CN118520357B - Bearing fault diagnosis method, system and storage medium based on characterization guidance - Google Patents
Bearing fault diagnosis method, system and storage medium based on characterization guidance Download PDFInfo
- Publication number
- CN118520357B CN118520357B CN202410985317.9A CN202410985317A CN118520357B CN 118520357 B CN118520357 B CN 118520357B CN 202410985317 A CN202410985317 A CN 202410985317A CN 118520357 B CN118520357 B CN 118520357B
- Authority
- CN
- China
- Prior art keywords
- activation
- layer
- loss
- fault diagnosis
- fault
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000003745 diagnosis Methods 0.000 title claims abstract description 48
- 238000000034 method Methods 0.000 title claims abstract description 40
- 238000012512 characterization method Methods 0.000 title claims abstract description 24
- 230000004913 activation Effects 0.000 claims abstract description 93
- 238000012549 training Methods 0.000 claims abstract description 32
- 238000010586 diagram Methods 0.000 claims abstract description 18
- 238000013507 mapping Methods 0.000 claims abstract description 7
- 230000004927 fusion Effects 0.000 claims abstract description 5
- 230000005764 inhibitory process Effects 0.000 claims abstract description 5
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 17
- 238000011176 pooling Methods 0.000 claims description 16
- 238000010606 normalization Methods 0.000 claims description 14
- 230000006870 function Effects 0.000 claims description 12
- 230000006378 damage Effects 0.000 claims description 11
- 238000004590 computer program Methods 0.000 claims description 6
- 238000001514 detection method Methods 0.000 claims description 6
- 230000001629 suppression Effects 0.000 claims description 2
- 238000012360 testing method Methods 0.000 description 10
- 230000008569 process Effects 0.000 description 7
- 230000007246 mechanism Effects 0.000 description 6
- 230000000007 visual effect Effects 0.000 description 6
- 230000006835 compression Effects 0.000 description 4
- 238000007906 compression Methods 0.000 description 4
- 238000005096 rolling process Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 238000013136 deep learning model Methods 0.000 description 3
- 230000007547 defect Effects 0.000 description 3
- 230000005284 excitation Effects 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 239000013598 vector Substances 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 2
- 230000036541 health Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 241000282412 Homo Species 0.000 description 1
- 208000027418 Wounds and injury Diseases 0.000 description 1
- 230000001133 acceleration Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 230000019771 cognition Effects 0.000 description 1
- 238000012733 comparative method Methods 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000005336 cracking Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000005553 drilling Methods 0.000 description 1
- 238000010892 electric spark Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000013209 evaluation strategy Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000002401 inhibitory effect Effects 0.000 description 1
- 208000014674 injury Diseases 0.000 description 1
- 230000004807 localization Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 238000011056 performance test Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
- 238000007794 visualization technique Methods 0.000 description 1
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a fault diagnosis method, system and storage medium based on signal characteristic characterization guidance, which uses a characteristic extractor to obtain a characteristic diagram of the last convolution layer output when constructing a fault model, and mapping the feature map into positive and negative class activation maps by adopting a class activation method, and then calculating the difference between the positive class activation map and the feature map of the training sample to obtain the activation function loss value of the positive class feature. And calculating the difference between the negative class activation graph of the input sample and the inhibition constant, and taking the difference value as the loss value of the negative class activation function. And carrying out weighted fusion on the negative class activation loss, the positive class activation loss and the classification loss of the classifier, and further optimizing model parameters of the fault diagnosis model. The invention can drive the main network of the fault diagnosis model to focus on the target category, express the remarkable region of the fault feature as the feature of the target category, and improve the diagnosis precision and stability of the model by improving the characterization capability of the model on the signal feature.
Description
Technical Field
The invention relates to the field of artificial intelligence and fault diagnosis, in particular to a bearing fault diagnosis method, system and storage medium based on characterization guidance.
