Methods, systems and devices for gearbox fault diagnosis
By combining the CBAM convolutional block attention module and the BiLSTM model with sliding window processing and softmax classification, the problems of delay and noise interference in gear fault diagnosis are solved, and high-precision fault identification and classification are achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- GUANGZHOU UNIVERSITY
- Filing Date
- 2023-03-03
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies are difficult to effectively identify the time delay of gear faults and are easily affected by noise interference, which affects fault diagnosis performance.
We employ a CNN and BiLSTM model with CBAM convolutional block attention module, combined with sliding window processing and softmax classification, to extract key features of gearbox vibration signals and perform fault identification.
It improves the predictive accuracy and anti-interference ability of gear fault diagnosis, and enables accurate classification of gear fault types.
Smart Images
Figure CN116296374B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of gearbox fault diagnosis, and specifically relates to a method, system and device for gearbox fault diagnosis. Background Technology
[0002] Gears are a critical component widely used in rotating machinery, often determining the lifespan of the equipment. Unexpected gear failures can lead to severe damage to rotating machinery. However, current technologies rarely consider the time delay of failure occurrence, which affects the performance of fault diagnosis and makes it difficult to handle noise interference. Therefore, effective and reliable fault identification is needed to ensure the normal operation of gears. Summary of the Invention
[0003] In view of the above-mentioned deficiencies of the prior art, the object of the present invention is to provide a method for diagnosing gearbox faults, characterized by comprising the following steps:
[0004] S1. Obtain the original vibration signal of the gearbox to obtain one-dimensional data;
[0005] S2. Input the one-dimensional dataset into a CNN containing a CBAM convolutional block attention module to obtain the signal of important features;
[0006] S3. Input the data processed by CNN into BiLSTM to obtain the features before and after capturing the time series information;
[0007] S4. Integrate the data processed in S3 into a two-dimensional image data through a sliding window process;
[0008] S5. The data processed by S4 is classified and fault identified through a fully connected layer and softmax.
[0009] In S1, the one-dimensional dataset includes: acceleration, rotation angle, and pressure.
[0010] In S2, the CBAM convolutional block attention module consists of a channel attention module and a spatial attention module.
[0011] In S4, the sliding window's architecture is configured as follows:
[0012] In sliding window processing, based on the original data X, a two-dimensional sample is obtained, and the calculation formula is as follows:
[0013] X={x t |t=1,2,3...n}(x t ∈R m , X∈R m×n )
[0014] Y = {y t|t=1,2,3...n}(y t ∈R, Y∈R n )
[0015] In the formula, x t For time, y t Let m be the observation vector of the system at time t, m be the number of observed attributes, n be the length of the observation time, and the sample representation of the original data be D = {(X, Y)}.
[0016] In S5, softmax classification maps the output vector of the fully connected layer to a range of 0 to 1, obtaining the probability that each sample is identified as a different category. The calculation process of the softmax function is as follows:
[0017]
[0018] Among them, P i (x) represents the probability that input x belongs to type i, m represents the total number of types, and xi represents the value of that type.
[0019] This invention provides a system for diagnosing gearbox faults, comprising:
[0020] Acquisition module: used to acquire the raw vibration signal of the gearbox to obtain a one-dimensional dataset;
[0021] Processing module: Integrates the spatial and temporal characteristics of the original vibration signal of the gearbox into a two-dimensional image data through sliding window processing;
[0022] Attention focus mechanism module: used to input one-dimensional datasets into CNN and BiLSTM models to obtain one-dimensional data features that focus on important features;
[0023] Recognition Module: Finally, the two-dimensional image obtained by the sliding window processing is put into the fully connected layer, and softmax is used for fault recognition.
[0024] The present invention provides a device for diagnosing gearbox faults, comprising: a memory, a processor, and a computer program, wherein the computer program is stored in the memory and executed on the processor.
[0025] The present invention provides a computer storage medium comprising a set of computer-executable instructions for implementing a gearbox fault diagnosis method as described in any one of the above.
[0026] The beneficial effects of this invention are: it can classify gearbox fault types, improve the prediction accuracy and anti-interference ability of fault diagnosis. Attached Figure Description
[0027] Figure 1This is a flowchart of a method for diagnosing gearbox faults according to the present invention.
[0028] Figure 2 This is a schematic diagram of the sliding window architecture of a gearbox fault diagnosis method according to the present invention.
[0029] Figure 3 This is a flowchart of the attention mechanism of the CBAM convolutional block attention module in a gearbox fault diagnosis method of the present invention.
