Bearing fault diagnosis method and system

Abstract

The invention discloses a bearing fault diagnosis method. The method comprises the following steps: acquiring existing bearing working data information; converting the acquired data information into a two-dimensional time-frequency image and constructing a training data set; constructing a bearing fault diagnosis primary model and training to obtain a bearing fault diagnosis model; and working data of the target bearing are acquired in real time, and fault diagnosis of the target bearing is carried out by using the bearing fault diagnosis model. The invention also discloses a system for realizing the bearing fault diagnosis method. According to the bearing fault diagnosis method, the bearing fault diagnosis model is constructed based on the convolutional network and the attention mechanism, and the model is trained by adopting the existing bearing data, so that the bearing fault diagnosis can be realized, the reliability is higher, and the accuracy is better.

Description

Technical Field

[0001] The invention belongs to the field of electrical automation, and specifically designs a bearing fault diagnosis method and system. Background Art

[0002] With the development of economy and technology and the improvement of people's living standards, electricity has become an indispensable secondary energy source in people's production and life, bringing endless convenience to people's production and life. Therefore, ensuring the stable and reliable supply of electricity has become one of the most important tasks of the power system.

[0003] Distribution network poles are important components of the power system. During the mechanized assembly of distribution network poles, the operating status of the bearings determines the operating reliability of the rotating mechanical equipment; therefore, effective detection and diagnosis of the operating status of the bearings can effectively avoid downtime accidents, thereby ensuring the effectiveness and reliability of the distribution network poles, and ultimately ensuring the stable and reliable operation of the power system.

[0004] At present, the bearing fault diagnosis scheme for distribution network pole assembly machinery generally uses time-frequency signal processing technology to extract fault features, and uses one-dimensional signals or converts one-dimensional signals into grayscale images and inputs them into convolutional neural network models for fault classification. However, this type of scheme is highly complex and has poor reliability and accuracy. Summary of the invention

[0005] One of the purposes of the present invention is to provide a bearing fault diagnosis method with high reliability and good accuracy.

[0006] A second objective of the present invention is to provide a system for implementing the bearing fault diagnosis method.

[0007] The bearing fault diagnosis method provided by the present invention comprises the following steps:

[0008] S1. Obtain existing bearing working data information;

[0009] S2. Based on wavelet transform, the data information obtained in step S1 is converted into a two-dimensional time-frequency image, and a training data set is constructed;

[0010] S3. Based on convolutional network and attention mechanism, a primary model for bearing fault diagnosis is constructed;

[0011] S4. Using the training data set obtained in step S2, the primary bearing fault diagnosis model constructed in step S3 is trained to obtain a bearing fault diagnosis model;

[0012] S5. Acquire the working data of the target bearing in real time, and use the bearing fault diagnosis model obtained in step S4 to perform fault diagnosis on the target bearing.

[0013] The step S2 converts the data information obtained in step S1 into a two-dimensional time-frequency image based on wavelet transform, which specifically includes the following steps:

[0014] Using continuous wavelet transform, convert the data information obtained in step S1 into a two-dimensional time-frequency image;

[0015] The continuous wavelet transform is performed using the following formula:

[0016]

[0017] Where CWT f (a, b) are the coefficients of continuous wavelet transform; a is the scale factor; b is the translation factor; f(t) is the original vibration signal of the bearing at time t; is the complex conjugate function of the wavelet function ψ(t).

[0018] The step S2 of constructing a training data set specifically includes the following steps:

[0019] In the constructed dataset, each data is marked, and the marks are divided into 9 fault states and normal states; among them, the 9 fault states include rolling element fault with a fault size of 0.1778mm, bearing inner ring fault with a fault size of 0.1778mm, bearing outer ring fault with a fault size of 0.1778mm, rolling element fault with a fault size of 0.3556mm, bearing inner ring fault with a fault size of 0.3556mm, bearing outer ring fault with a fault size of 0.3556mm, rolling element fault with a fault size of 0.5334mm, bearing inner ring fault with a fault size of 0.5334mm and bearing outer ring fault with a fault size of 0.5334mm.