Background
In the field of modern fault diagnosis, deep learning technology has attracted extensive attention in the industry and academia due to its excellent feature learning ability and efficient processing ability for large-scale data. Although the deep learning model has outstanding image processing and pattern recognition capabilities, it is considered a "black box" because of its high nonlinearity, multiple hidden layers, and a large amount of parameters, so that one cannot know exactly the decision basis behind it and the reliability of its internal structure. Especially in key fields such as automatic driving, industrial equipment health monitoring and the like, the decision-making result of the model directly relates to the life safety of people, and the trust problem of people on the model is deepened. Thus, the decision of the model requires not only high accuracy but also good stability and reliability so that its decision result can be understood and trusted by humans. To improve the interpretability of the model, researchers have tried to explain the decision mechanism of the model from different angles, however, one major limitation of the interpretation methods of the decision mechanism is that they only highlight areas where the model deems important for decision, but do not provide a practical reason for decision.
If an object is commonly co-occurring with certain visual features in a training sample, the model may learn to use the co-occurring features to represent the object. When common features are used that are not related to the target, these features are considered biased feature representations. In traditional studies, correction of characterization defects due to dataset bias or overfitting problems is a common problem. It may pose a potential threat to the task in a real scenario and also bring similar bias to the test sample that cannot be checked using conventional evaluation strategies.
Disclosure of Invention
The invention provides a bearing fault diagnosis method, a bearing fault diagnosis system and a storage medium based on characterization guidance, which are used for solving the technical problem of poor stability of the existing fault diagnosis method.
In order to solve the technical problems, the technical scheme provided by the invention is as follows:
a bearing fault diagnosis method based on characterization guidance, comprising:
constructing a fault diagnosis model, wherein the fault diagnosis model comprises a feature extractor and a classifier which are sequentially connected in series;
Obtaining vibration signals of a normal bearing and a fault bearing to construct a training sample set, and formulating a training strategy to train the fault diagnosis model;
the training strategy is:
Extracting a feature image output by the last convolution layer of the feature extractor, and acquiring positive and negative class activation images of an input sample by adopting a class activation mapping method; calculating the difference between the positive class activation graph and the feature graph of the training sample to be used as positive class activation loss; meanwhile, calculating the difference between a negative class activation graph and an inhibition constant of an input sample to be used as negative class activation loss; carrying out weighted fusion on the negative class activation loss, the positive class activation loss and the classification loss of the classifier, and optimizing model parameters of the fault diagnosis model according to a weighted fusion loss function;
And obtaining a vibration signal of the target bearing to construct a detection sample, and inputting the detection sample into a trained fault diagnosis model to obtain the fault type of the target bearing.
Preferably, the feature extractor comprises a plurality of serially connected residual layers, each residual layer comprises a plurality of serially connected residual blocks, each residual block comprises a plurality of serially connected convolution normalization units, and each convolution normalization unit comprises serially connected convolution layers and batch normalization layers; and in N residual layers positioned at the tail part of the series connection, a channel attention module is connected in the last batch normalization layer of each residual block.
Preferably, the feature extractor comprises 4 residual layers, each residual layer comprising 2 residual blocks, each residual block comprising 2 convolution normalization units; and in the last 2 residual layers positioned at the tail part of the series connection, a channel attention module is connected in the last batch normalization layer of each residual block.
Preferably, the channel attention module comprises an average pooling layer, a maximum pooling layer, a first full-connection layer, a second full-connection layer, a first activation layer and a second activation layer, wherein the output ends of the average pooling layer and the maximum pooling layer are connected with the input end of the first full-connection layer, the output end of the first full-connection layer is connected with the input end of the first activation layer, and the output end of the first activation layer is connected with the input end of the second full-connection layer; the output end of the second full-connection layer is connected with the input end of the second activation layer.
Preferably, the first activation layer adopts a ReLU activation function, and the second activation layer adopts a Sigmoid function.
Preferably, the positive class activation loss satisfies:
；
wherein, For a positive class of activation loss,In order to pay attention to the height of the force diagram,In order to pay attention to the width of the force diagram,A class activation graph for a positive class,Locating coordinates for pixel points in the class activation map and attention map; AM is an attention map;
The negative class activation loss satisfies:
；
wherein, For a negative class of activation loss,For the real label of the input image,In order to be a set of tags,For the set of all negative class labels:， Is the serial number of the negative type label, Is a suppression constant.
Preferably, the classification loss satisfies:
；
In order to classify the loss value(s), In order to be able to predict the value,To be a true value of the value,As a point of time in time it is,Sample tags are numbered.
