutional layer, the second convolutional layer, and the CBAM are connected in sequence, and the scale of the convolution kernel in the first convolutional layer is different from the scale of the convolution kernel in the second convolutional layer.

[0013] Optionally, the second branch also includes a third convolutional layer and a fourth convolutional layer, the third convolutional layer, the fourth convolutional layer, and the ECANet are connected in sequence, and the scale of the convolution kernel in the third convolutional layer is different from the scale of the convolution kernel in the fourth convolutional layer.

[0014] In the second aspect, a rolling bearing fault diagnosis device based on attention time-frequency feature extraction is also provided, including: a first acquisition module, used to acquire a vibration signal of a rolling bearing; a preprocessing module, used to preprocess the vibration signal to convert the vibration signal into a first time-frequency image; an input module, used to input the first time-frequency image into a first convolutional neural network, the first convolutional neural network includes a first branch, a second branch, and a fully connected layer, the first branch includes a convolutional attention module CBAM, the second branch includes an efficient channel attention module ECANet, the first branch is used to extract a first fault feature from the second time-frequency image, the second branch is used to extract a second fault feature from the second time-frequency image, the output of the first branch and the output of the second branch are connected to the fully connected layer, the fully connected layer is used to splice the first fault feature and the second fault feature, and output the fault diagnosis result of the second time-frequency image, the second time-frequency image is any time-frequency image input to the first convolutional neural network; a second acquisition module, used to obtain the fault diagnosis result of the first time-frequency image output by the first convolutional neural network.

[0015] Optionally, the preprocessing module is further used to perform short-time Fourier transform on the vibration signal to obtain a third time-frequency image; and perform bilinear interpolation processing on the third time-frequency image to obtain the first time-frequency image.

[0016] In the third aspect, a computer device is also provided, including: a memory and a processor, wherein at least one computer program is stored in the memory, and the at least one computer program is loaded and executed by the processor, so as to execute the rolling bearing fault diagnosis method based on attention time-frequency feature extraction described in the above embodiment.

[0017] In a fourth aspect, a computer-readable storage medium is also provided, in which at least one computer program is stored. The at least one computer program is loaded and executed by a processor, thereby executing the rolling bearing fault diagnosis method based on attention time-frequency feature extraction described in the above embodiment.

[0018] In a fifth aspect, a computer program product is provided, comprising a computer program / instruction, wherein when the computer program / instruction is executed by a processor, the method described in the first aspect is implemented.

[0019] The beneficial effects brought by the technical solution provided by the embodiments of the present disclosure include at least: In the disclosed embodiment, by adding CBAM and ECANet to the first convolutional neural network, and using CBAM and ECANet to process the time-frequency image input into the first convolutional neural network, the first fault feature and the second fault feature are obtained. CBAM and ECANet have strong attention performance, and can extract features in the time-frequency image that are meaningful for fault diagnosis, and filter out the noise in the time-frequency image. In this way, when the first fault feature and the second fault feature are used to diagnose the rolling bearing fault on the first time-frequency image, the accuracy of fault diagnosis can be effectively improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0021] Figure 1 A flowchart of a rolling bearing fault diagnosis method based on attention time-frequency feature extraction provided by an exemplary embodiment of the present disclosure is shown; Figure 2 A flowchart of a rolling bearing fault diagnosis method based on attention time-frequency feature extraction provided by another exemplary embodiment of the present disclosure is shown; Figure 3 It is a schematic diagram of the structure of the first convolutional neural network; Figure 4 It is the structural diagram of CBAM; Figure 5 This is a schematic diagram of the structure of ECANet; Figure 6 A schematic diagram of the structure of a rolling bearing fault diagnosis device based on attention time-frequency feature extraction provided by an exemplary embodiment of the present disclosure is shown; Figure 7 It is a schematic diagram of the structure of a computer device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0022] Unless otherwise defined, the technical or scientific terms used herein shall have the usual meanings understood by persons of ordinary skill in the field to which the present disclosure belongs. The words "first", "second", "third" and similar words used in the patent application specification and claims of the present disclosure do not indicate any order, quantity or importance, but are only used to distinguish different components. Similarly, words such as "one" or "one" do not indicate a quantitative limitation, but rather indicate the presence of at least one. Words such as "include" or "comprise" and similar words mean that the elements or objects appearing before "include" or "comprise" cover the elements or objects listed after "include" or "comprise" and their equivalents, and do not exclude other elements or objects. Words such as "connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect.

