Rolling bearing fault diagnosis method and system based on improved GAN-OSNet
By improving the GAN-OSNet method, data expansion and feature fusion of bearing fault samples is solved, and the accuracy and efficiency of bearing fault diagnosis under sample imbalance is achieved, achieving high-precision and low-cost fault diagnosis effect.
Patent Information
- Application Number
- CN202210572898.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-05-25
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2042-05-25
AI Technical Summary
The existing bearing fault diagnosis methods are insufficiently accurate and cost-effective when the sample is unbalanced, and fail to effectively integrate multi-scale features, affecting diagnostic accuracy and efficiency.
Using the improved GAN-OSNet method, data augmentation of a few types of fault samples is generated by generating an adversarial network, and an improved full-scale network is built, and the hollow space convolution pooling pyramid module, multi-head attention mechanism and CoordinateAttention module are used to form the most Optimal fault diagnosis model.
Achieve high-precision and low-cost bearing fault diagnosis in the case of unbalanced data samples, which improves diagnostic efficiency and feature extraction capabilities, and reduces fault diagnosis costs.
Smart Images

Figure CN114858455B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of bearing fault diagnosis, and in particular to a rolling bearing fault diagnosis method and system based on improved GAN-OSNet. Background Art
[0002] As the power source of many systems, rotating machinery is widely used in high-end equipment manufacturing fields such as aerospace, shipbuilding, and automobiles, and plays an extremely important role in modern industry. As the key part of rotating machinery, rolling bearings are prone to failures in long-term operation under complex working environments such as high intensity and variable loads. If the failure cannot be detected in a timely and accurate manner, it will cause huge economic losses and immeasurable consequences. Therefore, the research on rolling bearing fault diagnosis is of great significance to ensure the safety and reliability of equipment and reduce operation and maintenance costs, and can provide strong support for the maintenance decision-making management of subsequent equipment.
[0003] At present, the mainstream fault diagnosis methods can be divided into fault diagnosis based on signal analysis, fault diagnosis based on traditional machine learning, and fault diagnosis based on deep learning. With the rapid development of deep learning and the increasing amount of multi-source heterogeneous big data that can be collected on equipment operation, fault diagnosis based on deep learning has become the most widely used modeling method.
[0004] However, most of the existing fault diagnosis methods are aimed at situations where the fault samples are relatively ideal, and rarely involve fault diagnosis under non-ideal data conditions such as class imbalance, which is inconsistent with the actual situation and affects the diagnostic accuracy. In addition, most of the existing methods rarely study the effective fusion of multi-scale features extracted from samples, which affects the classification performance of the fault diagnosis model and the diagnostic accuracy is not high. In addition, the existing large-scale neural networks used for fault diagnosis are often accompanied by a large number of parameters and complex calculations, resulting in high fault diagnosis costs and low efficiency. It can be seen that there is no high-precision, low-cost and efficient bearing fault diagnosis method in the existing technology. Summary of the invention
[0005] 1. Technical issues to be resolved
[0006] In view of the shortcomings of the prior art, the present invention provides a rolling bearing fault diagnosis method and system based on improved GAN-OSNet, which solves the problems of insufficient accuracy, high cost and low efficiency when the sample is unbalanced.
[0007] (II) Technical solution
[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions:
[0009] In the first aspect, the present invention first proposes a rolling bearing fault diagnosis method based on improved GAN-OSNet, the method comprising:
[0010] Based on an improved generative adversarial network, data expansion is performed on real fault sample data whose sample quantity is less than a preset threshold; the generative network of the improved generative adversarial network is composed of two fully connected layers and two LSTM layers;
[0011] An improved full-scale network is constructed, and the improved full-scale network is trained to obtain an optimal fault diagnosis model; the improved full-scale network includes a main part of the original full-scale network stacked with multiple improved residual bottleneck structures; each of the improved residual bottleneck structures includes: a dilated spatial convolution pooling pyramid module and a multi-head attention mechanism added to the original residual bottleneck structure, and a CA module is used to replace the AG fusion mechanism in the original residual bottleneck structure;
[0012] The optimal fault diagnosis model is used to perform rolling bearing fault diagnosis.