Preferably, the detection sample and the training sample are single-channel time-frequency diagrams of the bearing, and the fault category comprises a fault position category and a damage depth category; the fault position categories are divided into an outer ring and an inner ring;
Obtaining vibration signals of a normal bearing and a fault bearing to construct a training sample set comprises the following steps:
obtaining vibration signals of various known fault categories from historical data, and extracting a single-channel time-frequency diagram from each vibration signal;
and constructing a training sample set by taking the fault category corresponding to each single-channel time-frequency diagram as a label.
A computer system comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the steps of the method described above when the computer program is executed by the processor.
A computer storage medium having stored thereon a computer program which when executed by a processor performs the steps of the above method.
The invention has the following beneficial effects:
1. When the fault diagnosis model is trained, the invention constrains the class activation diagram of the whole class, and can drive the backbone network to focus on the target class and express the visual salient region as the characteristic of the target class, thereby improving the diagnosis precision and stability of the model and simultaneously enabling the decision process of the model to be more interpretable.
2. In a preferred scheme, when the fault diagnosis model is designed, the channel attention module is introduced into the traditional residual structure, so that the information extraction and feature representation capability of the model are effectively enhanced, and the fault diagnosis precision is further improved.
In addition to the objects, features and advantages described above, the present invention has other objects, features and advantages. The invention will be described in further detail with reference to the accompanying drawings.
Drawings
The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the application. In the drawings:
FIG. 1 is a flow chart of a characterization guidance based bearing fault diagnosis method of the present invention.
Fig. 2 is a schematic structural view of the feature extractor of the present invention.
Fig. 3 is a schematic structural diagram of a channel attention module according to the present invention.
FIG. 4 is a flow chart of the training diagnostic model of the present invention.
Fig. 5 is a functional block diagram of a fault diagnosis system based on characterization guidance of the present invention.
Detailed Description
Embodiments of the invention are described in detail below with reference to the attached drawings, but the invention can be implemented in a number of different ways, which are defined and covered by the claims.
In this embodiment, as shown in fig. 1, the present embodiment provides a bearing fault diagnosis method based on characterization guidance, including the following steps:
s1, constructing a model input sample by using rolling bearing vibration signal data;
dividing samples of the rolling bearing vibration signal according to the same sampling point number; the divided one-dimensional signal is subjected to short-time fourier transform into a two-dimensional image.
In this example, the data used were from the kesixin Chu Daxue and pamphlet university rolling bearing datasets. For a data set of Kassi university, bearing vibration signals collected by a driving end acceleration sensor are selected, wherein the bearing vibration signals comprise bearings in four health states of normal, rolling body faults, inner ring faults and outer ring faults, the fault diameters of each fault type are respectively 0.007 inches, 0.014 inches and 0.021 inches, and ten fault types are provided. The data set expansion is performed by means of overlapping sampling, the moving step length is 400, and each sample comprises 2048 data points. Then, each sample is subjected to short-time Fourier transform to obtain a single-channel time-frequency diagram as the input of a model. 1000 samples in each class of the data set comprise 10000 samples, and are divided into a training set, a verification set and a test set according to the proportion of 6:2:2. For the university of Pade Boen dataset, all test bearings of this dataset were model 6203, sampling frequency was 64kHz. The fault bearing is divided into artificial damage and real damage, the artificial damage mainly comprises cracking caused by electric sparks, peeling caused by drilling and pitting caused by an electric engraving machine, and the real damage bearing is obtained through an accelerated life test bed. Dividing two types of damage data according to fault positions and damage degrees, wherein the fault positions are divided into an inner ring and an outer ring, the damage depth comprises 0 mm-2 mm and 2 mm-4.5 mm, and 5 types of fault types are formed together with normal data. 2000 samples are sampled for each fault category, the sample length is set to 2048, the short-time Fourier transform is performed on the collected samples as the data set of Kassi university, and the samples are divided into a training set, a verification set and a test set according to the ratio of 6:2:2.