[0023] In order to make the objectives, technical solutions and advantages of the present disclosure more clear, the embodiments of the present disclosure will be further described in detail below with reference to the accompanying drawings.

[0024] Figure 1 A flowchart of a rolling bearing fault diagnosis method based on attention time-frequency feature extraction provided by an exemplary embodiment of the present disclosure is shown, and the method can be executed by a computer device. Figure 1 , the method comprising: In step 101, a vibration signal of a rolling bearing is acquired.

[0025] The vibration signal acquired in step 101 is a real-time vibration signal of the rolling bearing, and the rolling bearing fault diagnosis is performed through the vibration signal.

[0026] In step 102, the vibration signal is preprocessed to convert the vibration signal into a first time-frequency image.

[0027] In step 103, the first time-frequency image is input into a first convolutional neural network.

[0028] The first convolutional neural network includes a first branch, a second branch, and a fully connected layer.

[0029] The first branch includes CBAM (Convolutional Block Attention Module), and the second branch includes ECANet (Efficient Channel Attention Networks).

[0030] The first branch is used to extract the first fault feature from the second time-frequency image, and the second branch is used to extract the second fault feature from the second time-frequency image. The output of the first branch and the output of the second branch are connected to the fully connected layer, and the fully connected layer is used to concatenate the first fault feature and the second fault feature, and output the fault diagnosis result of the second time-frequency image. Here, the second time-frequency image is any time-frequency image input to the first convolutional neural network.

[0031] In step 104, a fault diagnosis result of the time-frequency image output by the first convolutional neural network is obtained.

[0032] In the disclosed embodiment, by adding CBAM and ECANet to the first convolutional neural network, and using CBAM and ECANet to process the time-frequency image input into the first convolutional neural network, the first fault feature and the second fault feature are obtained. CBAM and ECANet have strong attention performance, and can extract features in the time-frequency image that are meaningful for fault diagnosis, and filter out the noise in the time-frequency image. In this way, when the first fault feature and the second fault feature are used to diagnose the rolling bearing fault on the first time-frequency image, the accuracy of fault diagnosis can be effectively improved.

[0033] Figure 2 A flowchart of a rolling bearing fault diagnosis method based on attention time-frequency feature extraction provided by another exemplary embodiment of the present disclosure is shown, and the method can be executed by a computer device. Figure 2 , the method comprising: In step 201, a vibration signal of a rolling bearing is acquired.

[0034] The vibration signal acquired in step 201 is a real-time vibration signal of the rolling bearing, and the rolling bearing fault diagnosis is performed through the vibration signal.

[0035] In step 202, a short-time Fourier transform is performed on the vibration signal to obtain a third time-frequency image.

[0036] In step 203, bilinear interpolation processing is performed on the third time-frequency image to obtain a first time-frequency image.

[0037] There are many implementation methods of short-time Fourier transform and bilinear interpolation in the related technologies, and the detailed description is omitted here.

[0038] Through step 203, the size of the third time-frequency image can be reduced.

[0039] In step 204, the first time-frequency image is input into a first convolutional neural network.

[0040] Figure 3 This is a schematic diagram of the structure of the first convolutional neural network. Figure 3The first convolutional neural network is described.

[0041] like Figure 3 As shown, the first convolutional neural network 300 includes: an input layer 301, a first branch 302, a second branch 303, a fully connected layer 304 and an output layer 305.

[0042] The input layer 301 is connected to the first branch 302 and the second branch 303 respectively. The output of the first branch 302 and the output of the second branch 303 are connected to the fully connected layer 304, and the fully connected layer 304 is connected to the output layer 305. The following describes each layer of the first convolutional neural network.

