Fault diagnosis method and system based on domain separation anti-fake label network

Abstract

The invention discloses a fault diagnosis method and system based on a domain separation anti-fake label network, and relates to the technical field of mechanical equipment fault diagnosis, and the method comprises the steps: obtaining a vibration source domain signal and a target domain signal of mechanical equipment, carrying out the feature extraction, and carrying out the iteration updating of a fake label; a DSAPN fault diagnosis model is constructed, a joint loss function of fault diagnosis is established, and the model is updated and iterated; and inputting a target domain sample set, fixing target domain sample classification points, and outputting a target domain diagnosis result. According to the fault diagnosis method based on the domain separation anti-fake label network, the advantages of the anti-fake label network and the neighborhood separation network are combined, the domain separation network is improved, a fake label mechanism is introduced, the problem that neighborhood features are difficult to separate is effectively solved, and the generalization ability and robustness of the model are remarkably improved. The method has better effects in the aspects of generalization ability, data demand and stability.

Description

Technical Field

[0001] The present invention relates to the technical field of mechanical equipment fault diagnosis, and in particular to a fault diagnosis method and system based on a domain separation-based pseudo-labeling network. Background Art

[0002] In recent years, deep learning technology has made significant progress in the field of bearing fault diagnosis, especially in automatic feature extraction and prediction accuracy improvement. However, when dealing with cross-operating conditions and cross-equipment diagnosis challenges, existing methods still have problems such as large data requirements and insufficient stability, which affects their promotion in practical applications.

[0003] In view of the above problems, the present invention proposes a domain separation adversarial pseudo-label network (DSAPN), which combines the advantages of domain adversarial learning network and domain separation learning network, and improves the domain separation learning network. Specifically, the present invention designs a network structure including private feature extractors and shared feature extractors for source and target domains, and efficiently extracts data features by combining convolutional neural networks with deep neural networks. At the same time, a gradient reversal layer is introduced to reduce the distribution difference between the source domain and the target domain. In addition, the classification accuracy of the target domain is further improved through the pseudo-label generation and update mechanism, thereby effectively solving the problems of large data demand and insufficient stability in the prior art. Summary of the invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is that the existing bearing fault diagnosis method based on deep learning has insufficient stability and the problem of large data demand when crossing working conditions and equipment.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: a fault diagnosis method based on domain separation against pseudo-label networks, comprising acquiring vibration source domain signals and target domain signals of mechanical equipment for feature extraction, and iteratively updating pseudo-labels; constructing a DSAPN fault diagnosis model, establishing a joint loss function for fault diagnosis, and iterating the model; inputting a target domain sample set, fixing the target domain sample classification points, and outputting the target domain diagnosis results.

[0007] As a preferred solution of the fault diagnosis method based on domain separation against pseudo-label network described in the present invention, wherein: the acquisition of vibration source domain signals and target domain signals of mechanical equipment for feature extraction includes performing feature extraction through a feature extractor composed of CNN+DNN, first extracting surface features through the CNN network, and then extracting deep features through the DNN network.

[0008] As a preferred solution of the fault diagnosis method based on domain separation against pseudo-label network described in the present invention, the iterative updating of pseudo-labels includes initializing the model, training the basic model through source domain data and target domain data, and then predicting the target domain unlabeled data through the trained model to generate pseudo-labels, selecting pseudo-labels whose confidence meets the preset standard for the next retraining of the model, and after each training, using the updated model to generate new pseudo-labels, and training, iteratively optimizing the model performance.

[0009] As a preferred solution of the fault diagnosis method based on domain separation against pseudo-label network described in the present invention, wherein: the joint loss function for establishing fault diagnosis includes the classification loss L of the category classifier Tc task , the classification loss L of the domain adaptation classifier Dc adapt , the classification loss L of the domain separator Ds sep and MMD loss.

[0010] As a preferred solution of the fault diagnosis method based on domain separation against pseudo-label network described in the present invention, the updating and iteration of the model includes using a back propagation algorithm to optimize the model through feature alignment between domains, and optimizing and iterating through a gradient descent strategy. When the iteration conditions are met, the training is completed to determine the DSAPN fault diagnosis model.