[0020] The primary model for bearing fault diagnosis is constructed based on the convolutional network and the attention mechanism described in step S3, which specifically includes the following steps:

[0021] The constructed primary model for bearing fault diagnosis includes feature extraction module, channel attention module, spatial attention module, aggregation module and classification module;

[0022] The feature extraction module, channel attention module, spatial attention module, aggregation module and classification module are connected in series in sequence;

[0023] A feature extraction module is constructed based on the convolution layer; the feature extraction module is used to extract the features of the input bearing data;

[0024] A channel attention module is constructed based on the attention mechanism; the channel attention module is used to extract higher-level features based on the features of the obtained bearing data;

[0025] A spatial attention module is constructed based on the attention mechanism; the spatial attention module is used to generate a spatial attention feature map according to the output of the channel attention module;

[0026] The aggregation module is used to fuse the obtained features;

[0027] The classification module is used to classify the fused features to realize the diagnosis of bearing faults.

[0028] The feature extraction module specifically includes the following steps:

[0029] The feature extraction module includes three branches; the input bearing data is processed through the three branches at the same time;

[0030] The first branch includes a first branch first convolution layer, a first branch first normalization layer, a first branch first activation function layer, a first branch second convolution layer, a first branch second normalization layer and a first branch second activation function layer, which are connected in series in sequence; wherein the convolution kernel of the first branch first convolution layer is 1×1, the step size is 1, the padding is 1, the input channel is 3, and the output channel is 32; the convolution kernel of the first branch second convolution layer is 4×4, the step size is 1, the padding is 1, the input channel is 32, and the output channel is 32;

[0031] The second branch includes the second branch first convolution layer, the second branch first normalization layer, the second branch first activation function layer, the second branch second convolution layer, the second branch second normalization layer, the second branch second activation function layer, the second branch third convolution layer, the second branch third normalization layer, the second branch third activation function layer, the second branch fourth convolution layer, the second branch fourth normalization layer, the second branch fourth activation function layer, the second branch fifth convolution layer, the second branch fifth normalization layer and the second branch fifth activation function layer, which are connected in series in sequence; wherein the convolution kernel of the second branch first convolution layer is 3×3, and the step length is The convolution kernel of the second convolution layer of the second branch is 3×3, the stride is 2, the padding is 1, the input channel is 3, and the output channel is 16; the convolution kernel of the third convolution layer of the second branch is 3×3, the stride is 1, the padding is 1, the input channel is 16, and the output channel is 32; the convolution kernel of the fourth convolution layer of the second branch is 3×3, the stride is 2, the padding is 1, the input channel is 32, and the output channel is 32; the convolution kernel of the fifth convolution layer of the second branch is 3×3, the stride is 1, the padding is 1, the input channel is 32, and the output channel is 64;

[0032] The third branch includes the third branch first convolution layer, the third branch first normalization layer, the third branch first activation function layer, the third branch second convolution layer, the third branch second normalization layer, the third branch second activation function layer, the third branch third convolution layer, the third branch first normalization layer, the third branch third activation function layer, the third branch fourth convolution layer, the third branch fourth normalization layer, the third branch fourth activation function layer, the third branch fifth convolution layer, the third branch fifth normalization layer and the third branch fifth activation function layer, which are connected in series in sequence; wherein, the convolution kernel of the third branch first convolution layer is 5×5, and the step length is The convolution kernel of the second convolution layer of the third branch is 5×5, the stride is 2, the padding is 2, the input channel is 3, and the output channel is 16; the convolution kernel of the third convolution layer of the third branch is 5×5, the stride is 1, the padding is 2, the input channel is 16, and the output channel is 16; the convolution kernel of the third convolution layer of the third branch is 5×5, the stride is 1, the padding is 2, the input channel is 16, and the output channel is 32; the convolution kernel of the fourth convolution layer of the third branch is 5×5, the stride is 2, the padding is 2, the input channel is 32, and the output channel is 32; the convolution kernel of the fifth convolution layer of the third branch is 5×5, the stride is 1, the padding is 2, the input channel is 32, and the output channel is 64.

[0033] The channel attention module specifically includes the following steps:

[0034] The features of the input channel attention module are divided into two paths: the first path is processed by the maximum pooling layer and the first convolution layer in turn to obtain the first channel features; the second path is processed by the average pooling layer and the second convolution layer in turn to obtain the second channel features; then the first channel features and the second channel features are added and processed by the activation function;

[0035] The processing process of the channel attention module is expressed as

[0036] Mc(F)=σ(f 1 × 1 (AvgPool(F))+f 1 × 1 (MaxPool(F)))

[0037] Where M c (F) is the output feature of the channel attention module; σ() is the Sigmoid activation function; f 1×1 () is the convolution layer processing function with a convolution kernel of 1×1; AvgPool() is the average pooling layer processing function; F is the input feature; MaxPool() is the maximum pooling layer processing function.