[0043] The input layer 301 is used to input the time-frequency image. Any vibration data needs to be processed in the aforementioned steps 202-203, converted into a time-frequency image, and then input into the first convolutional neural network 300 through the input layer 301.

[0044] Optionally, the first branch 302 includes a first convolutional layer, a second convolutional layer and a CBAM connected in sequence. The scale of the convolution kernel in the first convolutional layer is different from the scale of the convolution kernel in the second convolutional layer.

[0045] Through the first convolution layer and the second convolution layer with different convolution kernel scales, it is possible to extract the fault features hidden in the time-frequency image input to the input layer 301. For example, after the first time-frequency image is input to the first branch, the first convolution layer and the second convolution layer will sequentially perform convolutions of different scales on the first time-frequency image, thereby obtaining the first feature map of the first time-frequency image. Here, the first feature map is the feature map input into the CBAM. The first feature map is the fault feature hidden in the time-frequency image.

[0046] Figure 4 Figure 1 is a schematic diagram of the structure of CBAM. Figure 4 As shown, CBAM includes a channel attention module and a spatial attention module.

[0047] The channel attention module is used to process the first feature map A to obtain the channel attention feature B, and multiply the channel attention feature B with the first feature map A to obtain the second feature map C; the spatial attention module is used to process the second feature map C to obtain the spatial attention feature D, and multiply the spatial attention feature D with the second feature map C to obtain the first fault feature E output by CBAM.

[0048] The process of the channel attention module processing the first feature map A to obtain the channel attention feature B includes: sending the first feature map A to the artificial neural network ANN (Artificial Neural Network) after global maximum pooling and global average pooling respectively. Then, the output of the ANN is weighted and combined, and the channel attention feature B is obtained after the sigmoid activation function.

[0049] The process of the spatial attention module processing the second feature map C to obtain the spatial attention feature D includes: the second feature map C is processed by global maximum pooling and global average pooling in sequence, and then spliced ​​by channel, and then convolution processing and sigmoid activation are performed to obtain the spatial attention feature D.

[0050] By adding CBAM to the first branch, channel attention and spatial attention can be combined to achieve double refinement of the feature map of the input CBAM module. The first fault feature finally output can simultaneously pay attention to which channels and which spatial positions in the first feature map are meaningful, effectively reducing the interference of noise in the vibration signal of the rolling bearing on fault diagnosis.

[0051] After CBAM processes the first feature map, the output first fault feature contains rich feature information, which helps to improve the accuracy of fault identification.

[0052] There are many methods for implementing CBAM in the related art, and detailed description is omitted here.

[0053] Optionally, the second branch 303 includes a third convolutional layer, a fourth convolutional layer and ECANet connected in sequence.

[0054] The scale of the convolution kernel in the third convolution layer is different from the scale of the convolution kernel in the fourth convolution layer. The functions of the third convolution layer and the fourth convolution layer in the second branch 303 are the same as the functions of the first convolution layer and the second convolution layer in the first branch 302, and detailed description is omitted here.

[0055] Figure 5 This is a schematic diagram of the structure of ECANet. Figure 5 As shown in the figure, ECANet includes a global average pooling (GPA) module, a channel feature adaptive convolution module, and an activation function connected in sequence.

[0056] The GPA module is used to perform global average pooling on the third feature map, and input the third feature map after global average pooling into the channel feature adaptive convolution module. Here, the third feature map is the feature map input to ECANet.

[0057] The channel feature adaptive convolution module is used to perform a convolution operation on the third feature map after the global average pooling process to obtain a first convolution result, and input the first convolution result into the activation function to obtain a fourth feature map. In the channel feature adaptive convolution module, the size of the convolution kernel in the convolution operation is adaptively determined based on the channel feature of the third feature map.

[0058] In ECANet, the fourth feature map is used to multiply the third feature map to obtain the second fault feature output by ECANet.