[0011] As a preferred solution of the fault diagnosis method based on domain separation against pseudo-label network described in the present invention, wherein: the fixed target domain sample classification points include fixed category classifiers, domain adaptation classifiers and domain separation classifiers.

[0012] As a preferred solution of the fault diagnosis method based on domain separation against pseudo-label network described in the present invention, wherein: the output of target domain diagnosis result includes inputting the target domain sample into the DSAPN fault diagnosis model that has completed the fixed target domain sample classification point to output the target domain diagnosis result.

[0013] Another object of the present invention is to provide a fault diagnosis system based on domain separation against pseudo-label networks, which can reduce the model's demand for fault data by introducing a pseudo-label mechanism, thereby solving the problem of large data requirements in current deep learning-based bearing fault diagnosis methods.

[0014] As a preferred solution of the fault diagnosis system based on domain separation against pseudo-label network described in the present invention, it includes: a data acquisition module, a model construction module, and a prediction module; the data acquisition module is used to obtain the vibration source domain signal and the target domain signal of the mechanical equipment for feature extraction, and iteratively update the pseudo-label; the model construction module is used to construct a DSAPN fault diagnosis model, establish a joint loss function for fault diagnosis, and iteratively update the model; the prediction module is used to input a target domain sample set, fix the target domain sample classification point, and output the target domain diagnosis result.

[0015] A computer device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement a fault diagnosis method based on domain separation to counteract a pseudo-label network.

[0016] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a fault diagnosis method for a domain separation-based countermeasure pseudo-label network.

[0017] Beneficial effects of the invention: The fault diagnosis method based on domain separation adversarial pseudo-label network provided by the invention combines the advantages of adversarial network and neighborhood separation network, improves the domain separation network, introduces pseudo-label mechanism, effectively solves the problem of difficulty in separating neighborhood features, and significantly improves the generalization ability and robustness of the model. The invention achieves better results in generalization ability, data demand and stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0019] Figure 1 An overall flow chart of a fault diagnosis method based on domain separation against pseudo-label networks provided in the first embodiment of the present invention.

[0020] Figure 2 The overall structure diagram of a DSAPN network of a fault diagnosis method based on domain separation to combat pseudo-label networks provided in the first embodiment of the present invention.

[0021] Figure 3 A schematic diagram of a feature extractor of a fault diagnosis method based on domain separation against pseudo-label network provided in the second embodiment of the present invention.

[0022] Figure 4A pseudo-label iterative update diagram of a fault diagnosis method based on domain separation against pseudo-label networks provided by the second embodiment of the present invention. DETAILED DESCRIPTION

[0023] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0024] Example 1, reference Figure 1-Figure 2 , is an embodiment of the present invention, and provides a fault diagnosis method for a pseudo-label network based on domain separation, comprising:

[0025] Obtain the vibration source domain signal and target domain signal of the mechanical equipment for feature extraction, and iteratively update the pseudo-label. Build the DSAPN fault diagnosis model, establish the joint loss function of fault diagnosis, and iterate the model. Input the target domain sample set, fix the target domain sample classification point, and output the target domain diagnosis result.

[0026] Reference Figure 1 The following is the algorithm framework diagram of the DSAPN model. Model learning is divided into three steps: initialization, model training, and model prediction.

[0027] Initialization: The initialization strategy of network model training is very important for building a deep architecture. In order to ensure the quality of the learned network parameters, the classification loss L of the category classifier is calculated using shallow feature data. task To initialize various parameters.

[0028] DSAPN model training: The DSAPN model input is the source domain samples with labels and the target domain samples without labels. The purpose of model training is to minimize the loss function L. The back propagation algorithm is used for updating and the gradient descent is used for optimization, which is expressed as:

[0029]

[0030] in, is the gradient of the total loss with respect to the parameter θ, i+1 represents the parameter of the next iteration, λ is the adversarial coefficient, which is used to balance the classification loss and domain loss. α and β are the weights of MMD and separation loss. Through the above steps, the model is gradually optimized in each iteration to reduce the total loss, thereby improving the classification performance of the source domain and the target domain.