S2, constructing a fault diagnosis model, wherein the diagnosis model consists of a convolution feature extractor, a classifier and the like;
As shown in fig. 2, the feature extractor in this embodiment adopts a modified ResNet residual network structure, and has a total of 4 residual layers, where each residual layer contains 2 residual blocks, and each residual block is formed by connecting two 3*3 convolution layers and two batch normalization layers in series. Since 32 times downsampling of ResNet structures results in that the feature map output by the last layer only retains less spatial information, a hole convolution with a hole rate of 2 is applied in the 3 rd residual block, and a hole convolution with a hole rate of 4 is applied in the 4 th residual block. This reduces the downsampling factor to 8 times while maintaining the primordial receptive field. Furthermore, in the 3 rd and 4 th residual layers, a channel attention module is added after the second batch normalization layer of each residual block, respectively. The classifier consists of only one full connection layer.
The channel attention module improves the feature selection and representation capability of the network by dynamically adjusting the weights of the channels of the feature map. The channel attention module respectively uses global maximum pooling and global average pooling to aggregate the space information of the input feature images so as to obtain two global information abstracts, then respectively inputs the two aggregated features into two full-connection layers, then combines the output feature vectors element by element, and activates the feature vectors through a Sigmoid function so as to generate importance weights of all channels. Finally, the input feature map is weighted to obtain a feature map enhanced by the channel attention module.
Specifically, the structure of the channel attention module in this embodiment is shown in fig. 3, and includes two main steps: compression and excitation. The compression stage uses global average pooling to summarize the features of each channel. The excitation stage uses the information aggregated in the compression stage to generate a weight value for each channel feature. Finally, the generated weights are normalized and multiplied channel by channel with the input feature map. In addition, attention generated in combination with global average pooling and global maximum pooling tends to have a more vivid visual effect. An improved channel attention module is employed in this embodiment. In the compression stage, the feature images are taken as input, and the spatial information of the global average pooling and global maximum pooling aggregated feature images are respectively used to obtain two global information abstracts. In the excitation stage, two aggregated features are input to two fully connected layers, and the output feature vectors are subjected to element combination to generate importance weights of all channels, wherein the weights are used for re-weighting feature responses of all channels so as to better highlight important feature channels and inhibit unimportant channels.
Step S3, in the model training process, obtaining a feature map output by the last convolution layer of the model feature extractor, and obtaining positive and negative class activation maps of an input sample by adopting a class activation mapping method;
The class activation mapping is a characterization visualization method, and the feature areas of different classes in the image are characterized and guided by a visualization model. The basic idea of the method is to add a global average pooling layer at the last convolution layer of the model to compress the spatial dimension of the feature map, and input the compressed feature map to the full connection layer for classification. And weighting the weight of the specific category in the full-connection layer and the feature map output by the convolution layer to obtain a category activation map of the specific category, so as to locate the area on the image associated with the specific category. Class activation graphs of target classes are referred to as positive class activation graphs, and class activation graphs of non-target classes are collectively referred to as negative class activation graphs. The advantage of class activation mapping is that no additional annotation or supervision information is needed for training, as it is built entirely on top of the internal feature representation of the model. This makes it a simple and efficient object localization technique through class activation diagrams, we can intuitively understand the feature representations learned by the model for different classes.
And S4, in the model training process, calculating the difference between the positive class activation diagram and the feature diagram of the training sample as the positive class activation loss. Meanwhile, calculating the difference between a negative class activation graph and an inhibition constant of an input sample to be used as negative class activation loss;
In this embodiment, the objective of the loss function is to address the characteristic characterization defect resulting from dataset bias or overfitting. These characterization defects can lead to false alarms or false alarms of fault diagnosis of the deep learning model, thereby limiting the application of the deep learning model in practical engineering.
The attention mechanism mimics the human visual system and directs the model to focus on characterizing the key activated areas of the attention map, allowing the model to work in a manner consistent with human cognition. Currently, existing attention mechanisms can drive the model to focus on areas of discriminatory power in the image, but these attention mechanisms do not take into account the correlation between target class and non-target, making it difficult for the model to distinguish significantly the target features from other non-target features.
The present invention thus performs the constraint on the full class activation graph, with the training flow of the model shown in FIG. 5. After introducing the channel attention module, we use the feature map (attention map) output by the last convolution layer of the model to guide the features of the positive class activation map, thereby driving the model to associate the target class with the salient region. In addition, feature expression of the model to non-target classes is suppressed by minimizing redundant activation of the negative class activation graph.