[0059] By adding ECANet to the second branch, the coverage of local cross-channel interactions in the convolution process can be improved, so that the final second fault feature can not only maintain the dimension of the original third feature map, but also have better channel attention performance (it can pay attention to which channels in the third feature map are meaningful), thereby effectively reducing the interference of noise in the vibration signal of the rolling bearing on fault diagnosis.

[0060] There are many methods for implementing ECANet in related technologies, so detailed description is omitted here.

[0061] In the disclosed embodiment, by adding CBAM and ECANet to the first convolutional neural network, and using CBAM and ECANet to process the time-frequency image input to the first convolutional neural network respectively, the first fault feature and the second fault feature finally obtained can effectively filter the noise in the time-frequency image. In this way, when the first fault feature and the second fault feature are used to diagnose the rolling bearing fault, the accuracy of the fault diagnosis can be effectively improved.

[0062] In addition, since both CBAM and ECANet are lightweight modules, they occupy fewer computing resources compared to conventional attention modules, thereby reducing the computational complexity of rolling bearing fault diagnosis and achieving accurate fault diagnosis at a lower computing cost.

[0063] The fully connected layer 304 is used to concatenate the first fault feature and the second fault feature, and output the fault diagnosis result of the second time-frequency image. Here, the second time-frequency image is any time-frequency image input to the first convolutional neural network.

[0064] The output layer 305 is used to output the final fault diagnosis result.

[0065] After the first branch 302 and the second branch 303 extract the first fault feature and the second fault feature respectively, it is equivalent to having achieved the extraction of sufficiently rich and detailed fault-related features from the time-frequency image. At this time, the first fault feature and the second fault feature are sequentially input into the fully connected layer 304 and the output layer 305 to realize fault diagnosis of the time-frequency image input into the first convolutional neural network.

[0066] Before using the first convolutional neural network to perform fault diagnosis on the first time-frequency image, it is necessary to train the first convolutional neural network. The method includes the following four steps: In the first step, multiple historical vibration data of rolling bearings are obtained, and the multiple historical vibration data are divided into a training set, a validation set, and a test set.

[0067] The multiple historical vibration data also need to be preprocessed, which includes converting the historical vibration data into a time-frequency image and adding a label to each historical vibration data. For example, the method in steps 202-203 can be used to convert the historical vibration data into a time-frequency image.

[0068] Here, the historical vibration data is converted into a time-frequency image by Fourier transform and bilinear interpolation, which can reduce the size of the time-frequency image and thus reduce the training time of the first convolutional neural network.

[0069] In the second step, the first convolutional neural network is trained using the data in the training set.

[0070] In the third step, the time-frequency images in the test set are used to test the performance of the first convolutional neural network.

[0071] If the performance of the first convolutional neural network does not meet the requirements, execute step 4. If the performance of the first convolutional neural network meets the requirements, it means that the first convolutional neural network has been trained.

[0072] The fourth step is to fine-tune the first convolutional neural network using the data in the validation set.

[0073] After executing step 4, you need to jump to step 3.

[0074] Through the above steps 1 to 4, the first convolutional neural network can be trained.

[0075] In step 204, a fault diagnosis result of a first time-frequency image output by a first convolutional neural network is obtained.

[0076] After the first convolutional neural network training is completed, the first time-frequency image is input into the first convolutional neural network, and the first convolutional neural network outputs the fault diagnosis result of the first time-frequency image.

[0077] The following are device embodiments of the present application. For details not described in detail in the device embodiments, reference may be made to the above method embodiments.

[0078] Figure 6 FIG. 1 is a schematic diagram showing a rolling bearing fault diagnosis device based on attention time-frequency feature extraction provided by an exemplary embodiment of the present disclosure. Figure 6The rolling bearing fault diagnosis device 600 based on attention time-frequency feature extraction includes: a first acquisition module 601, a preprocessing module 602, an input module 603 and a second acquisition module 604.

[0079] The first acquisition module 601 is used to acquire the vibration signal of the rolling bearing.

[0080] The preprocessing module 602 is used to preprocess the vibration signal to convert the vibration signal into a first time-frequency image.