[0038] The spatial attention module specifically includes the following steps:

[0039] The features of the input spatial attention module are divided into two paths: the first path is processed by the maximum pooling layer in sequence to obtain the first spatial feature; the second path is processed by the average pooling layer to obtain the second spatial feature; then the first spatial feature and the second spatial feature are cascaded and processed by the convolution layer and activation function in sequence;

[0040] The processing of the spatial attention module is expressed as

[0041] Ms(F)=σ(f 7 × 7 ([AvgPool(F); MaxPool(F)]))

[0042] Where σ() is the Sigmoid activation function; f 7×7 () is the convolution layer processing function with a convolution kernel of 7×7; AvgPool() is the average pooling layer processing function; F is the input feature; MaxPool() is the maximum pooling layer processing function.

[0043] The aggregation module specifically includes the following steps:

[0044] A linear network is used as the aggregation module.

[0045] The classification module specifically includes the following steps:

[0046] A fully connected layer is used as the classification module.

[0047] The training described in step S4 specifically includes the following steps:

[0048] The cross entropy loss function is used as the total loss function;

[0049] The stochastic gradient descent algorithm was used to train the model, with the learning rate set to 0.01 and the number of iterations set to 100.

[0050] The present invention also provides a system for implementing the bearing fault diagnosis method, comprising a data acquisition module, a data processing module, a model construction module, a model training module and a fault diagnosis module; the data acquisition module, the data processing module, the model construction module, the model training module and the fault diagnosis module are connected in series in sequence; the data acquisition module is used to acquire the existing bearing working data information, and upload the data information to the data processing module; the data processing module is used to convert the acquired data information into a two-dimensional time-frequency image based on the received data information and based on the wavelet transform, and to construct a training data set, and to upload the data information to the model construction module; the model construction module is used to construct a primary model for bearing fault diagnosis based on the received data information and based on the convolutional network and the attention mechanism, and to upload the data information to the model training module; the model training module is used to train the constructed primary model for bearing fault diagnosis based on the received data information and the obtained training data set, to obtain a bearing fault diagnosis model, and to upload the data information to the fault diagnosis module; the fault diagnosis module is used to acquire the working data of the target bearing in real time based on the received data information, and to perform fault diagnosis of the target bearing using the obtained bearing fault diagnosis model.

[0051] The bearing fault diagnosis method and system provided by the present invention construct a bearing fault diagnosis model based on a convolutional network and an attention mechanism, and use existing bearing data to train the model. Therefore, the present invention can not only realize the diagnosis of bearing faults, but also has higher reliability and better accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 The figure is a schematic diagram of the method flow of the present invention.

[0053] Figure 2 Schematic diagram of the functional modules of the system of the present invention. DETAILED DESCRIPTION

[0054] like Figure 1 The method flow chart of the method of the present invention is shown as follows: The bearing fault diagnosis method disclosed in the present invention comprises the following steps:

[0055] S1. Obtain existing bearing working data information;

[0056] S2. Based on wavelet transform, the data information obtained in step S1 is converted into a two-dimensional time-frequency image, and a training data set is constructed; specifically, the steps include:

[0057] Using continuous wavelet transform, convert the data information obtained in step S1 into a two-dimensional time-frequency image;

[0058] The continuous wavelet transform is performed using the following formula:

[0059]

[0060] Where CWT f (a, b) are the coefficients of continuous wavelet transform; a is the scale factor; b is the translation factor; f(t) is the original vibration signal of the bearing at time t; is the complex conjugate function of the wavelet function ψ(t);

[0061] At the same time, in the constructed data set, each data is marked, and the marks are divided into 9 fault states and normal states; among them, the 9 fault states include rolling element fault with a fault size of 0.1778mm, bearing inner ring fault with a fault size of 0.1778mm, bearing outer ring fault with a fault size of 0.1778mm, rolling element fault with a fault size of 0.3556mm, bearing inner ring fault with a fault size of 0.3556mm, bearing outer ring fault with a fault size of 0.3556mm, rolling element fault with a fault size of 0.5334mm, bearing inner ring fault with a fault size of 0.5334mm and bearing outer ring fault with a fault size of 0.5334mm;