In order for the model to take the visually significant region as a feature representation of the target class, definePositive class activation loss to direct attention in an effort to focus on characterizing positive class activation features. Definition of the definitionAnd (3) for the negative class activation loss value, inhibiting the characterization of the negative class characteristics by the model by minimizing the redundant characteristics of the negative class activation graph.And (3) withAs a parameter update supervisor in the model training process. Order theA true label for an input image, whereinIn order to be a set of tags,A class activation graph for a positive class,The measurement is performed using the mean square error loss, defined as followsWhere AM is the attention profile, H, W is the height and width of the attention profile respectively,Pixel locations in the class activation map and attention map are located. In addition, set upFor the set of all negative class labels: Definition of WhereinTo suppress the constant, the level of activation of the negative class is generally set to 0.
And S5, forming a final training loss function to optimize the model by the weighted combination of the positive class activation loss, the negative class activation loss and the classification loss.
Used in the present embodimentAs a classification loss, it is defined asIn which, in the process,In order to be able to predict the value,Is a true value. Finally, the overall training loss function is obtained asIn which, in the process,AndTo controlAndThe superparameter of the weight can be adjusted according to the actual situation.
And (5) testing fault diagnosis performance. In the embodiment, a model based on ResNet and traditional CNN training is tested, and Gaussian white noise (SNR=4 dB to-4 dB) with different signal to noise ratios is added in a test set. Experiments were performed on a university of kesixi dataset based on ResNet model 18. The test results are: when the SNR is 4dB or more, the diagnostic accuracy can be maintained at 100%, and when the snr= -4dB, the high diagnostic accuracy of 93.25% can be maintained. Experiments were performed on a university of pamphlet dataset using the CNN model. Under the noise environment with SNR= -4dB, the diagnosis precision of 92.78% can be achieved for real damage data. Under the same conditions, the diagnostic accuracy of 73.99% can be achieved for complex data of the artificial injury. The measured values of this example are higher than those of other comparative methods, such as center loss and focus loss. From this, it can be seen that this embodiment can effectively enhance the robustness of the model to external disturbances. The above provides only a part of the test results of the present embodiment, not all of the test results. In fact, on the basis of the above, the present embodiment further performs a fault diagnosis performance test based on different image classification models and different loss function methods, and a visual analysis of an attention map, and the results indicate that: the present embodiment has a good performance regardless of the diagnostic accuracy, stability, or interpretability.
As shown in fig. 5, the present embodiment also provides a fault diagnosis system based on characterization guidance, including:
the sample acquisition module is used for constructing a model input sample by using rolling bearing vibration signal data;
the class activation map acquisition module acquires a feature map output by the last convolution layer of the model feature extractor, and acquires positive and negative class activation maps of an input sample by adopting a class activation mapping method;
The model training module is used for constructing a diagnosis model, and the diagnosis model consists of a convolution feature extractor, a classifier and the like; the difference between the positive class activation graph and the feature graph of the training sample is calculated as a positive class activation loss. Meanwhile, calculating the difference between a negative class activation graph and an inhibition constant of an input sample to be used as negative class activation loss; the weighted combination of the positive class activation loss, the negative class activation loss and the classification loss forms a final training loss function to optimize the model;
The fault diagnosis module is used for performing interpretable fault diagnosis on the input sample by using the model trained by the method.
The present system is based on the same inventive concept as the method shown in fig. 1, and thus reference is made to the relevant description of the previous method embodiments for specific processing operations of the respective modules.
In summary, the present invention uses attention seeking to guide the activation area of the target class activation graph by introducing the attention mechanism, thereby driving the model to associate the target class with the salient area. In addition, feature expression of the model to non-target classes is suppressed by minimizing redundant activation of the non-target class activation graph. The model can be helped to focus on the target category in the training process, and more reliable characteristic representation is automatically learned, so that the model diagnosis precision and stability are improved, and meanwhile, the decision is more visual interpretable.