[0081] The input module 603 is used to input the first time-frequency image into the first convolutional neural network. The first convolutional neural network includes a first branch, a second branch, and a fully connected layer. The first branch includes a convolutional attention module CBAM, and the second branch includes an efficient channel attention module ECANet. The first branch is used to extract the first fault feature from the second time-frequency image, and the second branch is used to extract the second fault feature from the second time-frequency image. The output of the first branch and the output of the second branch are connected to the fully connected layer. The fully connected layer is used to splice the first fault feature and the second fault feature, and output the fault diagnosis result of the second time-frequency image. The second time-frequency image is any time-frequency image input into the first convolutional neural network.

[0082] The second acquisition module 604 is used to obtain the fault diagnosis result of the first time-frequency image output by the first convolutional neural network.

[0083] Optionally, the preprocessing module 602 is further configured to perform short-time Fourier transform on the vibration signal to obtain a third time-frequency image; and perform bilinear interpolation processing on the third time-frequency image to obtain the first time-frequency image.

[0084] It should be noted that: when the rolling bearing fault diagnosis device based on attention time-frequency feature extraction provided in the above embodiment performs rolling bearing fault diagnosis, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the rolling bearing fault diagnosis device based on attention time-frequency feature extraction provided in the above embodiment and the rolling bearing fault diagnosis method embodiment based on attention time-frequency feature extraction belong to the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0085] The division of modules in the embodiments of the present disclosure is schematic and is only a logical function division. There may be other division methods in actual implementation. In addition, each functional module in each embodiment of the present disclosure may be integrated into a processor, or may exist physically separately, or two or more modules may be integrated into one module. The above-mentioned integrated modules may be implemented in the form of hardware or in the form of software functional modules.

[0086] If the integrated module is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present disclosure is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a terminal device (which can be a personal computer, mobile phone, or communication device, etc.) or a processor (processor) to execute all or part of the steps of the method of each embodiment of the present disclosure. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0087] Figure 7 Schematic diagram of the structure of the computer device provided by the embodiment of the present disclosure. Figure 7 As shown, the computer device 700 includes: a processor 701 and a memory 702 .

[0088] The processor 701 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 701 may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 701 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 701 may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 701 may also include an AI (Artificial Intelligence) processor, which is used to process computing operations related to machine learning.

[0089] The memory 702 may include one or more computer-readable storage media, which may be non-transitory. The memory 702 may also include a high-speed random access memory, and a non-volatile memory, such as one or more disk storage devices, flash memory storage devices. In some embodiments, the non-transitory computer-readable storage medium in the memory 702 is used to store at least one instruction, which is used to be executed by the processor 701 to implement the rolling bearing fault diagnosis method based on attention time-frequency feature extraction provided in the embodiments of the present disclosure.

[0090] Those skilled in the art will understand that Figure 7 The structure shown in the figure does not constitute a limitation on the computer device 700, and the computer device 700 may include more or less components than shown in the figure, or combine some components, or adopt a different arrangement of components.

[0091] The embodiments of the present disclosure also provide a non-temporary computer-readable storage medium. When the instructions in the storage medium are executed by the processor of a computer device, the computer device is enabled to execute the rolling bearing fault diagnosis method based on attention time-frequency feature extraction provided in the embodiments of the present disclosure.

[0092] The embodiments of the present disclosure also provide a computer program product, including a computer program / instruction, which, when executed by a processor, implements the rolling bearing fault diagnosis method based on attention time-frequency feature extraction provided in the embodiments of the present disclosure.