[0062] S3. Based on the convolutional network and attention mechanism, a primary model for bearing fault diagnosis is constructed; specifically, the following steps are included:

[0063] The constructed primary model for bearing fault diagnosis includes feature extraction module, channel attention module, spatial attention module, aggregation module and classification module;

[0064] The feature extraction module, channel attention module, spatial attention module, aggregation module and classification module are connected in series in sequence;

[0065] A feature extraction module is constructed based on the convolution layer; the feature extraction module is used to extract the features of the input bearing data;

[0066] A channel attention module is constructed based on the attention mechanism; the channel attention module is used to extract higher-level features based on the features of the obtained bearing data;

[0067] A spatial attention module is constructed based on the attention mechanism; the spatial attention module is used to generate a spatial attention feature map according to the output of the channel attention module;

[0068] The aggregation module is used to fuse the obtained features;

[0069] The classification module is used to classify the fused features to realize the diagnosis of bearing faults;

[0070] In specific implementation, the feature extraction module includes three branches; the input bearing data is processed through the three branches at the same time;

[0071] The first branch includes a first branch first convolution layer, a first branch first normalization layer, a first branch first activation function layer, a first branch second convolution layer, a first branch second normalization layer and a first branch second activation function layer, which are connected in series in sequence; wherein the convolution kernel of the first branch first convolution layer is 1×1, the step size is 1, the padding is 1, the input channel is 3, and the output channel is 32; the convolution kernel of the first branch second convolution layer is 4×4, the step size is 1, the padding is 1, the input channel is 32, and the output channel is 32;

[0072] The second branch includes the second branch first convolution layer, the second branch first normalization layer, the second branch first activation function layer, the second branch second convolution layer, the second branch second normalization layer, the second branch second activation function layer, the second branch third convolution layer, the second branch first normalization layer, the second branch third activation function layer, the second branch fourth convolution layer, the second branch fourth normalization layer, the second branch fourth activation function layer, the second branch fifth convolution layer, the second branch fifth normalization layer and the second branch fifth activation function layer, which are connected in series in sequence; wherein the convolution kernel of the second branch first convolution layer is 3×3, and the step length is The convolution kernel of the second convolution layer of the second branch is 3×3, the stride is 2, the padding is 1, the input channel is 3, and the output channel is 16; the convolution kernel of the third convolution layer of the second branch is 3×3, the stride is 1, the padding is 1, the input channel is 16, and the output channel is 32; the convolution kernel of the fourth convolution layer of the second branch is 3×3, the stride is 2, the padding is 1, the input channel is 32, and the output channel is 32; the convolution kernel of the fifth convolution layer of the second branch is 3×3, the stride is 1, the padding is 1, the input channel is 32, and the output channel is 64;

[0073] The third branch includes a third branch first convolution layer, a third branch first normalization layer, a third branch first activation function layer, a third branch second convolution layer, a third branch second normalization layer, a third branch second activation function layer, a third branch third convolution layer, a third branch third normalization layer, a third branch third activation function layer, a third branch fourth convolution layer, a third branch fourth normalization layer, a third branch fourth activation function layer, a third branch fifth convolution layer, a third branch fifth normalization layer and a third branch fifth activation function layer, which are connected in series in sequence; wherein, the convolution kernel of the first convolution layer of the third branch is 5×5, and the step length is The convolution kernel of the second convolution layer of the third branch is 5×5, the stride is 2, the padding is 2, the input channel is 3, and the output channel is 16; the convolution kernel of the third convolution layer of the third branch is 5×5, the stride is 1, the padding is 2, the input channel is 16, and the output channel is 16; the convolution kernel of the third convolution layer of the third branch is 5×5, the stride is 1, the padding is 2, the input channel is 16, and the output channel is 32; the convolution kernel of the fourth convolution layer of the third branch is 5×5, the stride is 2, the padding is 2, the input channel is 32, and the output channel is 32; the convolution kernel of the fifth convolution layer of the third branch is 5×5, the stride is 1, the padding is 2, the input channel is 32, and the output channel is 64;

[0074] Among them, the processing function of the convolution layer is expressed as