[0013] Preferably, the data expansion of the acquired real fault sample data whose sample quantity is less than a preset threshold based on the improved generative adversarial network includes:
[0014] S11, collecting vibration signals during the operation of the equipment as real sample data, wherein the real sample data includes real normal sample data and real fault sample data; then assigning different labels to the real normal sample data and each different category of real fault sample data, and dividing the real sample data after label assignment into a real sample data training set and a real sample data test set;
[0015] S12. Using an improved generative adversarial network, data expansion is performed on real fault sample data in the real sample data training set whose sample number is less than a preset threshold, to obtain a generated fault sample data set.
[0016] Preferably, the training of the improved full-scale network to obtain an optimal fault diagnosis model comprises:
[0017] The real sample data training set and the generated fault sample data set are used as a mixed training set to train the improved full-scale network to obtain an optimal fault diagnosis model.
[0018] Preferably, when the real sample data training set and the generated fault sample data set are used as a mixed training set, it also includes: using a reshape function to reshape the one-dimensional real sample data training set and the one-dimensional generated fault sample data set into two-dimensional data to form the mixed training set.
[0019] Preferably, the number of the improved residual bottleneck structures is 6.
[0020] In a second aspect, the present invention also proposes a rolling bearing fault diagnosis system based on improved GAN-OSNet, the system comprising:
[0021] A data expansion module, used for performing data expansion on real fault sample data whose sample quantity is less than a preset threshold based on an improved generative adversarial network; the generative network of the improved generative adversarial network consists of two fully connected layers and two LSTM layers;
[0022] A fault diagnosis model optimization module is used to construct an improved full-scale network and train the improved full-scale network to obtain an optimal fault diagnosis model; the improved full-scale network includes a main part of the original full-scale network stacked with multiple improved residual bottleneck structures; each of the improved residual bottleneck structures includes: a dilated spatial convolution pooling pyramid module and a multi-head attention mechanism added to the original residual bottleneck structure, and a CA module to replace the AG fusion mechanism in the original residual bottleneck structure;
[0023] The fault diagnosis module is used to perform rolling bearing fault diagnosis using the optimal fault diagnosis model.
[0024] Preferably, the data expansion module performs data expansion on real fault sample data whose sample quantity is less than a preset threshold based on the improved generative adversarial network, including:
[0025] S11, collecting vibration signals during the operation of the equipment as real sample data, wherein the real sample data includes real normal sample data and real fault sample data; then assigning different labels to the real normal sample data and each different category of real fault sample data, and dividing the real sample data after label assignment into a real sample data training set and a real sample data test set;
[0026] S12. Using an improved generative adversarial network, data expansion is performed on real fault sample data in the real sample data training set whose sample number is less than a preset threshold, to obtain a generated fault sample data set.
[0027] Preferably, in the fault diagnosis model optimization module, training the improved full-scale network to obtain the optimal fault diagnosis model includes:
[0028] The real sample data training set and the generated fault sample data set are used as a mixed training set to train the improved full-scale network to obtain an optimal fault diagnosis model.
[0029] Preferably, when the real sample data training set and the generated fault sample data set are used as a mixed training set, it also includes: using a reshape function to reshape the one-dimensional real sample data training set and the one-dimensional generated fault sample data set into two-dimensional data to form the mixed training set.
[0030] Preferably, the number of the improved residual bottleneck structures is 6.