[0030] Figure 4 This is a schematic diagram of the BiLSTM structure of a gearbox fault diagnosis method according to the present invention.
[0031] Figure 5 This is a schematic diagram of the Softmax structure of a gearbox fault diagnosis method according to the present invention.
[0032] Figure 6 This is a schematic diagram of the CNN structure of a gearbox fault diagnosis method according to the present invention.
[0033] Figure 7 This is a flowchart of a gear fault diagnosis system according to the present invention.
[0034] Figure 8 This is a schematic diagram of a gear fault diagnosis device according to the present invention. Detailed Implementation
[0035] The embodiments of the present invention will be described in detail below. The embodiments described below are implemented based on the technical solution of the present invention, and detailed implementation methods and specific operation processes are given. However, the protection scope of the present invention is not limited to the embodiments described below.
[0036] In view of the above-mentioned deficiencies of the prior art, the object of the present invention is to provide a method for diagnosing gearbox faults, characterized by comprising the following steps:
[0037] S1. Obtain the original vibration signal of the gearbox to obtain one-dimensional data;
[0038] S2. Input the one-dimensional dataset into a CNN containing a CBAM convolutional block attention module to obtain the signal of important features;
[0039] S3. Input the data processed by CNN into BiLSTM to obtain the features before and after capturing the time series information;
[0040] S4. Integrate the data processed in S3 into a two-dimensional image data through a sliding window process;
[0041] S5. The data processed by S4 is classified and fault identified through a fully connected layer and softmax.
[0042] In S1, the one-dimensional dataset includes: acceleration, rotation angle, and pressure.
[0043] In S2, the CBAM convolutional block attention module consists of a channel attention module and a spatial attention module.
[0044] In S4, the sliding window's architecture is configured as follows:
[0045] In sliding window processing, based on the original data X, a two-dimensional sample is obtained, and the calculation formula is as follows:
[0046] X={x t |t=1,2,3...n}(x t ∈R m , X∈R m×n )
[0047] Y = {y t |t=1,2,3...n}(y t ∈R, Y∈R n )
[0048] In the formula, x t For time, y t Let m be the observation vector of the system at time t, m be the number of observed attributes, n be the length of the observation time, and the sample representation of the original data be D = {(X, Y)}.
[0049] In S5, softmax classification maps the output vector of the fully connected layer to a range of 0 to 1, obtaining the probability that each sample is identified as a different category. The calculation process of the softmax function is as follows:
[0050]
[0051] Among them, P i (x) represents the probability that input x belongs to type i, m represents the total number of types, and xi represents the value of that type.
[0052] This invention provides a system for diagnosing gearbox faults, comprising:
[0053] Acquisition module: used to acquire the raw vibration signal of the gearbox to obtain a one-dimensional dataset;
[0054] Processing module: Integrates the spatial and temporal characteristics of the original vibration signal of the gearbox into a two-dimensional image data through sliding window processing;
[0055] Attention focus mechanism module: used to input one-dimensional datasets into CNN and BiLSTM models to obtain one-dimensional data features that focus on important features;
[0056] Recognition Module: Finally, the two-dimensional image obtained by the sliding window processing is put into the fully connected layer, and softmax is used for fault recognition.
[0057] The present invention provides a device for diagnosing gearbox faults, comprising: a memory, a processor, and a computer program, wherein the computer program is stored in the memory and executed on the processor.
[0058] The present invention provides a computer storage medium comprising a set of computer-executable instructions for implementing a gearbox fault diagnosis method as described in any one of the above.
[0059] This invention can classify gearbox fault types, thereby improving the predictive accuracy and anti-interference capability of fault diagnosis.
[0060] As attached Figure 1-8 As shown, according to an embodiment of the present invention, a method for diagnosing gearbox faults is provided. Figure 1 This is a flowchart of the gear fault diagnosis method using CNN and BiLSTM models according to an embodiment of the present invention, as shown below. Figure 1 As shown, it specifically includes:
[0061] S1. Obtain the original vibration signal of the gearbox to obtain one-dimensional data;
[0062] S2. Input the one-dimensional dataset into a CNN containing a CBAM convolutional block attention module to obtain the signal of important features;
[0063] S3. Input the data processed by CNN into BiLSTM to obtain the features before and after capturing the time series information;
[0064] S4. Integrate the data processed in S3 into a two-dimensional image data through a sliding window process;
[0065] S5. The data processed by S4 is classified and fault identified through a fully connected layer and softmax.
[0066] Specifically, S1 includes: acquiring the original vibration signal of the gearbox to obtain a one-dimensional dataset, which includes: acceleration, rotation angle and pressure.
[0067] S2 specifically includes:
[0068] S21. Input the acquired one-dimensional image data into a CNN containing the CBAM attention mechanism;
[0069] S22. One-dimensional data is passed through a one-dimensional convolutional layer to obtain a larger receptive field and more useful feature information. Then, through the CBAM attention concentration mechanism, CBAM can calculate attention weights along the spatial and channel dimensions, and then multiply them with the original features to obtain a new one-dimensional data. The new one-dimensional data redistributes the weight coefficients among different feature dimensions, improving the attention to important features and effectively suppressing useless information.
[0070] S23. The new one-dimensional data is fed into the Maxpooling layer to select the data with the most obvious features. This further improves the ability to suppress useless information.
[0071] S3 specifically includes:
[0072] S3. The one-dimensional data processed by CNN is put into BiLSTM. BiLSTM can learn the long-term correlation in time series information through special gating mechanism and structural features. The spatial features extracted from the CNN layer will be transmitted to the BiLSTM layer. BiLSTM will fully extract the timing features before and after vibration signal, solving the problem of CNN losing some relevant features in time.
[0073] S4 specifically includes:
[0074] S4. Integrate the data processed in S3 into a two-dimensional image data through a sliding window process.
[0075] S5 specifically includes:
[0076] S5. Finally, the two-dimensional image processed in S4 is put into a fully connected layer, and softmax is used for fault identification.
[0077] The specific implementation method is as follows:
[0078] Figure 3 This is a schematic diagram of the sliding window architecture for gear fault diagnosis using CNN and BiLSTM models according to an embodiment of the present invention.
[0079] Explanation of the sliding window architecture:
[0080] In sliding window processing, two-dimensional samples can be obtained based on the original data X (which has undergone some data preprocessing, such as normalization, by simultaneously considering its feature information and time delay information).
[0081] The calculation formula is as follows:
[0082] X={x t |t=1,2,3...n}(x t ∈R m , X∈R m×n )
[0083] Y = {y t |t = 1, 2, 3...n}(y t ∈R, Y∈R n )
[0084] In the formula, x t For time, y t Let m be the observation vector of the system at time t, m be the number of observed attributes, n be the length of the observation time, and the sample representation of the original data be D = {(X, Y)}.
[0085] CBAM attention mechanism: It consists of two modules: channel attention and spatial attention. The channel attention module allows the network to focus on the foreground of the image, making it more attentive to meaningful ground truth regions. The spatial attention module allows the network to focus on locations in the entire image rich in contextual information, as shown in the attached diagram. Figure 2 .
[0086] Channel Attention Module: Each channel of a feature represents a dedicated detector; therefore, channel attention focuses on which features are meaningful. To aggregate spatial features, max pooling is used to utilize diverse information.
[0087]
[0088] The input is a feature F representing the vibration signal of a gear. We first perform spatial max pooling to obtain 1×1×C channel descriptions. Then, we feed these features into a two-layer neural network. The first layer has C / t neurons with a ReLU activation function, and the second layer has C neurons. This two-layer neural network is shared. Next, we sum the two features and pass the sum through a sigmoid activation function to obtain the weight coefficients Mc. Finally, we multiply the weight coefficients by the original feature F to obtain the scaled new feature.
[0089] Spatial attention module: follows the channel attention module;
[0090]
[0091] After channel attention processing, the feature F is max-pooled to obtain a 3D channel description, and these two descriptions are concatenated according to the channels. Then, it is passed through a 7×7 convolutional layer with the sigmoid activation function to obtain the weight coefficients M. s Finally, multiplying the weight coefficients by the feature F' yields the scaled new feature.
[0092] Appendix Figure 6This is a schematic diagram of the CNN structure for gear fault diagnosis using CNN and BiLSTM models according to an embodiment of the present invention;
[0093] CNN is a feedforward neural network with convolutional computation and deep structure. One-dimensional CNN is very effective for extracting features from sequential data. Each hidden layer of a CNN includes a convolutional layer and a pooling layer. The convolutional computation process is as follows:
[0094]
[0095] In the formula: K and b are the weights and biases of the i-th filter in the first layer network, respectively, and x represents the i-th input in the first layer.
[0096] A CBAM attention module is added after the convolutional layer, followed by a pooling layer to select and filter the extracted features. Max pooling is the most commonly used pooling operation; it selects the maximum statistical measure to capture local features of the region, reducing the number of parameters. The max pooling transformation is described below:
[0097] p l(i，j) =max (j-1)w+1≤t≤jw , {a l(i，t)}
[0098] In the formula: a represents the t neuron of the i-th channel in layer 1, W is the width of pooling, and p represents the corresponding value of the neuron in layer 1 in the pooling operation.
[0099] Figure 4 This is a BiLSTM structure diagram of the gear fault diagnosis method using CNN and BiLSTM models according to an embodiment of the present invention.
[0100] Figure 5 This is a flowchart illustrating the identification mechanism of the gear fault diagnosis method using CNN and BiLSTM models according to an embodiment of the present invention.
[0101] The softmax classifier maps the output vector of the fully connected layer to a range of 0 to 1, obtaining the probability that each sample is identified as a different class, thus achieving the classification function. The calculation process of the softmax function is as follows:
[0102]
[0103] Among them, P i (x) represents the probability that input x belongs to type i, m represents the total number of types, and xi represents the value of that type.
[0104] Figure 5 This is a flowchart of the CBAM attention mechanism of the gear fault diagnosis method using CNN and BiLSTM models according to an embodiment of the present invention.
[0105] Gear fault diagnosis steps using CNN and BiLSTM models:
[0106] 1) Extract the state characteristics of the gear directly from the gear vibration signal;
[0107] 2) Based on the original data, two-dimensional image samples are obtained through sliding window processing;
[0108] 3) Input the two-dimensional image data into the convolutional layer of the CNN, extract the hidden feature information in the data, and add non-linear factors to the features to improve the expressive power of the features.
[0109] 4) Based on the established CBAM attention mechanism model, the data processed by the convolutional layer is fed into CBAM to obtain the weight coefficients M. s Finally, multiplying the weight coefficients by the feature F' yields the scaled new feature.
[0110] 5) Put the new two-dimensional image features into the max pooling layer, select the maximum statistic to obtain the local features of the region, and reduce the number of parameters.
[0111] 6) Connect the features trained by BiLSTM to the softmax layer to achieve fault classification.
[0112] A gear fault diagnosis system based on CNN and BiLSTM models is provided. Figure 7 This is a schematic diagram of a gear fault diagnosis system using CNN and BiLSTM models according to an embodiment of the present invention, as shown below. Figure 7 As shown, it specifically includes:
[0113] Acquisition module: used to acquire the raw vibration signal of the gearbox to obtain a one-dimensional dataset;
[0114] Attention Concentration Mechanism Module: Used to input one-dimensional datasets into CNN and BiLSTM models to obtain two-dimensional images that pay attention to important features;
[0115] Processing module: Integrates the spatial and temporal characteristics of the gearbox vibration signal into a two-dimensional image data through sliding window processing;
[0116] Recognition Module: Finally, the two-dimensional image obtained by the sliding window processing is put into the fully connected layer, and softmax is used for fault recognition.
[0117] The acquisition module is specifically used to: acquire the original vibration signal of the gearbox to obtain a one-dimensional dataset, wherein the one-dimensional dataset includes: acceleration, rotation angle and pressure.
[0118] The attention focus mechanism module is specifically used for:
[0119] 1. Input the acquired 2D image data into a CNN containing the CBAM attention mechanism;
[0120] 2. Two-dimensional image data is passed through a two-dimensional convolutional layer to obtain a larger receptive field and more useful feature information. Then, through the CBAM attention concentration mechanism, CBAM can calculate attention weights along the spatial and channel dimensions of the feature map, and then multiply them with the original feature map to obtain a new two-dimensional feature map. The new two-dimensional feature map redistributes the weight coefficients among different feature dimensions, improving attention to important features and effectively suppressing useless information.
[0121] 3. The new 2D feature map is fed into the Maxpooling layer to select the 2D map with the most obvious features. This further improves the ability to suppress useless information.
[0122] The processing module is specifically used to integrate the spatial and temporal characteristics of the gearbox vibration signal into a two-dimensional image data through sliding window processing.
[0123] The recognition module is specifically used to: finally put the two-dimensional image obtained by the sliding window processing into the fully connected layer, and use softmax for fault recognition.
[0124] The embodiments of the present invention are system embodiments corresponding to the above method embodiments. The specific operation of each module can be understood by referring to the description of the method embodiments, and will not be repeated here.