[0075] Cl=f(∑i(Wi*Fl-1,i)+b)

[0076] Where C l is the convolution feature output of the lth layer; f() is the activation function; W i is the feature map F l-1,i The corresponding weight item; F l-1,i is the i-th feature map of the l-1th layer; b is the bias value, which is used to adjust the output of the activation function;

[0077] The normalization layer is used to prevent gradient explosion and convert the data to a state with a mean of 0 and a variance of 1 to speed up convergence;

[0078] The features of the input channel attention module are divided into two paths: the first path is processed by the maximum pooling layer and the first convolution layer in turn to obtain the first channel features; the second path is processed by the average pooling layer and the second convolution layer in turn to obtain the second channel features; then the first channel features and the second channel features are added and processed by the activation function;

[0079] The processing process of the channel attention module is expressed as

[0080] Mc(F)=σ(f 1 × 1 (AvgPool(F))+f 1 × 1 (MaxPool(F)))

[0081] Where M c (F) is the output feature of the channel attention module; σ() is the Sigmoid activation function; f 1×1 () is the convolution layer processing function with a convolution kernel of 1×1; AvgPool() is the average pooling layer processing function; F is the input feature; MaxPool() is the maximum pooling layer processing function;

[0082] The features of the input spatial attention module are divided into two paths: the first path is processed by the maximum pooling layer in sequence to obtain the first spatial feature; the second path is processed by the average pooling layer to obtain the second spatial feature; then the first spatial feature and the second spatial feature are cascaded and processed by the convolution layer and activation function in sequence;

[0083] The processing of the spatial attention module is expressed as

[0084] Ms(F)=σ(f 7 × 7 ([AvgPool(F); MaxPool(F)]))

[0085] Where σ() is the Sigmoid activation function; f 7×7 () is the convolution layer processing function with a convolution kernel of 7×7; AvgPool() is the average pooling layer processing function; F is the input feature; MaxPool() is the maximum pooling layer processing function;

[0086] Among them, global maximum pooling and global average pooling operations are used to capture different types of statistical features;

[0087] Linear network is used as aggregation module;

[0088] Use fully connected layer as classification module;

[0089] S4. Using the training data set obtained in step S2, the primary bearing fault diagnosis model constructed in step S3 is trained to obtain a bearing fault diagnosis model;

[0090] During training, the cross entropy loss function is used as the total loss function;

[0091] The model was trained using the stochastic gradient descent algorithm, with the learning rate set to 0.01 and the number of iterations set to 100;

[0092] S5. Acquire the working data of the target bearing in real time, and use the bearing fault diagnosis model obtained in step S4 to perform fault diagnosis on the target bearing.

[0093] like Figure 2The figure shows a schematic diagram of the functional modules of the system of the present invention: the system for implementing the bearing fault diagnosis method disclosed in the present invention comprises a data acquisition module, a data processing module, a model construction module, a model training module and a fault diagnosis module; the data acquisition module, the data processing module, the model construction module, the model training module and the fault diagnosis module are connected in series in sequence; the data acquisition module is used to acquire the existing bearing working data information, and upload the data information to the data processing module; the data processing module is used to convert the acquired data information into a two-dimensional time-frequency image based on the received data information and wavelet transform, and construct a training data set, and upload the data information to the model construction module; the model construction module is used to construct a primary model for bearing fault diagnosis based on the received data information and convolutional network and attention mechanism, and upload the data information to the model training module; the model training module is used to train the constructed primary model for bearing fault diagnosis based on the received data information and the obtained training data set, obtain the bearing fault diagnosis model, and upload the data information to the fault diagnosis module; the fault diagnosis module is used to acquire the working data of the target bearing in real time based on the received data information, and use the obtained bearing fault diagnosis model to perform fault diagnosis on the target bearing.

Claims

1. A bearing fault diagnosis method comprising the following steps: S1. Obtain existing bearing working data information; S2. Based on wavelet transform, the data information obtained in step S1 is converted into a two-dimensional time-frequency image, and a training data set is constructed; S3. Based on convolutional network and attention mechanism, a primary model for bearing fault diagnosis is constructed; S4. Using the training data set obtained in step S2, the primary bearing fault diagnosis model constructed in step S3 is trained to obtain a bearing fault diagnosis model; S5. Acquire the working data of the target bearing in real time, and use the bearing fault diagnosis model obtained in step S4 to perform fault diagnosis on the target bearing.