[0031] (III) Beneficial effects
[0032] The present invention provides a rolling bearing fault diagnosis method and system based on improved GAN-OSNet. Compared with the prior art, it has the following beneficial effects:
[0033] 1. The present invention improves the existing generative adversarial network, and then uses the improved generative adversarial network to perform data expansion on the minority class fault sample data in the acquired real sample data to balance the data set; at the same time, an improved full-scale network is constructed, and the improved full-scale network is trained to obtain the optimal fault diagnosis model; finally, the above optimal fault diagnosis model is used to perform rolling bearing fault diagnosis. The present invention can perform high-precision, low-cost and high-efficiency diagnosis of rolling bearings under non-ideal data conditions such as unbalanced data samples, providing strong support for subsequent equipment maintenance decision management.
[0034] 2. The present invention integrates the LSTM network into the generator of the GAN network, and uses the improved GAN-LSTM network to expand the acquired fault sample data in a small amount. On the one hand, the GAN network is used to increase the amount of data, solving the problem of unbalanced proportion of training sample categories; on the other hand, the LSTM network has good processing characteristics for time series data, taking into account the time series characteristics of fault data, thereby optimizing the quality of generated samples. The present invention uses the GAN-LSTM network to obtain a large number of high-quality fault sample data, providing high-quality data support for subsequent fault classification.
[0035] 3. The present invention uses the full-scale network (OSNet) as the backbone of the bearing fault classification network, and improves the main structure of each residual bottleneck in OSNet, adds the atrous spatial convolutional pooling pyramid module (ASPP), a multi-head attention mechanism, and uses CoordinateAttention (CA) to replace the original AG fusion mechanism, which can better focus on key features that are helpful to the classification results, improve the network's feature extraction capability and the fusion effect of multiple features, and make the bearing fault diagnosis results based on this more accurate.
[0036] 4. The GAN-OSNet proposed in the present invention is a lightweight network. Compared with the existing mainstream fault diagnosis model based on deep learning, its model parameters are one order of magnitude smaller and the amount of calculation is smaller, but the fault diagnosis accuracy of the network can still reach or even exceed the existing mainstream model. It can be seen that it has good generalization ability and improves the fault diagnosis efficiency while reducing the fault diagnosis cost. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0038] Figure 1 It is a flow chart of the rolling bearing fault diagnosis method based on the improved GAN-OSNet in the present invention;
[0039] Figure 2 This is an embodiment diagram of a rolling bearing fault diagnosis method based on improved GAN-OSNet in an embodiment of the present invention;
[0040] Figure 3 This is a structural diagram of an improved GAN in an embodiment of the present invention;
[0041] Figure 4 It is a structural diagram of the improved OSNet in the embodiment of the present invention;
[0042] Figure 5 : is a diagram of the improved residual bottleneck structure in OSNet in an embodiment of the present invention;
[0043] Figure 6 It is the original residual bottleneck structure diagram in the prior art;
[0044] Figure 7 It is a diagram of the Lite3×3 structure in an embodiment of the present invention;
[0045] Figure 8 FIG. 4 is a structural diagram of ASPP in an embodiment of the present invention. DETAILED DESCRIPTION
[0046] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0047] The embodiment of the present application provides a rolling bearing fault diagnosis method and system based on improved GAN-OSNet, which solves the problems of insufficient accuracy, high cost and low efficiency in the bearing fault diagnosis method in the prior art when the sample is unbalanced, thereby achieving the purpose of accurately and efficiently completing bearing fault diagnosis at only a relatively low cost.
[0048] The technical solution in the embodiment of the present application is to solve the above technical problems, and the overall idea is as follows:
[0049] In order to solve the problems of low accuracy, high cost and low diagnostic efficiency in existing bearing fault diagnosis technology when the samples are unbalanced, the present invention first improves the existing generative adversarial network to facilitate data expansion of minority class fault sample data in the acquired real sample data to balance the data set, and then improves the full-scale network, and trains and optimizes the improved full-scale network based on the expanded data to obtain the optimal fault diagnosis model, and finally the optimal fault diagnosis model can be used to perform rolling bearing fault diagnosis.
[0050] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.