[0125] Device Example 1:
[0126] This invention provides a gear fault diagnosis device, such as... Figure 8 As shown, it includes: a memory 80, a processor 82, and a computer program stored on the memory 80 and executable on the processor 82. When the computer program is executed by the processor, it implements the steps in the above method embodiments.
[0127] Device Example 2:
[0128] This invention provides a computer-readable storage medium storing an information transmission implementation program, which, when executed by a processor 82, implements the steps described in the above method embodiments.
[0129] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.
Claims
1. A method for diagnosing gearbox faults, characterized in that, Includes the following steps: S1. Obtain the original vibration signal of the gearbox to obtain one-dimensional data; S2. Input one-dimensional data into a CNN containing a CBAM convolutional block attention module to obtain signals of important features; In S2, the CBAM convolutional block attention module consists of a channel attention module and a spatial attention module; S2 specifically includes: S21. Input the obtained one-dimensional data into a CNN containing a CBAM convolutional block attention module; S22. One-dimensional data is passed through a one-dimensional convolutional layer to obtain a larger receptive field and more useful feature information. Then, through the CBAM attention concentration mechanism, CBAM can calculate attention weights along the spatial and channel dimensions, and then multiply them with the original features to obtain a new one-dimensional data. The new one-dimensional data redistributes the weight coefficients among different feature dimensions, improves the attention to important features, and effectively suppresses useless information. S23. The new one-dimensional data is put into the Maxpooling pooling layer to select the data with the most obvious features, which further effectively improves the ability to suppress useless information. S3. Input the data processed by CNN into BiLSTM to obtain the features before and after capturing the time series information; S4. Integrate the data processed in S3 into a two-dimensional image data through a sliding window process; S5. The data processed in S4 is classified and fault identified through a fully connected layer and softmax. In S4, the sliding window's architecture is configured as follows: In the sliding window processing, based on the one-dimensional data, a two-dimensional sample is obtained, and the calculation formula is as follows: In the formula, For time, Let D be the one-dimensional data vector of the system at time t. When t=n, it means that the system is at time n. The two-dimensional sample is represented as D = {(X, Y)}.
2. The method according to claim 1, characterized in that, In S1, the one-dimensional data includes: acceleration, rotation angle, and pressure.
3. The method according to claim 1, characterized in that, In S5, softmax classification maps the output vector of the fully connected layer to a range of 0 to 1, obtaining the probability that each sample is identified as a different category. The calculation process of the softmax function is as follows: in, Let represent the probability that input x belongs to type i, m represent the total number of types, and xi represent the value of that type.
4. A system for diagnosing gearbox faults, characterized in that, include: Acquisition module: used to acquire the raw vibration signal of the gearbox to obtain a one-dimensional dataset; Processing module: Integrates the spatial and temporal characteristics of the original vibration signal of the gearbox into a two-dimensional image data through sliding window processing; Attention focus mechanism module: used to input one-dimensional data into CNN and BiLSTM models to obtain one-dimensional data features that focus on important features; Specifically, it includes: The acquired one-dimensional data is input into a CNN containing a CBAM convolutional block attention module; One-dimensional data is passed through a one-dimensional convolutional layer to obtain a larger receptive field and more useful feature information. Then, through the CBAM attention concentration mechanism, CBAM can calculate attention weights along the spatial and channel dimensions, and then multiply them with the original features to obtain a new one-dimensional data. The new one-dimensional data redistributes the weight coefficients among different feature dimensions, improves the attention to important features, and effectively improves the ability to suppress useless information. The new one-dimensional data is put into the Maxpooling pooling layer to select the data with the most obvious features, which further effectively improves the ability to suppress useless information. Recognition module: Finally, the two-dimensional image obtained by the sliding window processing is put into the fully connected layer, and softmax is used for fault recognition; In the recognition module, the sliding window architecture is as follows: In the sliding window processing, based on the one-dimensional data, a two-dimensional sample is obtained, and the calculation formula is as follows: In the formula, For time, Let D be the one-dimensional data vector of the system at time t. When t=n, it means that the system is at time n. The two-dimensional sample is represented as D = {(X, Y)}.
5. A device for diagnosing gearbox faults, characterized in that, include: The system includes a memory, a processor, and a computer program, wherein the computer program is stored in the memory and executed on the processor to implement the gearbox fault diagnosis method according to any one of claims 1 to 3.
6. A computer storage medium, characterized in that, The storage medium includes a set of computer-executable instructions to implement a gearbox fault diagnosis method according to any one of claims 1 to 3.