[0031] MMD Loss L MMD , parameter is θMMD In order to further reduce the distribution difference between the source domain and the target domain in space and improve the discriminative ability of the model, the MMD loss is introduced into the shared feature extractor to reduce the distribution difference while further expanding the generated shared features, which is expressed as:

[0032]

[0033] in, is the feature mapping function, is the feature of the mth source domain sample, is the feature of the nth target domain sample.

[0034] DSAPN model prediction: The optimized category classifier, domain adaptation classifier, and domain separation classifier constitute a stable performance fault diagnosis adaptive model, which inputs the target domain samples and outputs the fault diagnosis prediction results.

[0035] The overall structure of the proposed model is as follows Figure 2 As shown in the figure, it contains several key components: shared feature extractor, source domain private feature extractor, target domain private feature extractor, category classifier, domain adaptation classifier, domain separation classifier, and pseudo label generation iterator. These components together form a complete composite network. By training the network and optimizing the parameters to minimize the label prediction loss, more efficient and accurate bearing fault diagnosis can be achieved.

[0036] Furthermore, the category classifier Tc, parameterized by θ task The category classifier learns and classifies the shared invariant features extracted from the source domain and the target domain by the shared feature extractor. It classifies the source domain data samples during training and effectively classifies the unlabeled target domain data samples during testing. The classification loss L task , used to characterize the classifier training optimization process.

[0037]

[0038] Where N is the number of samples in the batch, C is the number of categories, and y ic is the true label of sample i, is the probability that sample i is predicted to be sample C.

[0039] It should be noted that the domain adaptation classifier Dc has parameters θ adapt The domain adaptation classifier learns the domain features extracted from the source domain and the target domain by the shared feature extractor, and outputs effective domain classification after training and optimization. Classification loss L adapt , used to characterize the training optimization process of the domain adaptation classifier.

[0040]

[0041] Among them, N s is the number of source domain samples, N t is the number of samples in the target domain, is the probability predicted by the domain classifier for source domain sample i, is the probability predicted by the domain classifier for the target domain sample i.

[0042] It should also be noted that the domain separator Ds has a parameter θ sep The domain separator learns the features extracted from the source domain private feature extractor, the target domain private feature extractor, and the shared feature extractor, and outputs an effective domain separation classification after training and optimization. The present invention defines the classification loss L sep , to characterize the domain separator training optimization process.

[0043]

[0044] Among them, 3 is the number of domain categories, corresponding to 3 domains, y ik is the true domain label of sample i, is the probability that sample i is predicted to be one of the three domains.

[0045] Example 2, reference Figure 3-Figure 4 , is an embodiment of the present invention, and provides a fault diagnosis method for a pseudo-label network based on domain separation, comprising:

[0046] S1: Obtain the vibration source domain signal and target domain signal of the mechanical equipment for feature extraction, and iteratively update the pseudo label.

[0047] Furthermore, obtaining the source domain signal and target domain signal of the mechanical equipment vibration for feature extraction includes extracting features through a feature extractor composed of CNN+DNN, first extracting surface features through the CNN network, and then extracting deep features through the DNN network. Such a combination not only makes feature separation easier, but also increases the depth of the network, which can solve the problem of expanding the adaptability of the network.

[0048] It should be noted that for the case where the source domain data and the target domain data are quite different, the present invention introduces pseudo labels, combined with iterative update strategies, to gradually optimize the model performance and improve the accuracy of the target domain. Iterative updating of pseudo labels includes initializing the model, training the basic model with source domain data and target domain data, and then predicting the target domain unlabeled data with the trained model to generate pseudo labels, selecting pseudo labels with confidence levels that meet preset standards for the next retraining of the model, and after each training, using the updated model to generate new pseudo labels, and training, iteratively optimizing model performance.

[0049] S2: Build a DSAPN fault diagnosis model, establish a joint loss function for fault diagnosis, and update and iterate the model.

[0050] Furthermore, the joint loss function of fault diagnosis is established, including the classification loss L of the category classifier Tc task , the classification loss L of the domain adaptation classifier Dc adapt , the classification loss L of the domain separator Ds sep and MMD loss.

[0051] The total loss L is expressed as:

[0052] L=L task -λL adapt +αL MMD +βL sep

[0053] It should be noted that updating and iterating the model includes using the back propagation algorithm to optimize the model through feature alignment between domains, and optimizing and iterating through the gradient descent strategy. When the iteration conditions are met, the training is completed and the DSAPN fault diagnosis model is determined.