[0051] Embodiment 1:
[0052] In the first aspect, the present invention first proposes a rolling bearing fault diagnosis method based on improved GAN-OSNet, see Figure 1 , the method comprising:
[0053] S1. Based on an improved generative adversarial network, data expansion is performed on real fault sample data whose sample quantity is less than a preset threshold; the generative network of the improved generative adversarial network is composed of two fully connected layers and two LSTM layers;
[0054] S2. construct an improved full-scale network, and train the improved full-scale network to obtain an optimal fault diagnosis model; the improved full-scale network includes stacking the main part of the original full-scale network with multiple improved residual bottleneck structures; each of the improved residual bottleneck structures includes: a dilated spatial convolution pooling pyramid module and a multi-head attention mechanism added to the original residual bottleneck structure, and a CA module to replace the AG fusion mechanism in the original residual bottleneck structure;
[0055] S3. diagnose rolling bearing faults using the optimal fault diagnosis model.
[0056] It can be seen that this embodiment improves the existing generative adversarial network, and then uses the improved generative adversarial network to perform data expansion on the minority class fault sample data in the acquired real sample data to balance the data set; at the same time, an improved full-scale network is constructed, and the improved full-scale network is trained to obtain the optimal fault diagnosis model; finally, the rolling bearing fault diagnosis is performed using the above optimal fault diagnosis model. The rolling bearing fault diagnosis method of this embodiment can perform high-precision, low-cost and high-efficiency diagnosis on rolling bearings under non-ideal data conditions such as unbalanced data samples, providing strong support for the maintenance decision-making management of subsequent equipment.
[0057] The following is combined with Figure 1-8 , and explanations of specific steps S1-S3 are provided to describe in detail the implementation process of an embodiment of the present invention.
[0058] S1. Based on an improved generative adversarial network, data expansion is performed on real fault sample data whose sample number is less than a preset threshold; the generative network of the improved generative adversarial network consists of two fully connected layers and two LSTM layers.
[0059] S11. Collect vibration signals during the operation of the equipment as real sample data, wherein the real sample data includes real normal sample data and real fault sample data; then assign different labels to the real normal sample data and each different category of real fault sample data, and divide the labeled real sample data into a real sample data training set and a real sample data test set.
[0060] 1.1) The vibration signals generated during the operation of the equipment are collected in chronological order to obtain real sample data. The real sample data includes real normal sample data and real fault sample data. Since most of the data collected during the operation of the equipment is normal data, and only a small amount of data is fault data, the real normal sample data in the real sample data collected above is far more than the real fault sample data, that is to say, the real sample data collected is an unbalanced data set. When collecting real sample data, the sliding window sampling technology is used for sampling, and the overlapping sampling is used to increase the number of samples and ensure the difference of the samples to obtain an unbalanced data set.
[0061] 1.2) Different labels are assigned to the collected real normal sample data and the real fault sample data of each category (all one-dimensional data), and then the entire real sample data is divided into a real sample data training set and a real sample data test set. When assigning different labels to the one-dimensional sample data, the normal data and each specific fault type are labeled, for example, ten categories (one normal + 9 fault types), the normal data label is 0, and the fault type label is 1-9.
[0062] S12. Using an improved generative adversarial network, data expansion is performed on real fault sample data corresponding to a smaller number of fault categories in the real sample data training set to obtain a generated fault sample data set.
[0063] Improve the original generative adversarial network (GAN network), that is, construct a GAN-LSTM network framework, and use the GAN-LSTM network framework to perform data expansion on the real fault sample data corresponding to the fault categories with a smaller number in the real sample data training set after the above processing. If the number of real fault sample data of all fault categories in the real sample data training set is small (for example, less than a certain preset number threshold, the threshold can be flexibly set according to actual needs), then data expansion is performed on the real fault sample data corresponding to all fault categories. Specifically,
[0064] 2.1) Construct an improved generative adversarial network, namely the GAN-LSTM network framework. Figure 3 In this embodiment, the GAN-LSTM network framework is composed of a generator network G and a discriminator network D, and the generator network G is composed of two fully connected layers and two LSTM layers; the discriminator network D is composed of a plurality of overall stacks of linear layers + BN layers + activation functions + dropout combinations.
[0065] 2.2) Train the improved generative adversarial network and perform data augmentation.
[0066] 2.2.1) Train the generative network G and discriminative network D in the improved generative adversarial network.
[0067] The randomly generated noise vector z is used as input, and the generated fault sample G(z) is obtained through the generative network G. Gaussian noise is added to the generated fault sample G(z) to improve sample diversity and avoid the generated sample being single and resulting in low quality. The training set of real fault samples and the generated fault samples are input into the discriminant network D together, and the discriminant score is output.
[0068] 2.2.2) The cross entropy loss function is optimized and solved using the Adam optimization algorithm. Based on the adversarial mechanism, the network parameters of the improved generative adversarial network are continuously trained until the loss function converges to the optimal state, thereby obtaining the trained generative network G and discriminative network D.
[0069] 2.2.3) The randomly generated noise vector z is input into the trained generative network G to generate fault data, thereby expanding the minority class fault data to balance the data set and provide data support for subsequent fault classification.
[0070] S2. Construct an improved full-scale network (OSNet network), and use a mixed training set to train the improved OSNet network to obtain an optimal fault diagnosis model; the mixed training set includes the real sample data training set and the generated fault sample data set; the improved OSNet network includes the main part of the original OSNet network stacked with multiple improved residual bottleneck structures; each of the improved residual bottleneck structures includes: an atrous spatial convolutional pooling pyramid module (ASPP) and a multi-head attention mechanism (multi-head attention) added to the original residual bottleneck structure, and a CoordinateAttention (CA) module is used to replace the original AG fusion mechanism in the original residual bottleneck structure.
[0071] 1) Build an improved full-scale network (OSNet network).
[0072] In order to further improve the feature extraction capability of the fault classification model and enable the model to better focus on key features that are helpful to the classification results; at the same time, in order to simplify the complexity of model parameters and reduce the amount of calculation, in this embodiment, an improved full-scale network (OSNet network) framework is constructed.
[0073] See also Figure 4 In the improved OSNet network of this embodiment, when performing dynamic fusion of multi-scale features, first, a 7×7 ordinary convolution layer is used for preliminary feature extraction, and then the main part is stacked by multiple improved residual bottleneck structures, and finally, after global average pooling, the classification result is output through the fully connected layer. Preferably, in this embodiment, the number of improved residual bottleneck structures is set to 6. Figure 5 , Figure 5 The improved residual bottleneck structure in this embodiment is improved. The main structure of each improved residual bottleneck structure includes: a dilated spatial convolutional pooling pyramid module (ASPP), 4 lite3×3 feature streams, a multi-head attention mechanism, a Coordinate Attention (CA) module, and a residual structure. Figure 6 is the original residual bottleneck structure before improvement. Figure 7 This is the structural diagram of Lite3×3 in the residual bottleneck structure.
[0074] Add the Atrous Spatial Convolution Pooling Pyramid (ASPP) module to the bottleneck, such as Figure 8The ASPP module uses dilated convolutions with different expansion factors (rates) to capture different scale features of data at multiple scales, which can expand the receptive field of the convolution kernel to achieve free multi-scale feature extraction, and effectively fuse shallow and deep features to obtain more powerful feature expression and improve classification accuracy.
[0075] A multi-head attention mechanism is added to the bottleneck to focus on the correlation of features at different scales. Since the outputs of the four feature streams are extracted at different depths, the correlation of features at different scales is not taken into account. Therefore, the present invention uses self-attention, an important component of the transformer, to perform feature correlation, which is beneficial to the effectiveness of feature extraction and focuses more on key features that are helpful for classification, thereby improving classification accuracy.
[0076] The Coordinate Attention (CA) module is used to replace the original AG fusion mechanism. Due to the original AG fusion method, it only multiplies the corresponding weights for each channel of the feature map, and only re-measures the importance of each channel, while ignoring the impact of the position information in the same feature map on the classification effect. After replacing the original AG fusion mechanism with the Coordinate Attention (CA) module, not only the relationship between the channels is considered, but also the position information of the feature space. It can encode the spatial information in the horizontal and vertical directions, so that the network can pay attention to a wide range of position information, which helps the model to better extract key features without bringing too much calculation, and is flexible and lightweight enough.
[0077] When performing fault classification, the working process of the improved full-scale network (OSNet network) is as follows: the input of each improved residual bottleneck first passes through the atrous spatial convolution pooling pyramid module (ASPP) to capture the different scale features of the data, and then uses the four feature streams of lite3×3 for full-scale feature extraction; then the four output feature maps are input into the multi-head attention mechanism for feature correlation, and the CA module is used to replace the original AG fusion mechanism for dynamic feature fusion to better extract key features; the fused feature map is then restored to the number of channels at the time of input through a 1×1 convolution, and is added to the input through a residual structure as the output of the residual bottleneck module.
[0078] 2) Train and optimize the improved full-scale network (OSNet) mentioned above.
[0079] 2.1) The acquired real sample data training set and the above-mentioned generated fault sample data set are mixed, and the one-dimensional sample data is reshaped into two-dimensional data by reshaping as a new training set for the improved OSNet classification network. Each time, samples with a batch size of batch-size are used as the input of the classification network.
[0080] 2.2) After the input sample undergoes preliminary feature extraction in the 7×7 convolutional layer and activation function relu, an activation feature map of a certain size is output, and then multi-scale feature extraction and fusion are performed through 6 residual bottleneck structures in sequence. Finally, after a 1*1 convolution and global average pooling, the data is flattened and input into the fully connected layer to obtain the fault classification result;
[0081] 2.3) According to the classification results, the feature extraction process is continuously optimized in reverse, and the cross entropy loss function is optimized and solved using the Adam optimization algorithm until the loss function converges to the optimal value;
[0082] 2.4) Save the network parameters of the best network after training and optimization, obtain the global optimal network, and use the global optimal network as the final fault diagnosis model.
[0083] After each epoch of training, the real sample data test set is used to verify whether the current network parameters are better than the previous optimal network. If so, the current network parameters are saved as the optimal network. Otherwise, continue training for the next epoch. The network's feature extraction capability is continuously trained until the set epoch is trained, the network training is completed, and the global optimal network is obtained for accurate fault diagnosis.
[0084] S3. Use the optimal fault diagnosis model obtained after the above training to perform rolling bearing fault diagnosis.
[0085] Finally, the optimal fault diagnosis model obtained above can be used to diagnose rolling bearing faults.
[0086] At this point, the entire process of the rolling bearing fault diagnosis method based on the improved GAN-OSNet in this embodiment is completed.
[0087] Embodiment 2:
[0088] In a second aspect, the present invention also provides a rolling bearing fault diagnosis system based on improved GAN-OSNet, the system comprising:
[0089] A data expansion module, used for performing data expansion on real fault sample data whose sample quantity is less than a preset threshold based on an improved generative adversarial network; the generative network of the improved generative adversarial network consists of two fully connected layers and two LSTM layers;
[0090] A fault diagnosis model optimization module is used to construct an improved full-scale network and train the improved full-scale network to obtain an optimal fault diagnosis model; the improved full-scale network includes a main part of the original full-scale network stacked with multiple improved residual bottleneck structures; each of the improved residual bottleneck structures includes: a dilated spatial convolution pooling pyramid module and a multi-head attention mechanism added to the original residual bottleneck structure, and a CA module to replace the AG fusion mechanism in the original residual bottleneck structure;
[0091] The fault diagnosis module is used to perform rolling bearing fault diagnosis using the optimal fault diagnosis model.
[0092] Optionally, the data expansion module performs data expansion on real fault sample data whose sample quantity is less than a preset threshold based on the improved generative adversarial network, including:
[0093] S11, collecting vibration signals during the operation of the equipment as real sample data, wherein the real sample data includes real normal sample data and real fault sample data; then assigning different labels to the real normal sample data and each different category of real fault sample data, and dividing the real sample data after label assignment into a real sample data training set and a real sample data test set;
[0094] S12. Using an improved generative adversarial network, data expansion is performed on real fault sample data in the real sample data training set whose sample number is less than a preset threshold, to obtain a generated fault sample data set.
[0095] Optionally, in the fault diagnosis model optimization module, training the improved full-scale network to obtain an optimal fault diagnosis model includes:
[0096] The real sample data training set and the generated fault sample data set are used as a mixed training set to train the improved full-scale network to obtain an optimal fault diagnosis model.
[0097] Optionally, when using the real sample data training set and the generated fault sample data set as a mixed training set, it also includes: using a reshape function to reshape the one-dimensional real sample data training set and the one-dimensional generated fault sample data set into two-dimensional data to form the mixed training set.
[0098] Optionally, the number of the improved residual bottleneck structures is 6.
[0099] It can be understood that the rolling bearing fault diagnosis system based on improved GAN-OSNet provided in an embodiment of the present invention corresponds to the above-mentioned rolling bearing fault diagnosis method based on improved GAN-OSNet. The explanations, examples, beneficial effects and other parts of the relevant contents can refer to the corresponding contents in the rolling bearing fault diagnosis method based on improved GAN-OSNet, and will not be repeated here.
[0100] In summary, compared with the prior art, the present invention has the following beneficial effects:
[0101] 1. The present invention improves the existing generative adversarial network, and then uses the improved generative adversarial network to perform data expansion on the minority class fault sample data in the acquired real sample data to balance the data set; at the same time, an improved full-scale network is constructed, and the improved full-scale network is trained to obtain the optimal fault diagnosis model; finally, the above optimal fault diagnosis model is used to perform rolling bearing fault diagnosis. The present invention can perform high-precision, low-cost and high-efficiency diagnosis of rolling bearings under non-ideal data conditions such as unbalanced data samples, providing strong support for subsequent equipment maintenance decision management.
[0102] 2. The present invention integrates the LSTM network into the generator of the GAN network, and uses the improved GAN-LSTM network to expand the acquired fault sample data in a small amount. On the one hand, the GAN network is used to increase the amount of data, solving the problem of unbalanced proportion of training sample categories; on the other hand, the LSTM network has good processing characteristics for time series data, taking into account the time series characteristics of fault data, thereby optimizing the quality of generated samples. The present invention uses the GAN-LSTM network to obtain a large number of high-quality fault sample data, providing high-quality data support for subsequent fault classification.
[0103] 3. The present invention uses the full-scale network (OSNet) as the backbone of the bearing fault classification network, and improves the main structure of each residual bottleneck in OSNet, adds the atrous spatial convolutional pooling pyramid module (ASPP), a multi-head attention mechanism, and uses CoordinateAttention (CA) to replace the original AG fusion mechanism, which can better focus on key features that are helpful to the classification results, improve the network's feature extraction capability and the fusion effect of multiple features, and make the bearing fault diagnosis results based on this more accurate.
[0104] 4. The GAN-OSNet proposed in the present invention is a lightweight network. Compared with the existing mainstream fault diagnosis model based on deep learning, its model parameters are one order of magnitude smaller and the amount of calculation is smaller, but the fault diagnosis accuracy of the network can still reach or even exceed the existing mainstream model. It can be seen that it has good generalization ability and improves the fault diagnosis efficiency while reducing the fault diagnosis cost.
[0105] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.
[0106] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A rolling bearing fault diagnosis method based on improved GAN-OSNet, characterized in that: The method comprises: Based on an improved generative adversarial network, data expansion is performed on real fault sample data whose sample quantity is less than a preset threshold; the generative network of the improved generative adversarial network is composed of two fully connected layers and two LSTM layers; An improved full-scale network is constructed, and the improved full-scale network is trained to obtain an optimal fault diagnosis model; the improved full-scale network includes a main part of the original full-scale network stacked with multiple improved residual bottleneck structures; each of the improved residual bottleneck structures includes: a dilated spatial convolution pooling pyramid module and a multi-head attention mechanism added to the original residual bottleneck structure, and a CA module is used to replace the AG fusion mechanism in the original residual bottleneck structure; The optimal fault diagnosis model is used to perform rolling bearing fault diagnosis.
2. The method according to claim 1, characterized in that The method of performing data expansion on real fault sample data whose number of samples is less than a preset threshold based on the improved generative adversarial network includes: S11, collecting vibration signals during the operation of the equipment as real sample data, wherein the real sample data includes real normal sample data and real fault sample data; then assigning different labels to the real normal sample data and each different category of real fault sample data, and dividing the real sample data after label assignment into a real sample data training set and a real sample data test set; S12. Using an improved generative adversarial network, data expansion is performed on real fault sample data in the real sample data training set whose sample number is less than a preset threshold, to obtain a generated fault sample data set.
3. The method according to claim 2, characterized in that The step of training the improved full-scale network to obtain an optimal fault diagnosis model comprises: The real sample data training set and the generated fault sample data set are used as a mixed training set to train the improved full-scale network to obtain an optimal fault diagnosis model.
4. The method according to claim 3, characterized in that When the real sample data training set and the generated fault sample data set are used as a mixed training set, the method further includes: using a reshape function to reshape the one-dimensional real sample data training set and the one-dimensional generated fault sample data set into two-dimensional data to form the mixed training set.
5. The method according to claim 1, characterized in that The number of the improved residual bottleneck structures is 6.
6. A rolling bearing fault diagnosis system based on improved GAN-OSNet, characterized in that: The system comprises: A data expansion module, used for performing data expansion on real fault sample data whose sample quantity is less than a preset threshold based on an improved generative adversarial network; the generative network of the improved generative adversarial network consists of two fully connected layers and two LSTM layers; A fault diagnosis model optimization module is used to construct an improved full-scale network and train the improved full-scale network to obtain an optimal fault diagnosis model; the improved full-scale network includes a main part of the original full-scale network stacked with multiple improved residual bottleneck structures; each of the improved residual bottleneck structures includes: a dilated spatial convolution pooling pyramid module and a multi-head attention mechanism added to the original residual bottleneck structure, and a CA module to replace the AG fusion mechanism in the original residual bottleneck structure; The fault diagnosis module is used to perform rolling bearing fault diagnosis using the optimal fault diagnosis model.
7. The system according to claim 6, characterized in that The data expansion module performs data expansion on real fault sample data whose sample quantity is less than a preset threshold based on the improved generative adversarial network, including: S11, collecting vibration signals during the operation of the equipment as real sample data, wherein the real sample data includes real normal sample data and real fault sample data; then assigning different labels to the real normal sample data and each different category of real fault sample data, and dividing the real sample data after label assignment into a real sample data training set and a real sample data test set; S12. Using an improved generative adversarial network, data expansion is performed on real fault sample data in the real sample data training set whose sample number is less than a preset threshold, to obtain a generated fault sample data set.
8. The system according to claim 7, characterized in that The fault diagnosis model optimization module trains the improved full-scale network to obtain the optimal fault diagnosis model, including: The real sample data training set and the generated fault sample data set are used as a mixed training set to train the improved full-scale network to obtain an optimal fault diagnosis model.
9. The system according to claim 8, characterized in that When the real sample data training set and the generated fault sample data set are used as a mixed training set, the method further includes: using a reshape function to reshape the one-dimensional real sample data training set and the one-dimensional generated fault sample data set into two-dimensional data to form the mixed training set.
10. The system according to claim 6, characterized in that The number of the improved residual bottleneck structures is 6.