2. The bearing fault diagnosis method according to claim 1, characterized in that The step S2 converts the data information obtained in step S1 into a two-dimensional time-frequency image based on wavelet transform, which specifically includes the following steps: Using continuous wavelet transform, convert the data information obtained in step S1 into a two-dimensional time-frequency image; The continuous wavelet transform is performed using the following formula: Where CWT f (a, b) are the coefficients of continuous wavelet transform; a is the scale factor; b is the translation factor; f(t) is the original vibration signal of the bearing at time t; is the complex conjugate function of the wavelet function ψ(t).

3. The bearing fault diagnosis method according to claim 2 is characterized in that The step S2 of constructing a training data set specifically includes the following steps: In the constructed dataset, each data is marked, and the marks are divided into 9 fault states and normal states; among them, the 9 fault states include rolling element fault with a fault size of 0.1778mm, bearing inner ring fault with a fault size of 0.1778mm, bearing outer ring fault with a fault size of 0.1778mm, rolling element fault with a fault size of 0.3556mm, bearing inner ring fault with a fault size of 0.3556mm, bearing outer ring fault with a fault size of 0.3556mm, rolling element fault with a fault size of 0.5334mm, bearing inner ring fault with a fault size of 0.5334mm and bearing outer ring fault with a fault size of 0.5334mm.

4. The bearing fault diagnosis method according to claim 3 is characterized in that The primary model for bearing fault diagnosis is constructed based on the convolutional network and the attention mechanism described in step S3, which specifically includes the following steps: The constructed primary model for bearing fault diagnosis includes feature extraction module, channel attention module, spatial attention module, aggregation module and classification module; The feature extraction module, channel attention module, spatial attention module, aggregation module and classification module are connected in series in sequence; Build a feature extraction module based on the convolutional layer; The feature extraction module is used to extract the features of the input bearing data; A channel attention module is constructed based on the attention mechanism; the channel attention module is used to extract higher-level features based on the features of the obtained bearing data; A spatial attention module is constructed based on the attention mechanism; the spatial attention module is used to generate a spatial attention feature map according to the output of the channel attention module; The aggregation module is used to fuse the obtained features; The classification module is used to classify the fused features to realize the diagnosis of bearing faults.

5. The bearing fault diagnosis method according to claim 4, characterized in that The feature extraction module specifically includes the following steps: The feature extraction module includes three branches; the input bearing data is processed through the three branches at the same time; The first branch includes a first branch first convolution layer, a first branch first normalization layer, a first branch first activation function layer, a first branch second convolution layer, a first branch second normalization layer and a first branch second activation function layer, which are connected in series in sequence; wherein the convolution kernel of the first branch first convolution layer is 1×1, the step size is 1, the padding is 1, the input channel is 3, and the output channel is 32; the convolution kernel of the first branch second convolution layer is 4×4, the step size is 1, the padding is 1, the input channel is 32, and the output channel is 32; The second branch includes the second branch first convolution layer, the second branch first normalization layer, the second branch first activation function layer, the second branch second convolution layer, the second branch second normalization layer, the second branch second activation function layer, the second branch third convolution layer, the second branch third normalization layer, the second branch third activation function layer, the second branch fourth convolution layer, the second branch fourth normalization layer, the second branch fourth activation function layer, the second branch fifth convolution layer, the second branch fifth normalization layer and the second branch fifth activation function layer, which are connected in series in sequence; wherein the convolution kernel of the second branch first convolution layer is 3×3, and the step length is The convolution kernel of the second convolution layer of the second branch is 3×3, the stride is 2, the padding is 1, the input channel is 3, and the output channel is 16; the convolution kernel of the third convolution layer of the second branch is 3×3, the stride is 1, the padding is 1, the input channel is 16, and the output channel is 32; the convolution kernel of the fourth convolution layer of the second branch is 3×3, the stride is 2, the padding is 1, the input channel is 32, and the output channel is 32; the convolution kernel of the fifth convolution layer of the second branch is 3×3, the stride is 1, the padding is 1, the input channel is 32, and the output channel is 64; The third branch includes the third branch first convolution layer, the third branch first normalization layer, the third branch first activation function layer, the third branch second convolution layer, the third branch second normalization layer, the third branch second activation function layer, the third branch third convolution layer, the third branch first normalization layer, the third branch third activation function layer, the third branch fourth convolution layer, the third branch fourth normalization layer, the third branch fourth activation function layer, the third branch fifth convolution layer, the third branch fifth normalization layer and the third branch fifth activation function layer, which are connected in series in sequence; wherein, the convolution kernel of the third branch first convolution layer is 5×5, and the step length is The convolution kernel of the second convolution layer of the third branch is 5×5, the stride is 2, the padding is 2, the input channel is 3, and the output channel is 16; the convolution kernel of the third convolution layer of the third branch is 5×5, the stride is 1, the padding is 2, the input channel is 16, and the output channel is 16; the convolution kernel of the third convolution layer of the third branch is 5×5, the stride is 1, the padding is 2, the input channel is 16, and the output channel is 32; the convolution kernel of the fourth convolution layer of the third branch is 5×5, the stride is 2, the padding is 2, the input channel is 32, and the output channel is 32; the convolution kernel of the fifth convolution layer of the third branch is 5×5, the stride is 1, the padding is 2, the input channel is 32, and the output channel is 64.

6. The bearing fault diagnosis method according to claim 5, characterized in that The channel attention module specifically includes the following steps: The features of the input channel attention module are divided into two paths: the first path is processed by the maximum pooling layer and the first convolution layer in turn to obtain the first channel features; the second path is processed by the average pooling layer and the second convolution layer in turn to obtain the second channel features; then the first channel features and the second channel features are added and processed by the activation function; The processing process of the channel attention module is expressed as Mc(F)=σ(f 1 × 1 (AvgPool(F))+f 1 × 1 (MaxPool(F))) Where M c (F) is the output feature of the channel attention module; σ() is the Sigmoid activation function; f 1×1 () is the convolution layer processing function with a convolution kernel of 1×1; AvgPool() is the average pooling layer processing function; F is the input feature; MaxPool() is the maximum pooling layer processing function.

7. The bearing fault diagnosis method according to claim 6, characterized in that The spatial attention module specifically includes the following steps: The features of the input spatial attention module are divided into two paths: the first path is processed by the maximum pooling layer in sequence to obtain the first spatial feature; the second path is processed by the average pooling layer to obtain the second spatial feature; then the first spatial feature and the second spatial feature are cascaded and processed by the convolution layer and activation function in sequence; The processing of the spatial attention module is expressed as Ms(F)=σ(f 7 × 7 ([AvgPool(F);MaxPool(F)])) Where σ() is the Sigmoid activation function; f 7×7 () is the convolution layer processing function with a convolution kernel of 7×7; AvgPool() is the average pooling layer processing function; F is the input feature; MaxPool() is the maximum pooling layer processing function.

8. The bearing fault diagnosis method according to claim 7, characterized in that The aggregation module specifically includes the following steps: Linear network is used as the aggregation module; The classification module specifically includes the following steps: A fully connected layer is used as the classification module.

9. The bearing fault diagnosis method according to claim 8, characterized in that The training described in step S4 specifically includes the following steps: The cross entropy loss function is used as the total loss function; The model was trained using the stochastic gradient descent algorithm, with the learning rate set to 0.01 and the number of iterations set to 100.

10. A system for implementing the bearing fault diagnosis method according to any one of claims 1 to 9, characterized in that It includes a data acquisition module, a data processing module, a model building module, a model training module and a fault diagnosis module; the data acquisition module, the data processing module, the model building module, the model training module and the fault diagnosis module are connected in series in sequence; the data acquisition module is used to acquire the existing bearing working data information and upload the data information to the data processing module; the data processing module is used to convert the acquired data information into a two-dimensional time-frequency image based on the received data information and wavelet transform, and construct a training data set, and upload the data information to the model building module; The model building module is used to build a primary model for bearing fault diagnosis based on the received data information, based on the convolutional network and attention mechanism, and upload the data information to the model training module; The model training module is used to train the constructed primary model for bearing fault diagnosis based on the received data information and the obtained training data set, obtain the bearing fault diagnosis model, and upload the data information to the fault diagnosis module; the fault diagnosis module is used to obtain the working data of the target bearing in real time based on the received data information, and use the obtained bearing fault diagnosis model to perform fault diagnosis on the target bearing.

Images