The above description is only of the preferred embodiments of the present invention and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (10)
1. A bearing fault diagnosis method based on characterization guidance, comprising:
constructing a fault diagnosis model, wherein the fault diagnosis model comprises a feature extractor and a classifier which are sequentially connected in series;
Obtaining vibration signals of a normal bearing and a fault bearing to construct a training sample set, and formulating a training strategy to train the fault diagnosis model;
the training strategy is:
Extracting a feature image output by the last convolution layer of the feature extractor, and acquiring positive and negative class activation images of an input sample by adopting a class activation mapping method; calculating the difference between the positive class activation graph and the feature graph of the training sample to be used as positive class activation loss; meanwhile, calculating the difference between a negative class activation graph and an inhibition constant of an input sample to be used as negative class activation loss; carrying out weighted fusion on the negative class activation loss, the positive class activation loss and the classification loss of the classifier, and optimizing model parameters of the fault diagnosis model according to a weighted fusion loss function;
And obtaining a vibration signal of the target bearing to construct a detection sample, and inputting the detection sample into a trained fault diagnosis model to obtain the fault type of the target bearing.
2. The bearing fault diagnosis method based on characterization guidance according to claim 1, wherein the feature extractor comprises a plurality of serially connected residual layers in turn, each residual layer comprises a plurality of serially connected residual blocks in turn, each residual block comprises a plurality of serially connected convolution normalization units in turn, and each convolution normalization unit comprises serially connected convolution layers in turn and a batch normalization layer; and in N residual layers positioned at the tail part of the series connection, a channel attention module is connected in the last batch normalization layer of each residual block.
3. The bearing fault diagnosis method based on characterization guidance of claim 2, wherein the feature extractor comprises 4 residual layers, each residual layer comprising 2 residual blocks, each residual block comprising 2 convolution normalization units; and in the last 2 residual layers positioned at the tail part of the series connection, a channel attention module is connected in the last batch normalization layer of each residual block.
4. The characterization guidance-based bearing fault diagnosis method according to claim 3, wherein the channel attention module comprises an average pooling layer, a maximum pooling layer, a first fully-connected layer, a second fully-connected layer, a first activation layer, and a second activation layer, wherein the output ends of the average pooling layer and the maximum pooling layer are connected with the input end of the first fully-connected layer, the output end of the first fully-connected layer is connected with the input end of the first activation layer, and the output end of the first activation layer is connected with the input end of the second fully-connected layer; the output end of the second full-connection layer is connected with the input end of the second activation layer.
5. The method for diagnosing bearing faults based on characterization guidance of claim 4 in which the first activation layer employs a ReLU activation function and the second activation layer employs a Sigmoid function.
6. The method for diagnosing bearing faults based on characterization guidance of any of claims 1 to 5 in which the positive class activation loss satisfies:
；
wherein, For a positive class of activation loss,In order to pay attention to the height of the force diagram,In order to pay attention to the width of the force diagram,A class activation graph for a positive class,Locating coordinates for pixel points in the class activation map and attention map; AM is an attention map;
The negative class activation loss satisfies:
；
wherein, For a negative class of activation loss,For the real label of the input image,In order to be a set of tags,For the set of all negative class labels:， Is the serial number of the negative type label, Is a suppression constant.
7. The characterization guidance-based bearing fault diagnosis method according to claim 6, wherein the classification loss satisfies:
；
In order to classify the loss value(s), In order to be able to predict the value,To be a true value of the value,As a point of time in time it is,Sample tags are numbered.
8. The bearing fault diagnosis method based on characterization guidance according to any one of claims 1 to 5, wherein the detection sample and the training sample are single-channel time-frequency diagrams of the bearing, and the fault category comprises a fault location category and a damage depth category; the fault position categories are divided into an outer ring and an inner ring;
Obtaining vibration signals of a normal bearing and a fault bearing to construct a training sample set comprises the following steps:
obtaining vibration signals of various known fault categories from historical data, and extracting a single-channel time-frequency diagram from each vibration signal;
and constructing a training sample set by taking the fault category corresponding to each single-channel time-frequency diagram as a label.
9. A computer system comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the steps of any of the methods of the preceding claims 1 to 8 when the computer program is executed by the processor.
