Rolling bearing cross-working-condition composite fault diagnosis method based on MDMAAN and MK-MMD

Abstract

The invention discloses a rolling bearing cross-working-condition coincidence fault diagnosis method based on MDMAAN and MK-MMD, and the method employs a multi-source-domain multi-head attention mechanism method, employs the multi-head attention mechanism to combine with a convolutional neural network to carry out the feature extraction of a plurality of input source domains and target domains, and carries out the weight distribution of the extracted features. And selecting a multi-core maximum mean value difference as a metric function of each pair of source domain and target domain, selecting proper parameter configuration, and constructing an MDMAAN model and a fault diagnosis process based on the parameter configuration so as to complete the composite fault diagnosis of the rolling bearing under the cross-working condition. By using rolling bearing public experimental data, experimental verification shows that compared with other deep transfer learning fault diagnosis models, the diagnosis accuracy obtained by the model is the best, and the effectiveness and the excellence of the method in rolling bearing cross-working-condition composite fault diagnosis are verified. The provided method can well solve the problem of cross-working-condition composite fault diagnosis of the rolling bearing.

Description

Technical Field

[0001] The present invention belongs to the technical field of rotating machinery fault diagnosis, and relates to a rolling bearing fault diagnosis method, and in particular to a rolling bearing cross-working condition composite fault diagnosis method based on MDMAAN and MK-MMD. Background Art

[0002] Rotating machinery is an integral part of the machinery manufacturing industry, and its proper operation is highly dependent on the health of its bearings. Rolling bearings, the hinges of machinery, are widely used in virtually all types of rotating machinery, such as transmissions, rolling mills, wind turbines, high-speed trains, and gas turbines. Compared to other mechanical components, rolling bearings are more susceptible to failure due to the harsh operating conditions of high temperatures, high speeds, and high loads they encounter over time. Failure to detect faults promptly often leads to equipment downtime, resulting in significant economic losses and even safety incidents. To reduce maintenance costs and prevent casualties, timely and accurate fault diagnosis is essential before immeasurable losses occur.

[0003] Transfer learning fault diagnosis plays a crucial role in fault prediction and health management. This is primarily due to its ability to effectively address data scarcity, imbalance, and poor generalization. In some real-world industrial applications, access to equipment failure data is often limited. Transfer learning fault diagnosis can transfer existing pre-trained models to new tasks, effectively improving model performance and accuracy.

[0004] The application of transfer learning fault diagnosis in PHM not only improves equipment operating efficiency and reduces equipment failure rates, but also saves companies significant maintenance costs and improves their economic benefits. This paper analyzes the specific application of deep transfer learning algorithms in rolling bearing fault diagnosis, improving the accuracy and generalization of fault diagnosis. This not only provides a new technical solution for the intelligent operation and maintenance of rotating machinery, but also promotes the advancement of intelligent manufacturing technology and enhances the intelligence level of mechanical equipment.

[0005] Domain adaptation is key to transfer learning. Its primary goal is to map data from the source and target domains to the same feature space, minimizing the distances in the feature space. In this way, domain adaptation helps improve model performance in new domains, thereby avoiding the domain discrepancy problem that often occurs in transfer learning. For unlabeled target domains, the main strategy is to guide feature learning by minimizing the metric loss between the source and target domain feature distributions. Metric functions are the core of feature-based deep transfer learning, and different metrics have varying effects on data matching. Metric functions include Maximum Mean Discrepancy (MMD), Kullback-Leibler Deviation, and Wasserstein Deviation, with MMD being the most widely used. Methods using MMD for domain adaptation include Domain Adaptive Neural Networks (DaNN), Deep Domain Confusion (DDC), and Domain Adaptive Networks (DAN). Zhang et al. used a sliding framework to collect samples and applied them to a transfer learning neural network to improve fault classification performance under different operating conditions. Jiang et al. applied multi-feature fusion and a joint adaptive transfer learning network to achieve bearing fault diagnosis. Han et al. used adversarial learning to guide the feature generator to provide transferable features to achieve rolling bearing fault diagnosis.

[0006] Currently, most transfer learning-based fault diagnosis research focuses on single-source domain transfer. However, in real-world scenarios, labeled data is often collected from multiple different source domains. This data not only has a different distribution from the target domain but also from each other, resulting in multiple source domains. Therefore, neither merging all source domains nor selecting a single source domain can achieve ideal fault diagnosis results. Multi-source domain adaptive fault diagnosis methods have emerged. Yu et al. proposed a multi-source domain adaptive method with strong generalization and robustness. The fault diagnosis method based on this method is well-suited for fault diagnosis tasks under variable conditions. Zhuang et al. proposed a new multi-source domain adaptive method, the Deep Negative Correlation Multi-source Domain Adaptation Network, for mechanical fault diagnosis under different operating conditions. Huang et al. designed a joint loss for aligning the distribution of feature and label information to mitigate the negative impact of inconsistent label spaces. Yang et al. used SSAE and K-means to extract better domain-invariant and discriminative features from multi-source domain data, resulting in a fault diagnosis model that outperforms single-source domain transfer. Xu et al. proposed a multi-source unsupervised domain adaptation network. This network uses a feature extractor to learn feature representations and constructs a domain discriminator for each source domain. Through adversarial training, domain-invariant features are learned to diagnose fault types in the target domain. Zhang et al. organically combined data from multiple source domains, proposed a dual alignment mechanism, and reduced the prediction inconsistency of multiple classifiers. The method demonstrated excellent cross-domain diagnostic performance in multiple experimental cases. To achieve multi-domain adaptation, Wang et al. proposed a domain-specific distribution alignment module that uses intra-domain and inter-domain alignment strategies to reduce the offset between the source and target domains. Wei et al. proposed a weighted domain adaptation network. This network assigns weights to different source domains by measuring the MMD loss between each source and target domain. A threshold for domain adaptation is also set to prevent negative transfer. The effectiveness of multi-source domain adaptation is verified in case studies across speed and load. A literature review reveals that multi-source domain adaptation fault diagnosis methods have significant advantages when processing data from multiple different source domains. However, the challenge of properly assigning weights to multiple source domains to further improve diagnostic accuracy remains to be addressed.

[0007] In summary, domain adaptation has been extensively experimented with and applied in the field of fault diagnosis, and can effectively perform fault diagnosis. However, in practical applications, fault diagnosis may involve a variety of different fault types. Single-source domain adaptation often ignores this diversity, which may reduce the accuracy of fault diagnosis. This paper takes the perspective of multi-source domain adaptive fault diagnosis and combines the multi-head attention mechanism to assign reasonable weights to multiple source domain inputs. Multi-kernel Maximum Mean Discrepancy (MK-MMD) is used to measure the loss between each source domain and the target domain. A Multi-source Domain Multi-head Attention Adaptation Network (MDMAAN) is constructed to explore the effectiveness and generalization ability of this model in the field of fault diagnosis. Summary of the Invention

[0008] The purpose of the present invention is to construct a multi-source domain multi-head attention adaptive network model by optimizing the source domain classification loss, domain invariant representation and the difference between classifiers, and to combine the multi-source domain multi-head attention adaptive network with the multi-core maximum mean difference to provide a rolling bearing cross-working condition composite fault diagnosis method based on MDMAAN and MK-MMD.

[0009] The purpose of the present invention is achieved through the following technical solutions:

[0010] A rolling bearing cross-operating condition composite fault diagnosis method based on MDMAAN and MK-MMD includes the following steps:

[0011] Step 1: Divide the dataset into multiple source domain data and target domain data according to different working conditions, and divide them into training sets and test sets; extract common features from the input data and map the extracted original features to the common feature space; splice them together after processing by the attention mechanism, and obtain the output of the multi-head attention mechanism through linear transformation; assign weights to the extracted features, perform specific feature extraction on the assigned features and map them to the feature space;

[0012] Step 2: Minimize the distribution difference between the source and target domains in the feature space through multi-kernel maximum mean difference (MK-MMD) and analyze the domain adaptation loss; analyze the classification loss of the source domain through the source domain label, analyze the classification loss of the prediction results of the target domain, analyze the total loss and update the MDMAAN model through backpropagation;

[0013] Step 3: The MDMAAN model consists of two feature extraction units and two fault diagnosis units. The fault diagnosis unit consists of multiple Softmax classifiers, and the total loss is composed of the source domain classification loss, the MK-MMD loss, and the classifier classification loss. The dataset is classified according to different working conditions and input into the trained MDMAAN model. The diagnostic effect is evaluated using the test set data and the target domain labels to complete the MDMAAN cross-working condition composite fault diagnosis.

[0014] Compared with the prior art, the present invention has the following advantages:

[0015] 1. The MDMAAN model has high diagnostic accuracy, especially for complex fault diagnosis of rolling bearings. It can use laboratory data to realize cross-operating fault diagnosis in actual working scenarios. The MDMAAN model also has good comprehensive diagnostic effect, with high identification accuracy and robustness.

[0016] 2. In the diagnosis results of the MDMAAN model, only a very small number of fault categories were misclassified. The clustering effect of the model is better than that of other models. The MDMAAN model has higher fault diagnosis accuracy under multi-source domains and cross-working conditions, thus proving the stability and superiority of the MDMAAN model.

[0017] 3. Experimental verification shows that compared with the CNN model, CORAL model and MFSAN model, MDMAAN can effectively identify the complex fault types of rolling bearings and improve the efficiency of fault diagnosis by using the multi-kernel maximum mean difference MK-MMD to measure the distance between the source domain and the target domain in a specific feature space. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is the fault diagnosis process of the MDMAAN model;

[0019] Figure 2 It is a rolling bearing cross-operating condition composite fault diagnosis model based on MDMAAN and MK-MMD;

[0020] Figure 3 Comparison of diagnostic accuracy of MDMAAN model with different numbers of convolutional layers;

[0021] Figure 4 Comparison of loss values ​​of different convolutional layers of the MDMAAN model;

[0022] Figure 5 It is the main component of the experimental bench;

[0023] Figure 6 It is a bearing failure component;

[0024] Figure 7The CNN model results for Experiment A are as follows: a) CNN feature visualization and b) CNN confusion matrix.

[0025] Figure 8 Figure 1 shows the CORAL model results for Experiment A. a) CORAL feature visualization and b) CORAL confusion matrix.

[0026] Figure 9 Figure 1 shows the MFSAN model results for Experiment A. a) MFSAN feature visualization and b) MFSAN confusion matrix.

[0027] Figure 10 Figure 1 shows the MDMAAN model results for Experiment A. a) MDMAAN feature visualization and b) MDMAAN confusion matrix.

[0028] Figure 11 CNN model results for Experiment J: a) CNN feature visualization and b) CNN confusion matrix.

[0029] Figure 12 Figure 1 shows the CORAL model results for Experiment J. a) CORAL feature visualization and b) CORAL confusion matrix.

[0030] Figure 13 Figure 1 shows the MFSAN model results for Experiment J. a) MFSAN feature visualization and b) MFSAN confusion matrix.

[0031] Figure 14 Figure 2 shows the MDMAAN model results for Experiment J. a) MDMAAN feature visualization and b) MDMAAN confusion matrix. DETAILED DESCRIPTION

[0032] The technical solution of the present invention is further described below with reference to the accompanying drawings, but is not limited thereto. Any modification or equivalent replacement of the technical solution of the present invention that does not depart from the spirit and scope of the technical solution of the present invention should be included in the scope of protection of the present invention.

[0033] The present invention provides a rolling bearing cross-working condition composite fault diagnosis method based on MDMAAN and MK-MMD. First, an 8-head attention mechanism is proposed, which uses 8 identical attention functions to simultaneously process inputs to obtain different feature information at the same position. The design of this structure allows each attention mechanism to optimize the different feature parts of each position, thereby balancing the deviations that may be caused by the same attention mechanism. The 8-head attention is combined with feature extraction to allocate more weight to valuable information. At the same time, the multi-core maximum mean difference is used to minimize the distribution difference between the source domain and the target domain, so as to better solve the cross-domain fault identification. Finally, a rolling bearing cross-working condition composite fault diagnosis model based on MDMAAN and MK-MMD is proposed. In order to improve the fault diagnosis accuracy, two feature extraction units and two fault diagnosis units are used. The flow chart is as follows: Figure 1 The specific implementation steps are as follows:

[0034] Step 1: Build a dual-feature extraction network structure model:

[0035] Step 1: Divide the data set into n source domain data Ds1, Ds2, ..., Ds according to different working conditions n and target domain data Dt, and divide them into training and test sets; extract features from the input data, and then use 8 attention functions to process the input data simultaneously, assigning different weights to various features so that the model can focus on fault characteristics;

[0036] Step 1 and 2: extract features from the model input data through a common feature extractor and map the extracted original features into a common feature space;

[0037] Step 1 and 3: Use the multi-head attention mechanism to distribute the extracted original features according to the weights, divide each set of source domain data and target domain data into multiple sample pairs, input the sample pairs into a specific feature extractor for feature extraction and mapping to a specific feature space.

[0038] The calculation formula of the multi-head attention mechanism is as follows:

[0039] Multihead(Q,K,V)=Concat(head1,…,head k ,…,head h )W o

[0040] Where: Concat(·) is the vertical concatenation operation of the matrix, the weight matrix head i The expression is:

[0041]

[0042] Where: X is the input, weight matrix is the scaling factor, d k =d model / h.

[0043] Step 2: Build a multi-core maximum mean difference algorithm:

[0044] Step 2.1 Multi-kernel Maximum Mean Difference: Use different kernels to enhance the maximum mean difference, thereby selecting the optimal kernel function and analyzing the domain adaptation loss;

[0045] The calculation formula of the multi-core maximum mean difference is as follows:

[0046]

[0047] Where: φ(·) represents the feature map, P and Q represent two probability distributions, and the maximum mean difference of multiple cores is It is defined as the RKHS distance between P and Q if and only if When , the distributions P and Q are equal.

[0048] The mean difference calculation formula is as follows:

[0049]

[0050] Where: β u Represents weights, imposes constraints on them to ensure the derived multi-core is characteristic, and m represents m different weighted sums.

[0051] Step 2: Analyze the classification loss of the source domain through the source domain label, analyze the classification loss of the prediction results of the target domain, analyze the total loss and update the MDMAAN model through backpropagation.

[0052] Step 3: Construct a rolling bearing cross-operating condition composite fault diagnosis model based on MDMAAN and MK-MMD:

[0053] Step 3: The MDMAAN model consists of two feature extraction units and two fault diagnosis units. The fault diagnosis unit consists of multiple Softmax classifiers. The total loss is composed of the source domain classification loss, MK-MMD loss, and classifier classification loss.

[0054] Step 3.2: Rolling bearing cross-operating condition composite fault diagnosis using MDMAAN and MK-MMD:

[0055] The dataset from step 1 is classified according to different working conditions and input into the trained MDMAAN model. The diagnostic effect is evaluated using the validation set data and the target domain labels to complete the MDMAAN cross-working condition composite fault diagnosis.

[0056] Step 3. Evaluation of cross-operating-condition composite fault diagnosis: When conducting cross-operating-condition composite fault diagnosis of rolling bearings, the diagnostic effectiveness was evaluated using test set data and target domain labels. The MDMAAN model performed well for diagnosing single inner race faults, single outer race faults, and combined inner and outer race faults of rolling bearings, demonstrating its effectiveness in diagnosing cross-operating-condition composite faults of rolling bearings.

[0057] Steps 3 and 4: Evaluation of the diagnostic effects of different models: MFSAN, CNN, and CORAL models were selected as comparison models to obtain feature visualizations and confusion matrix diagrams of the diagnostic model outputs. The comparison showed that the MDMAAN model had a better overall diagnostic effect, with higher recognition accuracy and robustness.

[0058] Example:

[0059] In this embodiment, the rolling bearing cross-operating condition composite fault diagnosis method based on MDMAAN and MK-MMD includes the following steps:

[0060] Step 1: Build a dual feature extraction network structure model;

[0061] Step 2: Construct a multi-core maximum mean difference algorithm;

[0062] Step 3: Construct a rolling bearing cross-working condition composite fault diagnosis model based on MDMAAN and MK-MMD.

[0063] The rolling bearing cross-operating condition composite fault diagnosis model based on MDMAAN and MK-MMD proposed in this embodiment is as follows: Figure 2 shown.

[0064] The Paderborn University publicly available PU bearing fault dataset was used. The Paderborn University test bench consists of a drive motor, torque measurement shaft, rolling bearing test module, flywheel, and load motor. The tested bearings are 6203 bearings. Faults include real-world damage caused by vandalism, electric engraving, EDM, drilling, and fatigue. The sampling frequency is 64 kHz. The PU dataset contains six normal bearings, 12 artificially damaged bearings, and 14 bearings damaged by accelerated life testing. Due to the complexity of the data and the rich variety of bearing fault types in the PU dataset, data from one normal bearing and selected data from accelerated life test bearings K005, KA15, KA16, KB23, KB24, KB27, KI14, and KI21 were selected from the PU dataset for the experiments in this section. Faulty bearings include inner race faults, outer race faults, and combined inner and outer race faults, as shown in Table 1. The speed and load of the rolling bearings used in the PU data collection experiments are shown in Table 2.

[0065] Table 1 Description of PU partially faulty bearings

[0066]

[0067] Table 2 Operating parameters of PU data acquisition experiment

[0068]

[0069] As can be seen from Table 2, the PU dataset has four different speeds, load torques, and radial forces. Based on these four different working conditions, the experiment can set up nine cross-working condition multi-source domain migration fault diagnosis tasks, as shown in Table 3. The data is divided into 5000 samples, each with a length of 1024.

[0070] Table 3 Multi-source domain fault diagnosis experiment settings

[0071]

[0072]

[0073] 80% of each dataset was randomly selected as the training set, and 20% as the test set. A set of bearing fault data was randomly selected as the test set. The number of iterations was set to 30, the initial learning rate was 0.01, and SGD was used as the optimization algorithm with a momentum of 0.9. Each model was trained for 30 cycles for fault diagnosis. A learning rate decay strategy was also used to adjust the learning rate, and the weight decay value was initialized to 0.0005. The loss function was cross-entropy. In fault diagnosis, overall accuracy was used as the evaluation metric, defined as the ratio of the number of correctly classified samples to the total number of samples. The initial model parameter settings are shown in Table 4. The effects of different convolutional layers, different training batches, and different numbers of source domains on the fault diagnosis of the MDMAAN model were studied.

[0074] Table 4 Experimental settings of MDMAAN model parameters

[0075]

[0076]

[0077] By setting the experimental conditions in Table 3, we compare the MDMAAN model with different convolution layers. This paper compares the models from two to seven layers of convolution, and selects the most appropriate number of layers based on the model's diagnostic accuracy and loss curve. Figure 3 As shown, the loss value comparison is Figure 4 shown.

[0078] By comparing different numbers of convolutional layers, it is found that with the increase in the number of layers, the diagnostic accuracy of the MDMAAN model increases the fastest and reaches the highest level when there are five convolutional layers, the loss value converges the fastest, and the iteration speed is the fastest. This paper selects five convolutional layers for multi-source domain cross-working condition fault diagnosis experiments.

[0079] The test bench is made of our own bearing failure test bench BPS, which consists of a digital force display, a motor speed controller, a motor, a radial hydraulic loading system, an axial hydraulic loading system, a main shaft, two support bearings 6210 and 18720, a 1-inch bearing to be tested, a force arm beam adjustment device, and a hydraulic device for disassembling the bearing. Figure 5 As shown. The experiment collects data through a three-axis acceleration sensor with a sampling frequency of 25.6KHz. Experiments are conducted on single faults and composite faults under different degrees of bearing damage. In the experiment, the radial acceleration sensor is used to obtain the normal state information of a rolling bearing and the vibration acceleration information of 12 bearing fault states. The 12 bearing fault states include 9 single-point faults and 3 composite faults. The composite fault is a combination of the three faults of the inner ring, outer ring and rolling element. Figure 6 Taking the speed of 900 r / min as an example, the basic parameters of the bearing and the fault characteristic frequency are shown in Table 5, the experimental parameters of different fault states are shown in Table 6, and the operating condition parameters collected in the experiment are shown in Table 7.

[0080] Table 5 Main parameters and fault characteristic frequencies of ER-16K bearings

[0081]

[0082]

[0083] Table 6 Experimental parameter settings for different fault states

[0084]

[0085] Table 7 Experimental working condition parameter settings

[0086]

[0087] (1) Comparative experimental verification of rolling bearing cross-operating condition composite fault diagnosis under different diagnostic models

[0088] Using PU experimental data as an example, the proposed method is validated. First, the diagnostic accuracy of the MDMAAN model for cross-operating fault diagnosis demonstrates the superiority of this method. The diagnostic results are shown in Table 5. Second, a comparison is made between multiple deep transfer learning fault diagnosis models. Feature visualizations and confusion matrix plots of the label classification of the MDMAAN rolling bearing deep transfer learning fault diagnosis model further demonstrate the proposed model's strong fault diagnosis capabilities.

[0089] Table 5 Diagnostic results of the MDMAAN model for different working conditions

[0090]

[0091] Table 5 shows that the MDMAAN model achieves an average recognition accuracy of 98.87% for rolling bearing cross-operating-condition combined fault diagnosis, demonstrating its effectiveness in cross-operating-condition rolling bearing fault diagnosis. Comparisons across different operating conditions show that the MDMAAN model performs slightly better for cross-load fault diagnosis than for cross-speed fault diagnosis. This suggests that speed has a greater influence than load on cross-operating-condition rolling bearing fault diagnosis.

[0092] A group of experiments A in Table 5 was randomly selected. The source domain 1 working condition of the experiment was a speed of 1500 r / min, a load torque of 0.7 Nm, and a radial force of 1000 N. The source domain 2 working condition was a speed of 900 r / min, a load torque of 0.7 Nm, and a radial force of 1000 N. The target domain working condition was a speed of 1500 r / min, a load torque of 0.1 Nm, and a radial force of 1000 N. A comparison of deep transfer learning fault diagnosis was performed. The compared models were MFSAN, CNN, and CORAL models. The diagnosis results of different models are shown in Table 6. The feature visualization diagram and confusion matrix diagram output by the diagnosis model are shown in Figure 6. Figure 7 、 Figure 8 、 Figure 9 and Figure 10 shown.

[0093] Table 6 Accuracy of different diagnostic models under Experiment A

[0094]

[0095] Through the comparison of the accuracy of the MDMAAN, MFSAN, CNN and CORAL models in Table 6, it is found that the MDMAAN model is significantly better than the other three models in the diagnosis of rolling bearing faults under cross-operating conditions. The diagnostic accuracy of the MDMAAN model is 4.75% higher than that of the MFSAN model, the diagnostic accuracy of the MDMAAN model is 12.87% higher than that of the CNN model, and the diagnostic accuracy of the MDMAAN model is 9.87% higher than that of the CORAL model. Figure 7 a) and Figure 7 b) It is not difficult to find that the CNN model has poor diagnostic effect on K005, KB23 and KB24, especially 30% of KB24 fault categories are diagnosed as KB23, and 24% of KB24 fault categories are diagnosed as KI14; Figure 8 a) and Figure 8 b) It can be seen that the CORAL model also misdiagnoses the KB24 fault category, and diagnoses 37% of the fault categories as KB23. Figure 9a) and Figure 9 b) It can be seen that the MFSAN model has a good diagnosis effect on each fault category. A small number of fault labels are classified into other fault categories, but the diagnosis effect is obviously not as good as that of the MDMAAN model. Figure 10 a) and Figure 10 b) As can be seen, the MDMAAN model performs well in diagnosing rolling bearing faults across all categories. The results show that the target domain is clustered into eight clusters, corresponding to the eight fault states within the target domain, including single inner race faults, single outer race faults, and combined inner and outer race faults. Only a very small number of combined inner and outer race faults fall into other fault categories. The MDMAAN model also outperforms the other three models in single fault diagnosis. The MDMAAN model achieves the highest diagnostic accuracy, especially for combined rolling bearing faults. It can leverage laboratory data to perform cross-condition fault diagnosis in real-world scenarios. The MDMAAN model also demonstrates excellent comprehensive diagnostic performance, with high identification accuracy and robustness.

[0096] (2) Experimental verification of rolling bearing cross-operating condition composite fault diagnosis based on MDMAAN

[0097] Based on the 13 bearing types in Table 8 and the nine operating conditions in Table 7, the following 10 multi-source domain cross-operating condition fault diagnosis experimental tasks can be set. To reduce errors, the same cross-operating condition experimental tasks are divided into two different sets of data for fault diagnosis. Experiments A and B have the same experimental tasks, Experiments C and D have the same experimental tasks, Experiments E and F have the same experimental tasks, Experiments G and H have the same experimental tasks, and Experiments I and J have the same experimental tasks. The specific source domain and target domain operating conditions are shown in Table 8.

[0098] Table 7 Experimental working condition parameter settings

[0099]

[0100] Table 8 MDMAAN model cross-operating condition fault diagnosis experiment task settings

[0101]

[0102] The fault diagnosis results of the cross-working condition fault diagnosis experimental task of deep transfer learning of the MDMAAN model in Table 8 are shown in Table 9.

[0103] Table 9 Results of the MDMAAN model cross-operating condition fault diagnosis experimental task

[0104]

[0105] For Experiment J in Table 9, the CNN, CORAL, MFASN, and MDMAAN models were compared. The operating conditions of source domain 1 were 900 rpm and 200 kg radial load, the operating conditions of source domain 2 were 1500 rpm and 0 kg radial load, and the operating conditions of the target domain were 2400 rpm and 100 kg radial load. The diagnosis results of different models are shown in Table 10. The feature visualization and confusion matrix output by the diagnosis model are shown in Figure 10. Figure 11 、 Figure 12 、 Figure 13 and Figure 14 shown.

[0106] Table 10 Accuracy of different diagnostic models under Experiment A

[0107]

[0108] Through the comparison of the accuracy of the MDMAAN, MFSAN, CNN and CORAL models in Table 10, it is found that the MDMAAN model is significantly better than the other three models in rolling bearing fault diagnosis under cross-operating conditions. The diagnostic accuracy of the MDMAAN model is higher than that of the MFSAN model by 7.23%, the CNN model by 12.49% and the CORAL model by 13.05%, respectively, indicating the effectiveness and excellence of the model in multi-source domain cross-operating fault diagnosis. Figure 11 a) and Figure 11 b) It can be found that 54% of the B10 rolling element moderate faults and 50% of the O03 outer ring moderate faults were misclassified in the diagnosis results of the CNN model. Figure 12 a) and Figure 12 b) It can be found that 50% of the B15 rolling element weight faults are misclassified in the diagnosis results of the CORAL model, and a small number of other faults are misclassified. Figure 13 a) and Figure 13 b) It can be seen that 49% of the B15 rolling weight faults were misclassified in the diagnosis results of the MFSAN model, and a small number of other faults were misclassified. Figure 14 a) and Figure 14 b) It can be seen that only a very small number of fault categories are misclassified in the diagnosis results of the MDMAAN model. The results show that the target domain is clustered into thirteen clusters, corresponding to the thirteen fault states of the target domain, and the clustering effect of the model is better than that of other models, indicating that the MDMAAN model has higher fault diagnosis accuracy under multi-source domain and cross-working conditions, thus proving the stability and superiority of the MDMAAN model.

Claims

1. A rolling bearing cross-operating condition composite fault diagnosis method based on MDMAAN and MK-MMD, characterized by The method comprises the following steps: Step 1: Divide the dataset into multiple source domain data and target domain data according to different working conditions, and divide it into training set and test set; extract common features from the input data and map the extracted original features to the common feature space; After being processed by the attention mechanism, they are spliced ​​together and linearly transformed to obtain the output of the multi-head attention mechanism. The extracted features are weighted, and specific features are extracted from the assigned features and mapped to a specific feature space. Step 2: Minimize the distribution difference between the source and target domains through multi-kernel maximum mean difference (MK-MMD) and analyze the domain adaptation loss; analyze the classification loss of the source domain through the source domain label, analyze the classification loss of the prediction results of the target domain, analyze the total loss and update the MDMAAN model through backpropagation; Step 3: The MDMAAN model consists of two feature extraction units and two fault diagnosis units. The fault diagnosis unit consists of multiple Softmax classifiers, and the total loss is composed of the source domain classification loss, the MK-MMD loss, and the classifier classification loss. The dataset is classified according to different working conditions and input into the trained MDMAAN model. The diagnostic effect is evaluated using the validation set data and the target domain labels to complete the MDMAAN cross-working condition composite fault diagnosis.

2. The rolling bearing cross-operating condition composite fault diagnosis method based on MDMAAN and MK-MMD according to claim 1 is characterized in that The specific steps of step one are as follows: Step 1: Divide the data set into n source domain data Ds1, Ds2, ..., Ds according to different working conditions n and target domain data Dt, and divide them into training and test sets; extract features from the input data, and then use 8 attention functions to process the input data simultaneously, assigning different weights to various features so that the model can focus on fault characteristics; Step 1 and 2: extract features from the model input data through a common feature extractor and map the extracted original features into a common feature space; Step 1 and 3: Use the multi-head attention mechanism to distribute the extracted original features according to the weights, divide each set of source domain data and target domain data into multiple sample pairs, input the sample pairs into a specific feature extractor for feature extraction and mapping to a specific feature space.

3. The rolling bearing cross-operating condition composite fault diagnosis method based on MDMAAN domain MK-MMD according to claim 2 is characterized in that The calculation formula of the multi-head attention mechanism is as follows: Multihead(Q,K,V)=Concat(head1,…,head k ,…,head h )W o Where: Concat(·) is the vertical concatenation operation of the matrix, the weight matrix head i The expression is: Where: X is the input, weight matrix is the scaling factor, d k =d model / h.

4. The rolling bearing cross-operating condition composite fault diagnosis method based on MDMAAN and MK-MMD according to claim 1 is characterized in that The specific steps of step 2 are as follows: Step 2.1 Multi-kernel Maximum Mean Difference: Use different kernels to enhance the maximum mean difference, thereby selecting the optimal kernel function and analyzing the domain adaptation loss; Step 2: Analyze the classification loss of the source domain through the source domain label, analyze the classification loss of the prediction results of the target domain, analyze the total loss and update the MDMAAN model through backpropagation.

5. The rolling bearing cross-operating condition composite fault diagnosis method based on MDMAAN and MK-MMD according to claim 4 is characterized in that The calculation formula of the multi-core maximum mean difference is as follows: Where: φ(·) represents the feature map, P and Q represent two probability distributions, and the maximum mean difference of multiple cores is It is defined as the RKHS distance between P and Q if and only if When , the distributions P and Q are equal. The mean difference calculation formula is as follows: Where: β u Represents weights, imposes constraints on them to ensure the derived multi-core is characteristic, and m represents m different weighted sums.

6. The rolling bearing cross-operating condition composite fault diagnosis method of MDMAAN and MK-MMD according to claim 1 is characterized in that The specific steps of step three are as follows: Step 3: The MDMAAN model consists of two feature extraction units and two fault diagnosis units. The fault diagnosis unit consists of multiple Softmax classifiers. The total loss is composed of the source domain classification loss, MK-MMD loss, and classifier classification loss. Step 3.2: Rolling bearing cross-operating condition composite fault diagnosis using MDMAAN and MK-MMD: The data set in step 1 is classified according to different working conditions and input into the trained MDMAAN model to complete the MDMAAN cross-working condition composite fault diagnosis; Step 3. Evaluation of cross-operating-condition composite fault diagnosis: When conducting cross-operating-condition composite fault diagnosis of rolling bearings, the diagnostic effectiveness was evaluated using test set data and target domain labels. The MDMAAN model performed well for diagnosing single inner race faults, single outer race faults, and combined inner and outer race faults of rolling bearings, demonstrating its effectiveness in diagnosing cross-operating-condition composite faults of rolling bearings. Steps 3 and 4: Evaluation of the diagnostic effects of different models: MFSAN, CNN, and CORAL models were selected as comparison models to obtain feature visualizations and confusion matrix diagrams of the diagnostic model outputs. The comparison showed that the MDMAAN model had a better overall diagnostic effect, with higher recognition accuracy and robustness.

Images