[0093] The above description is only an optional embodiment of the present disclosure and is not intended to limit the present disclosure. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A rolling bearing fault diagnosis method based on attention time-frequency feature extraction, characterized in that: The method comprises: Obtain vibration signals of rolling bearings; Preprocessing the vibration signal to convert the vibration signal into a first time-frequency image; Input the first time-frequency image into a first convolutional neural network, the first convolutional neural network includes a first branch, a second branch, and a fully connected layer, the first branch includes a convolutional attention module CBAM, the second branch includes an efficient channel attention module ECANet, the first branch is used to extract a first fault feature from the second time-frequency image, the second branch is used to extract a second fault feature from the second time-frequency image, the output of the first branch and the output of the second branch are connected to the fully connected layer, the fully connected layer is used to splice the first fault feature and the second fault feature, and output a fault diagnosis result of the second time-frequency image, the second time-frequency image is any time-frequency image input to the first convolutional neural network; Obtain a fault diagnosis result of the first time-frequency image output by the first convolutional neural network.

2. The method according to claim 1, characterized in that The CBAM includes a channel attention module and a spatial attention module; The channel attention module is used to process the first feature map to obtain a channel attention feature, and multiply the channel attention feature with the first feature map to obtain a second feature map, where the first feature map is the feature map input into the CBAM; The spatial attention module is used to process the second feature map to obtain a spatial attention feature, and multiply the spatial attention feature with the second feature map to obtain a first fault feature output by the CBAM.

3. The method according to claim 1, characterized in that The ECANet includes a global average pooling GPA module, a channel feature adaptive convolution module and an activation function connected in sequence; The GPA module is used to perform global average pooling processing on the third feature map, and input the third feature map after the global average pooling processing into the channel feature adaptive convolution module, and the third feature map is the feature map input into the ECANet; The channel feature adaptive convolution module is used to perform a convolution operation on the third feature map after the global average pooling process to obtain a first convolution result, and input the first convolution result into an activation function to obtain a fourth feature map, wherein the size of the convolution kernel in the convolution operation is adaptively determined based on the channel feature of the third feature map; In the ECANet, the fourth characteristic map is used to multiply the third characteristic map to obtain a second fault feature output by the ECANet.

4. The method according to any one of claims 1 to 3, characterized in that: The preprocessing of the vibration signal to convert the vibration signal into a first time-frequency image includes: Performing short-time Fourier transform on the vibration signal to obtain a third time-frequency image; Perform bilinear interpolation processing on the third time-frequency image to obtain the first time-frequency image.

5. The method according to any one of claims 1 to 3, characterized in that: The first branch also includes a first convolutional layer and a second convolutional layer, the first convolutional layer, the second convolutional layer, and the CBAM are connected in sequence, and the scale of the convolution kernel in the first convolutional layer is different from the scale of the convolution kernel in the second convolutional layer.

6. The method according to any one of claims 1 to 3, characterized in that: The second branch also includes a third convolutional layer and a fourth convolutional layer, the third convolutional layer, the fourth convolutional layer, and the ECANet are connected in sequence, and the scale of the convolution kernel in the third convolutional layer is different from the scale of the convolution kernel in the fourth convolutional layer.

7. A rolling bearing fault diagnosis device based on attention time-frequency feature extraction, characterized in that: The device comprises: A first acquisition module, used to acquire a vibration signal of a rolling bearing; A preprocessing module, used for preprocessing the vibration signal to convert the vibration signal into a first time-frequency image; An input module, used to input the first time-frequency image into a first convolutional neural network, the first convolutional neural network includes a first branch, a second branch, and a fully connected layer, the first branch includes a convolutional attention module CBAM, the second branch includes an efficient channel attention module ECANet, the first branch is used to extract a first fault feature from the second time-frequency image, the second branch is used to extract a second fault feature from the second time-frequency image, the output of the first branch and the output of the second branch are connected to the fully connected layer, the fully connected layer is used to splice the first fault feature and the second fault feature, and output a fault diagnosis result of the second time-frequency image, the second time-frequency image is any time-frequency image input to the first convolutional neural network; The second acquisition module is used to obtain the fault diagnosis result of the first time-frequency image output by the first convolutional neural network.

8. A computer device, characterized in that: The computer device comprises: a memory and a processor, wherein at least one computer program is stored in the memory, and the at least one computer program is loaded and executed by the processor to implement the method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores at least one computer program, and the at least one computer program is loaded and executed by a processor to implement the method according to any one of claims 1 to 6.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images