[0054] S3: Input the target domain sample set, fix the target domain sample classification point, and output the target domain diagnosis result.

[0055] Furthermore, the fixed target domain sample classification points include fixed category classifiers, domain adaptation classifiers, and domain separation classifiers.

[0056] It should be noted that outputting the target domain diagnosis result includes inputting the target domain sample into the DSAPN fault diagnosis model that has completed the fixed target domain sample classification point to output the target domain diagnosis result.

[0057] Example 3, an embodiment of the present invention, provides a fault diagnosis method based on domain separation to combat pseudo-label networks. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0058] First, according to the bearing fault diagnosis method based on domain separation against pseudo-label network proposed in this example, experiments were conducted on the Case Western Reserve University bearing dataset and the Paderborn bearing dataset. Cross-operating condition experiments were conducted on the Case Western Reserve University bearing dataset, and cross-device experiments were conducted using the Case Western Reserve University bearing dataset as the source domain and the Paderborn bearing dataset as the target domain. For cross-operating condition experiments, the method of the present invention: the bearing fault diagnosis method based on the domain separation against pseudo-label network is compared with the experimental results of support vector machines (Support Victor Machines, SVM), deep neural networks (Deep Neural Networks, DNN), deep neural networks with wide convolutions (Deep Convolutional Neural Networks with Wide First-layer Kernels, WDCNN), and one-dimensional convolutional neural networks (1-Dimensional Convolutional Neural Networks, 1DCNN). For cross-device experiments, the method of the present invention: the bearing fault diagnosis method based on the domain separation against pseudo-label network is compared with SVM, WDCNN, deep adversarial neural network (Domain Adaptive Neural Network, DANN), and domain separation reconstruction adversarial network (Domain Separation Reconstruction Adversarial Network, DSRAN). From all the experimental results, it can be seen that the present invention has a significant improvement in accuracy. It should be noted that the above-mentioned fault diagnosis method is only illustrated by the above-mentioned benchmark data set for experimental examples. In practical applications, the above-mentioned method can be applied to different scenarios for experimental analysis as needed.

[0059] For the single data set cross-operating condition experiment, the present invention does not introduce a pseudo-label iteration mechanism, and uses the original network for the experiment. The experiment selects the Western Reserve University bearing data set to verify the stability of the model and the prediction accuracy under different operating conditions.

[0060] The CWRU bearing data set is composed of vibration signals of bearing faults collected by acceleration sensors at the drive end and fan end of the motor. Among them, the bearing used at the drive end is SKF6205, the sampling frequency is 12kHz and 48kHz, and the bearing used at the fan end is SKF6203, and the sampling frequency is 12kHz. The collected fault signals can be divided into inner ring fault (IF), outer ring fault (OF), rolling element fault (B) and normal (N) according to the bearing status. For different fault states, they can be divided into slight, medium and severe faults, and the corresponding fault diameters are 0.007 inches, 0.014 inches and 0.021 inches respectively. The load condition and speed of the motor are also different. The motor load includes 0 horsepower, 1 horsepower, 2 horsepower and 3 horsepower, and the motor speed is 1797rpm, 1772rpm, 1750rpm and 1730rpm. So 10 health states are obtained. As shown in Table 1.

[0061] The experiment of the present invention selects different fault positions and different health status vibration signals of the rolling bearing drive end with a sampling frequency of 12kHz for experimental research, and defines a data set containing different loads and different speeds. The detailed description of the data set is shown in Table 1. In Table 1, the experiment selects the drive end bearing fault vibration signals of different fault positions and different health states of 12kHz as the data set. Data set A consists of 2000 samples of 10 bearing health states at 1hp and 1772rpm, data set B consists of 2000 samples of 10 bearing health states at 2hp and 1750rpm, and data set C consists of 2000 samples of 10 bearing health states at 3hp and 1730rpm. Among them, there are 200 samples of health status, each sample contains 1200 data points. Considering that each original data contains about 12,000 data points, when 1200 data points are taken as one sample, only about 100 samples can be generated, which does not reach the 200 samples required for the experiment of the present invention, so the sampling overlap sampling technology is used to expand the number of data samples of the present invention, and the sampling overlap rate is 50%. After the above processing operations, a sample that meets the experimental requirements of the present invention is formed.

[0062] Table 1 List of cross-operating condition experimental data

[0063]

[0064] In order to test the domain adaptation performance of the present invention, A→B indicates that dataset A is the source domain, dataset B is the target domain, and so on. Therefore, for datasets A, B, and C, there are 6 domain adaptation problems. The algorithm proposed in the present invention is compared with SVM, DNN, WDCNN, and 1DCNN. The experimental results of the final accuracy are shown in Table 2.

[0065] Table 2 Comparison of cross-operating condition diagnosis results and methods

[0066]

[0067]

[0068] From Table 2, it can be seen that the average accuracy of the algorithm proposed in the present invention is 99.9%, which is significantly better than other methods. In the five domain migration scenarios of A→B, B→A, A→C, B→C and C→B, the fault diagnosis accuracy of the present invention has achieved the optimal domain adaptation effect, reaching 100%. In the domain migration scenario of C→A, although it did not reach 100%, it is better than other methods.

[0069] For cross-device diagnosis experiments, the present invention introduces a pseudo-label iteration mechanism and uses the updated network to conduct experiments. The experiment selects the CWRU bearing dataset as the source domain data and the Paderborn (PDB) bearing dataset as the target domain data to verify the prediction accuracy of the model for different datasets.

[0070] The PDB bearing dataset experiments on 32 different bearings, of which 12 are artificially damaged bearings, 14 are real loss bearings, and the other 6 are healthy bearings. 20 original vibration signals are collected for each bearing, with a sampling frequency of 64kHz, each signal lasting 4s, and about 256,000 points are recorded. Its fault signals can be divided into inner ring fault, outer ring fault and normal according to the bearing fault state. The motor speed is 1500rpm and 900rpm, the load torque is 0.7Nm and 0.1Nm, and the radial force is 1kN and 0.4kN.

[0071] The source domain data D1 of this experiment consists of 6000 samples of 3 bearing states at the fan end of Western Reserve University at 1hp and 1772rpm, and D2 consists of 6000 samples of 3 bearing states at the fan end of Western Reserve University at 2hp and 1750rpm. Considering the large difference between the PDB bearing dataset and the Western Reserve University bearing dataset, the present invention selects the dataset produced by artificial damage in the PDB bearing dataset, which is similar to the Western Reserve University bearing data. Because the artificial damage data also includes electrospark machining and artificial electro-engraving, the present invention produces target domain datasets F1 and F2. F1 consists of 7800 samples of 3 bearing states of the PDB bearing dataset under the conditions of electrospark machining, motor speed of 1500rpm, load torque of 0.7Nm, and radial force of 0.4kN. F2 consists of 7800 samples of three bearing conditions of the PDB bearing dataset under artificial electric etching, motor speed of 1500rpm, load torque of 0.7Nm, and radial force of 0.4kN, as shown in Table 3.

[0072] Table 3 Cross-device experimental data list

[0073]

[0074]

[0075] In order to test the domain adaptation performance of the present invention across devices, the present invention conducted 4 groups of cross-operating condition experiments, namely D1→F1, D1→F2, D2→F1 and D2→F2. Taking D1→F1 as an example, D1 is the source domain data with labels, F1 is the target domain data without labels, and so on. The algorithm proposed in the present invention is compared with SVM, DANN, WDCNN, and DSRAN. The final experimental results of the accuracy are shown in Table 4.

[0076] Table 4 Comparison of cross-device diagnostic results and methods

[0077] method D1→F1 D1→F2 D2→F1 D2→F2 Average SVM 34.0％ 35.9％ 40.2％ 39.2％ 37.3％ DANN 83.2％ 80.3％ 73.9％ 76.2％ 52.7％ WDCNN 51.0％ 47.2％ 56.7％ 55.9％ 78.4％ DSRAN 92.4％ 92.7％ 87.7％ 85.4％ 89.5％ DSAPN 95.7％ 96.5％ 95.5％ 94.7％ 95.6％

[0078] From Table 4, it can be seen that the average accuracy of the algorithm proposed in the present invention is 95.6%, which is significantly better than other methods. In the domain migration scenarios of D1→F1, D1→F2, and D2→F1, the fault diagnosis accuracy of the present invention has achieved a good domain adaptation effect, reaching 96%. In the domain migration scenario of D2→F2, although it did not reach the accuracy of other domain migrations, it is still slightly better than other methods.

[0079] Embodiment 4, an embodiment of the present invention provides a fault diagnosis system based on domain separation against pseudo-label network, including a data acquisition module, a model building module, and a prediction module.

[0080] The data acquisition module is used to obtain the vibration source domain signal and target domain signal of the mechanical equipment for feature extraction and iterative update of pseudo labels. The model building module is used to build the DSAPN fault diagnosis model, establish the joint loss function of fault diagnosis, and update the model iteratively. The prediction module is used to input the target domain sample set, fix the target domain sample classification point, and output the target domain diagnosis result.

[0081] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium Including: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), Random Access Various media that can store program codes, such as RAM (Random Access Memory), disks or optical disks.

[0082] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0083] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0084] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc. It should be noted that the above embodiments are only used to illustrate the technical solution of the present invention and are not limited. Although the present invention is described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.

[0085] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A fault diagnosis method based on domain separation against pseudo-label network, characterized in that: include: Obtain the vibration source domain signal and target domain signal of the mechanical equipment for feature extraction, and iteratively update the pseudo label; Construct the DSAPN fault diagnosis model, establish the joint loss function of fault diagnosis, and update and iterate the model; Input the target domain sample set, fix the target domain sample classification point, and output the target domain diagnosis result.

2. The fault diagnosis method based on domain separation against pseudo-label network according to claim 1, characterized in that: The method of obtaining the vibration source domain signal and the target domain signal of the mechanical equipment for feature extraction includes performing feature extraction through a feature extractor composed of CNN+DNN, first extracting surface features through the CNN network, and then extracting deep features through the DNN network.

3. The fault diagnosis method based on domain separation against pseudo-label network according to claim 2, characterized in that: The iterative updating of pseudo labels includes initializing the model, training the basic model through source domain data and target domain data, predicting the target domain unlabeled data through the trained model to generate pseudo labels, selecting pseudo labels whose confidence meets the preset standard for the next retraining of the model, and after each training, using the updated model to generate new pseudo labels, and training to iteratively optimize model performance.

4. The fault diagnosis method based on domain separation against pseudo-label network according to claim 3, characterized in that: The joint loss function for establishing fault diagnosis includes the classification loss L of the category classifier Tc task , the classification loss L of the domain adaptation classifier Dc adapt , the classification loss L of the domain separator Ds sep and MMD loss.

5. The fault diagnosis method based on domain separation against pseudo-label network according to claim 4, characterized in that: The updating and iterating of the model includes using a back propagation algorithm to optimize the model through feature alignment between domains, and optimizing and iterating through a gradient descent strategy. When the iteration conditions are met, the training is completed to determine the DSAPN fault diagnosis model.

6. The fault diagnosis method based on domain separation against pseudo-label network according to claim 5, characterized in that: The fixed target domain sample classification points include a fixed category classifier, a domain adaptation classifier and a domain separation classifier.

7. The fault diagnosis method based on domain separation against pseudo-label network according to claim 6, characterized in that: The outputting of the target domain diagnosis result includes inputting the target domain sample into a DSAPN fault diagnosis model that has completed the fixed target domain sample classification point and outputting the target domain diagnosis result.

8. A system using the fault diagnosis method based on domain separation against pseudo-label network as claimed in any one of claims 1 to 7, characterized in that: Including data collection module, model building module, and prediction module; The data acquisition module is used to obtain mechanical equipment vibration source domain signals and target domain signals for feature extraction and iterative update of pseudo labels; The model building module is used to build a DSAPN fault diagnosis model, establish a joint loss function for fault diagnosis, and update and iterate the model; The prediction module is used to input a target domain sample set, fix the target domain sample classification points, and output a target domain diagnosis result.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the fault diagnosis method based on domain separation against pseudo-label network described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the fault diagnosis method based on domain separation against pseudo-label network according to any one of claims 1 to 7 are implemented.

Images