10. A computer storage medium having stored thereon a computer program, characterized in that the program when executed by a processor realizes the steps of the method of any of the preceding claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410985317.9A CN118520357B (en) | 2024-07-23 | 2024-07-23 | Bearing fault diagnosis method, system and storage medium based on characterization guidance |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410985317.9A CN118520357B (en) | 2024-07-23 | 2024-07-23 | Bearing fault diagnosis method, system and storage medium based on characterization guidance |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN118520357A CN118520357A (en) | 2024-08-20 |
| CN118520357B true CN118520357B (en) | 2024-09-24 |
Family
ID=92285475
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410985317.9A Active CN118520357B (en) | 2024-07-23 | 2024-07-23 | Bearing fault diagnosis method, system and storage medium based on characterization guidance |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118520357B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114821097A (en) * | 2022-04-07 | 2022-07-29 | 西南交通大学 | Multi-scale feature image classification method based on transfer learning |
| CN114970605A (en) * | 2022-05-06 | 2022-08-30 | 大连理工大学 | Multi-mode feature fusion neural network refrigeration equipment fault diagnosis method |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US12051229B2 (en) * | 2021-01-14 | 2024-07-30 | Tata Consultancy Services Limited | System and method for attention-based surface crack segmentation |
| WO2024065777A1 (en) * | 2022-09-30 | 2024-04-04 | Siemens Aktiengesellschaft | Method, apparatus, electronic device, and storage medium for diagnosing industrial fault |
| CN118312879B (en) * | 2024-04-08 | 2024-09-24 | 南京理工大学 | Proportion servo valve fault diagnosis method based on attention convolution capsule network |
-
2024
- 2024-07-23 CN CN202410985317.9A patent/CN118520357B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114821097A (en) * | 2022-04-07 | 2022-07-29 | 西南交通大学 | Multi-scale feature image classification method based on transfer learning |
| CN114970605A (en) * | 2022-05-06 | 2022-08-30 | 大连理工大学 | Multi-mode feature fusion neural network refrigeration equipment fault diagnosis method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN118520357A (en) | 2024-08-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112763214B (en) | Rolling bearing fault diagnosis method based on multi-label zero-sample learning | |
| CN112906644B (en) | Mechanical fault intelligent diagnosis method based on deep migration learning | |
| CN110334764B (en) | Rotary machine intelligent fault diagnosis method based on integrated depth self-encoder | |
| CN110110768B (en) | Rolling bearing fault diagnosis method based on parallel feature learning and multiple classifiers | |
| CN111753985B (en) | Image deep learning model testing method and device based on neuron coverage rate | |
| CN112308147B (en) | Rotary machinery fault diagnosis method based on multi-source domain anchor adapter integrated migration | |
| CN111562108A (en) | Rolling bearing intelligent fault diagnosis method based on CNN and FCMC | |
| CN111539152B (en) | Rolling bearing fault self-learning method based on two-stage twin convolutional neural network | |
| Jiang et al. | A machine vision-based realtime anomaly detection method for industrial products using deep learning | |
| CN116894187A (en) | Gear box fault diagnosis method based on deep migration learning | |
| CN114429152A (en) | Rolling bearing fault diagnosis method based on dynamic index antagonism self-adaption | |
| CN111275108A (en) | Method for performing sample expansion on partial discharge data based on generation countermeasure network | |
| CN114997211A (en) | Cross-working-condition fault diagnosis method based on improved countermeasure network and attention mechanism | |
| CN116401603A (en) | Multi-mode bearing fault intelligent diagnosis method based on transfer learning | |
| CN116702076A (en) | Small sample migration learning fault diagnosis method, system, computer and storage medium based on CNN feature fusion | |
| CN112836719A (en) | Indicator diagram similarity detection method fusing two classifications and three groups | |
| Han et al. | A novel intelligent fault diagnosis method based on dual convolutional neural network with multi-level information fusion | |
| Lee et al. | Deep convolutional neural network with new training method and transfer learning for structural fault classification of vehicle instrument panel structure | |
| CN118520357B (en) | Bearing fault diagnosis method, system and storage medium based on characterization guidance | |
| CN117191396A (en) | Gear box fault diagnosis method based on two-stage migration | |
| CN115935187B (en) | Nuclear sensitivity alignment network-based mechanical fault diagnosis method under variable working conditions | |
| CN114383846B (en) | Bearing composite fault diagnosis method based on fault label information vector | |
| CN111783941A (en) | Mechanical equipment diagnosis and classification method based on probability confidence degree convolutional neural network | |
| CN116541771A (en) | Unbalanced sample bearing fault diagnosis method based on multi-scale feature fusion | |
| CN116399592A (en) | Bearing fault diagnosis method based on channel attention dual-path feature extraction |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |