Bearing fault diagnosis method and device based on digital twinborn and convergent attention visual perception

Through digital twins and aggregated attention visual perception models, automatic, real-time and accurate diagnosis of bearing failures is achieved, and the accuracy and efficiency problems caused by artificial dependence in traditional methods are solved, thereby improving the accuracy and efficiency of diagnosis.

CN119086067APending Publication Date: 2024-12-06元始智能科技(南通)有限公司
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411374192.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-29
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

Traditional bearing fault diagnosis relies on manual experience and professional knowledge, resulting in limited accuracy, reliability and efficiency of diagnostic results, making it difficult to achieve automatic, real-time and accurate fault diagnosis.

Method used

Using a method based on digital twins and aggregated attention visual perception, the aggregated attention visual perception model is constructed by obtaining the actual operating parameters and simulated operating parameters of the bearing, and iterative training is carried out to generate a fault diagnosis model to achieve automatic, real-time and accurate faults. diagnosis.

Benefits of technology

It improves the accuracy and efficiency of bearing fault diagnosis, reduces human error, can adapt to complex and changeable fault modes, enhances early fault detection capabilities, and solves the problems of accuracy, reliability and inefficiency in traditional methods.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119086067A_ABST
    Figure CN119086067A_ABST
Patent Text Reader

Abstract

The invention provides a bearing fault diagnosis method and device based on digital twinborn and convergent attention visual perception, and the method comprises the steps: obtaining the actual operation parameters of a target bearing, and obtaining the target input information according to the actual operation parameters of the target bearing; inputting the target input information into a fault diagnosis model to obtain fault diagnosis prediction information of the target bearing; wherein the fault diagnosis model is obtained by training an aggregation attention visual perception model based on actual operation parameters and simulated operation parameters of the sample bearing and fault category labels corresponding to the actual operation parameters and the simulated operation parameters; the simulation operation parameters are obtained by simulating a dynamical model in a digital twin model corresponding to the sample bearing; the convergent attention visual perception model is constructed based on fusion of the convergent attention model and the visual perception model. According to the invention, the fault diagnosis of the bearing can be automatically and accurately carried out in real time through the fault diagnosis model.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a bearing fault diagnosis method and device based on digital twin and aggregated attention visual perception. Background Art

[0002] In the rapidly developing field of online wireless Internet of Things, motors are also widely used, and monitoring the health of bearings has become the key to rapid repair and precise maintenance. Therefore, it is necessary to study effective bearing fault diagnosis methods to accurately predict bearing failures in the early stages and replace them in time to prevent serious damage and catastrophic consequences to the motor system due to failures.

[0003] Traditional bearing fault diagnosis usually relies on manual experience and professional knowledge to diagnose and test each bearing one by one. However, manual experience and professional knowledge have certain subjective factors, which leads to certain limitations on the accuracy, reliability and efficiency of the diagnostic results.

[0004] Therefore, there is an urgent need for a bearing fault diagnosis method and device based on digital twins and aggregated attention visual perception to solve the above problems. Summary of the invention

[0005] The present invention provides a bearing fault diagnosis method and device based on digital twin and aggregated attention visual perception, so as to solve the defects of accuracy, reliability and efficiency of diagnosis results in the prior art, and realize automatic, real-time and accurate bearing fault diagnosis.

[0006] The present invention provides a bearing fault diagnosis method based on digital twin and aggregated attention visual perception, comprising: Acquire actual operating parameters of a target bearing, and acquire target input information according to the actual operating parameters of the target bearing; Inputting the target input information into a fault diagnosis model to obtain fault diagnosis prediction information of the target bearing; Among them, the fault diagnosis model is obtained by training the aggregated attention visual perception model based on the actual operating parameters and simulated operating parameters of the sample bearing, and the fault category labels corresponding to the actual operating parameters and the simulated operating parameters; the simulated operating parameters are obtained by simulating the dynamic model in the digital twin model corresponding to the sample bearing; the aggregated attention visual perception model is constructed based on the fusion of the aggregated attention model and the visual perception model.

[0007] According to a bearing fault diagnosis method based on digital twin and aggregated attention visual perception provided by the present invention, the fault diagnosis model is trained based on the following steps: Taking the preset fault state and the preset constraint conditions as input parameters, the dynamic model in the digital twin model corresponding to the sample bearing is simulated to obtain the simulated operating parameters of the sample bearing; the preset fault state includes a single fault state and a compound fault state, and the preset constraint conditions are conditions for constraining the operating conditions of the bearing; Marking the actual operating parameters and the simulated operating parameters of the sample bearing by fault category, and obtaining the fault category labels corresponding to the actual operating parameters and the simulated operating parameters respectively; Preprocessing the actual operation parameters and the simulated operation parameters respectively to obtain first sample input information corresponding to the actual operation parameters and second sample input information corresponding to the simulated operation parameters; the preprocessing includes continuous wavelet transform processing; Taking the first sample input information and the second sample input information as sample data, and taking the fault category labels corresponding to the actual operating parameters and the simulated operating parameters respectively as sample labels, a sample data set is constructed; The aggregated attention visual perception model is iteratively trained according to the sample data set to obtain the fault diagnosis model.

[0008] According to a bearing fault diagnosis method based on digital twin and aggregated attention visual perception provided by the present invention, the aggregated attention visual perception model is iteratively trained according to the sample data set to obtain the fault diagnosis model, including: Inputting the first sample input information and the second sample input information in the sample data set into the aggregated attention visual perception model to obtain the fault diagnosis prediction information corresponding to the actual operating parameters and the simulated operating parameters respectively; Acquire a first loss function according to a maximum mean difference between the first sample input information and the second sample input information; Obtaining a second loss function according to the deviation between the fault diagnosis prediction information corresponding to the actual operating parameters and the simulated operating parameters and the fault category labels corresponding to the actual operating parameters and the simulated operating parameters; The first loss function and the second loss function are combined to iteratively train the aggregated attention visual perception model to obtain the fault diagnosis model.

[0009] According to a bearing fault diagnosis method based on digital twin and aggregated attention visual perception provided by the present invention, the first loss function and the second loss function are combined to iteratively train the aggregated attention visual perception model to obtain the fault diagnosis model, including: Performing weighted addition on the first loss function and the second loss function to obtain a target loss function; According to the target loss function, the aggregated attention visual perception model is iteratively trained to obtain the fault diagnosis model.

[0010] According to a bearing fault diagnosis method based on digital twin and aggregated attention visual perception provided by the present invention, the dynamic model is calculated based on the following formula: ; in, , , , , are the horizontal translation displacement, vertical translation displacement, mass, damping and stiffness of the inner ring of the virtual bearing in the digital twin model, respectively. and are the first-order derivative corresponding to the horizontal translation displacement and the first-order derivative corresponding to the vertical translation displacement of the inner ring, respectively. and are respectively the second-order derivative corresponding to the horizontal translation displacement and the second-order derivative corresponding to the vertical translation displacement of the inner ring; , , , , are the horizontal translation displacement, vertical translation displacement, mass, damping and stiffness of the outer ring of the virtual bearing, respectively. and are respectively the first-order derivative corresponding to the horizontal translation displacement and the first-order derivative corresponding to the vertical translation displacement of the outer ring, and are respectively the second-order derivative corresponding to the horizontal translation displacement and the second-order derivative corresponding to the vertical translation displacement of the outer ring; , , , are the vertical translation displacement, mass, damping and stiffness of the unit resonator of the virtual bearing, respectively, and are respectively the first-order derivative and the second-order derivative of the vertical translation displacement of the unit resonator; is the radial load applied to the inner ring; and are the Hertzian contact forces in the horizontal and vertical directions of the virtual bearing, respectively; is the acceleration due to gravity.

[0011] According to a bearing fault diagnosis method based on digital twin and aggregated attention visual perception provided by the present invention, the aggregated attention visual perception model includes a preprocessing layer, a first feature layer, a downsampling layer, an aggregated attention layer, a second feature layer, a pooling layer and a fully connected layer; Wherein, the preprocessing layer is constructed based on the convolutional layer and the residual layer; Each of the first feature layer and the second feature layer is constructed based on a first feature block and a second feature block, the first feature block is constructed based on a plurality of convolutional layers with different convolutional kernel sizes, an SE attention layer, and a feedforward network layer; the second feature block is constructed based on a plurality of convolutional layers with different convolutional kernel sizes and a feedforward network layer; The downsampling layer is constructed based on the second feature block, a plurality of convolutional layers with different convolutional kernel sizes, and a feedforward network layer.

[0012] According to a bearing fault diagnosis method based on digital twin and aggregated attention visual perception provided by the present invention, the target input information is obtained according to the actual operating parameters of the target bearing, including: Performing continuous wavelet transform processing on the actual operating parameters of the target bearing to obtain time-frequency features of multiple different scales; The target input information is obtained according to the time-frequency features of multiple different scales.

[0013] The present invention also provides a bearing fault diagnosis device based on digital twin and aggregated attention visual perception, comprising: An acquisition unit, used to acquire actual operating parameters of a target bearing, and acquire target input information according to the actual operating parameters of the target bearing; A diagnosis prediction unit, used to input the target input information into a fault diagnosis model to obtain fault diagnosis prediction information of the target bearing; Among them, the fault diagnosis model is obtained by training the aggregated attention visual perception model based on the actual operating parameters and simulated operating parameters of the sample bearing, and the fault category labels corresponding to the actual operating parameters and the simulated operating parameters; the simulated operating parameters are obtained by simulating the dynamic model in the digital twin model corresponding to the sample bearing; the aggregated attention visual perception model is constructed based on the fusion of the aggregated attention model and the visual perception model.

[0014] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, a bearing fault diagnosis method based on digital twins and aggregated attention visual perception as described in any one of the above-mentioned methods is implemented.

[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a bearing fault diagnosis method based on digital twins and aggregated attention visual perception as described in any one of the above.

[0016] The present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements a bearing fault diagnosis method based on digital twins and aggregated attention visual perception as described in any one of the above.

[0017] The bearing fault diagnosis method and device based on digital twin and aggregated attention visual perception provided by the present invention iteratively train the aggregated attention visual perception model constructed by fusion of the aggregated attention model and the visual perception model based on the simulated operating parameters of the sample bearing obtained by simulating the dynamic model in the digital twin model, and the actual operating parameters of the sample bearing. The fault diagnosis model obtained by training focuses on the key areas in the time-frequency diagram for global perception, while being able to perform fine-grained perception of each pixel point, thereby automatically, real-time and accurately diagnosing bearing faults, effectively solving the problems of accuracy, reliability and low efficiency of diagnosis results caused by human subjective factors. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are 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.

[0019] Figure 1 It is a flow chart of the bearing fault diagnosis method based on digital twin and aggregated attention visual perception provided by the present invention.

[0020] Figure 2 It is a schematic diagram of a time domain signal before continuous wavelet transformation and a time-frequency diagram after continuous wavelet transformation provided by the present invention.

[0021] Figure 3 It is a flow chart of the fault diagnosis model training provided by the present invention.

[0022] Figure 4 It is a structural schematic diagram of the digital twin model corresponding to the sample bearing provided by the present invention.

[0023] Figure 5 It is a structural schematic diagram of the aggregated attention visual perception model provided by the present invention.

[0024] Figure 6It is a confusion matrix diagram comparing the bearing fault diagnosis method provided by the present invention with the prior art method based on deep neural network.

[0025] Figure 7 It is a structural schematic diagram of a bearing fault diagnosis device based on digital twin and aggregated attention visual perception provided by the present invention.

[0026] Figure 8 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. 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.

[0028] Bearing failure is one of the most common failures in rotating machinery, accounting for 41% of all types of motor failures. Under harsh operating conditions such as rapid changes in load and speed, tiny cracks and fissures on the raceways and rolling elements will gradually worsen. During the period of bearing failure, if the fault prediction is not carried out in time, it may cause abnormal operating conditions, serious damage to the machine and its ancillary equipment, and even serious personal injury. In order to extend the service life of the motor, save the cost of regular replacement of parts, and reduce the economic losses caused by sudden accidents, the field of bearing fault diagnosis has received widespread attention. In the rapidly developing field of online wireless Internet of Things, motors are also widely used, and monitoring the health of bearings has become the key to rapid repair and precise maintenance. Therefore, it is necessary to study effective bearing fault diagnosis methods to accurately predict bearing failures in the early stage and replace them in time to prevent serious damage and catastrophic consequences of motor systems due to failures.

[0029] Traditional analytical methods for bearing fault diagnosis have some disadvantages. These methods usually rely on manual experience and expertise, which may be affected by subjective factors, resulting in certain limitations on the accuracy and reliability of the diagnostic results. In addition, traditional methods may take a long time to analyze and judge, and are inefficient. Moreover, they may have difficulty accurately diagnosing complex bearing fault situations and fail to detect potential problems in a timely manner. At the same time, traditional methods may encounter difficulties when processing large amounts of data and find it difficult to effectively extract useful information.

[0030] Therefore, the traditional technology relies on manual experience and professional knowledge to perform diagnostic tests on each bearing one by one, which will lead to certain limitations on the accuracy, reliability and efficiency of the diagnostic results.

[0031] In this regard, an embodiment of the present application provides a bearing fault diagnosis method based on digital twins and aggregated attention visual perception. The method iteratively trains an aggregated attention visual perception model constructed by fusion of an aggregated attention model and a visual perception model based on the simulated operating parameters of a sample bearing obtained by simulating a dynamic model in a digital twin model, as well as the actual operating parameters of the sample bearing, so as to automatically output bearing fault diagnosis prediction information based on the fault diagnosis model obtained through training. This effectively solves the shortcomings of traditional analysis methods for bearing fault diagnosis through artificial intelligence methods, specifically, automatically extracts features from data through machine learning algorithms, improves the objectivity and accuracy of diagnosis, and reduces human errors. At the same time, artificial intelligence technology can quickly process and analyze large amounts of data, improve diagnostic efficiency, and can adapt to complex and changeable fault modes, enhance the ability to detect early faults, and thus improve the accuracy and real-time performance of the overall diagnosis.

[0032] Figure 1 is a flow chart of a bearing fault diagnosis method based on digital twin and aggregated attention visual perception provided by the present invention; Figure 1 As shown, the method includes step 110 and step 120.

[0033] Step 110, obtaining actual operating parameters of a target bearing, and obtaining target input information according to the actual operating parameters of the target bearing.

[0034] The target bearing is the bearing for which fault diagnosis is required; the actual operating parameters include multi-dimensional time series parameters formed by collecting the parameters generated by the target bearing during actual operation at each sampling point in the current cycle, including but not limited to acceleration vibration signals in the horizontal and vertical directions, which are not specifically limited in the embodiments of the present application. The sampling frequency of each cycle can be configured according to actual needs, such as 64kHz.

[0035] Optionally, when fault diagnosis of the target bearing is required, the actual operating parameters of the target bearing may be obtained first; here, the actual operating parameters of the target bearing may be input by the user, or shared through an acquisition device, and the embodiments of the present application do not specifically limit this.

[0036] Next, the actual operating parameters of the target bearing are preprocessed to obtain target input information; the so-called preprocessing includes but is not limited to format conversion of the actual operating parameters according to the model type of the fault diagnosis model and time-frequency feature extraction of the actual operating parameters, which is not specifically limited in the embodiments of the present application.

[0037] Exemplarily, acquiring target input information according to actual operating parameters of the target bearing includes: Performing continuous wavelet transform processing on the actual operating parameters of the target bearing to obtain time-frequency features of multiple different scales; The target input information is obtained according to the time-frequency features of multiple different scales.

[0038] Optionally, since continuous wavelet transform processing is used to convert actual operating parameters from time domain signals into time-frequency signals (or time-frequency diagrams, or time-frequency features); in addition, converting actual operating parameters into time-frequency diagrams through continuous wavelet transform has significant advantages: first, continuous wavelet transform can effectively decompose vibration signals, reveal the time-frequency characteristics of different frequency components, and visualize complex signals; second, video diagrams intuitively display the changing trend of vibration signals over time, which is convenient for rapid identification of fault characteristics; third, this conversion method improves the efficiency and accuracy of signal analysis, and helps to achieve real-time monitoring and early warning of bearing faults.

[0039] Therefore, before the actual operating parameters of the target bearing are input into the fault diagnosis model, the embodiment of the present application may first perform continuous wavelet transform processing on the actual operating parameters of the target bearing to decompose the actual operating parameters of the target bearing to obtain multiple time-frequency features of different scales, such as Figure 2 shown.

[0040] After obtaining time-frequency features of multiple different scales, they are spliced ​​together to form target input information so that the fault diagnosis model can identify fault characteristics more quickly and accurately, thereby improving the efficiency and accuracy of fault analysis and helping to achieve real-time monitoring and early warning of bearing faults.

[0041] Among them, for any signal , that is, the operating parameters of the bearing, its continuous wavelet transform can be realized by the following formula: ; in, For signal The continuous wavelet transform result of . is the translation factor; is the scale factor, is the wavelet basis function, is the time variable; the calculation formula of the wavelet basis function is as follows: .

[0042] Step 120, input the target input information into the fault diagnosis model to obtain the fault diagnosis prediction information of the target bearing; wherein the fault diagnosis model is obtained by training the aggregated attention visual perception model based on the actual operating parameters and simulated operating parameters of the sample bearing, and the fault category labels corresponding to the actual operating parameters and the simulated operating parameters; the simulated operating parameters are obtained by simulating the dynamic model in the digital twin model corresponding to the sample bearing; the aggregated attention visual perception model is constructed based on the fusion of the aggregated attention model and the visual perception model.

[0043] Optionally, after obtaining the target input information, the target input information may be input into a fault diagnosis model so that the fault diagnosis model extracts fault features and diagnoses fault categories of the target input information, thereby outputting corresponding fault diagnosis prediction information. The fault diagnosis prediction information here may be single fault information or composite fault information formed by a mixture of multiple single faults, which is not specifically limited in the embodiments of the present application.

[0044] The fault diagnosis model here can be trained based on the following steps: Firstly, a digital twin model of a large number of sample bearings under multi-source fault categories is established to effectively reflect the operating characteristics of the bearings, and then simulate and generate the simulated operating parameters of the sample bearings under different fault states; Next, a sample data set is constructed based on the actual operating parameters and simulated operating parameters of the sample bearing, as well as the fault category labels corresponding to the actual operating parameters and the simulated operating parameters. The so-called sample data set construction can be constructed after performing at least one preprocessing on the actual operating parameters and the simulated operating parameters of the sample bearing; the so-called at least one preprocessing includes but is not limited to at least one of missing data filling, data enhancement, and continuous wavelet transform processing.

[0045] In addition, a converged attention visual perception model is constructed; the converged attention visual perception model here can be a model based on the visual perception model, the feature extraction module of the visual perception model is pruned, and the converged attention model is added to the pruned feature extraction module, so that the fault diagnosis model trained thereby is more lightweight and can be more focused on the key areas in the time-frequency diagram, so as to achieve global perception while being able to perform fine-grained perception of each pixel point, thereby better capturing the fault characteristics of the bearing and improving the accuracy of bearing fault diagnosis.

[0046] The initialization of the aggregated attention visual perception model here can be achieved by pre-training based on the operating parameters and fault diagnosis labels of equipment in general fields, or it can be achieved by random initial parameters, and the embodiments of the present application do not specifically limit this.

[0047] Next, based on the sample data set, the aggregated attention visual perception model constructed by fusing the aggregated attention model and the visual perception model is iteratively trained to obtain a model that can accurately deduce and output the fault diagnosis prediction information of the bearing from the actual operating parameters of the bearing.

[0048] The method provided in the embodiment of the present application iteratively trains the aggregated attention visual perception model constructed by fusing the aggregated attention model and the visual perception model based on the simulated operating parameters of the sample bearing obtained by simulating the dynamic model in the digital twin model, and the actual operating parameters of the sample bearing, so that the fault diagnosis model obtained by training can focus on the key areas in the time-frequency diagram for global perception while performing fine-grained perception of each pixel point, thereby automatically, real-time and accurately diagnosing bearing faults, effectively solving the problems of accuracy, reliability and low efficiency of diagnosis results caused by human subjective factors.

[0049] In some embodiments, the fault diagnosis model is trained based on the following steps: Taking the preset fault state and the preset constraint conditions as input parameters, the dynamic model in the digital twin model corresponding to the sample bearing is simulated to obtain the simulated operating parameters of the sample bearing; the preset fault state includes a single fault state and a compound fault state, and the preset constraint conditions are conditions for constraining the operating conditions of the bearing; Marking the actual operating parameters and the simulated operating parameters of the sample bearing by fault category, and obtaining the fault category labels corresponding to the actual operating parameters and the simulated operating parameters respectively; Preprocessing the actual operation parameters and the simulated operation parameters respectively to obtain first sample input information corresponding to the actual operation parameters and second sample input information corresponding to the simulated operation parameters; the preprocessing includes continuous wavelet transform processing; Taking the first sample input information and the second sample input information as sample data, and taking the fault category labels corresponding to the actual operating parameters and the simulated operating parameters respectively as sample labels, a sample data set is constructed; The aggregated attention visual perception model is iteratively trained according to the sample data set to obtain the fault diagnosis model.

[0050] Figure 3 is a flow chart of the fault diagnosis model training provided by the present invention; Figure 3 As shown, the training steps of the fault diagnosis model further include: Step 310, since deep learning fault diagnosis methods usually rely on a training data set containing all potential fault mode samples, these samples may not be available in a specific industrial environment, which limits the model performance; in this regard, an embodiment of the present application builds a digital twin model corresponding to the sample bearing, and uses simulation technology to simulate the collection environment of the operating parameters of the bearing in the actual operating scenario, thereby generating virtual operating parameters under different fault conditions, thereby enriching the training data set, thereby solving the problem of specific samples not being available.

[0051] Figure 4 is a structural schematic diagram of a digital twin model corresponding to a sample bearing provided by the present invention; Figure 4 As shown in the figure, the digital twin model (or bearing fault simulation test bench) corresponding to the sample bearing is built with the actual attribute information of the sample bearing, including but not limited to the type, model, structure, etc. of the sample bearing; and during the experiment, a variety of preset fault states are configured, such as outer ring fault (or OF), inner ring fault (or IF), and composite fault of outer ring fault and inner ring fault, and preset constraints are imposed, including but not limited to speed constraints, radial force constraints and load torque constraints, so as to simulate the dynamic model in the digital twin model corresponding to the sample bearing, and collect the operating parameters of the virtual bearing in the digital twin model at a preset sampling frequency (such as 64kHz), such as the acceleration vibration signal in the vertical direction, or the acceleration vibration signals in the horizontal and vertical directions, so as to obtain the simulated operating parameters of the sample bearing.

[0052] Among them, the speed constraint, radial force constraint and load torque constraint can be configured according to actual needs, such as setting the speed to 1500r / min, the radial force to 1000N, and the load torque to 0.7Nm.

[0053] The sample bearing may be a bearing of the same type or a different type as the target bearing, and may be configured according to actual needs. For example, an SKF6203 deep groove ball bearing may be used as a sample bearing, etc. This embodiment of the present application does not specifically limit this.

[0054] Here, the dynamic model in the digital twin model can be constructed based on the degree of freedom dynamic model of each component simulating the bearing in the digital twin model.

[0055] The degree of freedom dynamic model here includes, but is not limited to, a degree of freedom dynamic model simulating the inner ring and outer ring of the bearing under horizontal translation displacement and vertical translation displacement respectively, and a degree of freedom dynamic model of the unit resonator under vertical translation displacement.

[0056] Exemplarily, the kinetic model is calculated based on the following formula: ; in, , , , , are the horizontal translation displacement, vertical translation displacement, mass, damping and stiffness of the inner ring of the virtual bearing in the digital twin model, respectively. and are the first-order derivative corresponding to the horizontal translation displacement and the first-order derivative corresponding to the vertical translation displacement of the inner ring, respectively. and are respectively the second-order derivative corresponding to the horizontal translation displacement and the second-order derivative corresponding to the vertical translation displacement of the inner ring; , , , , are the horizontal translation displacement, vertical translation displacement, mass, damping and stiffness of the outer ring of the virtual bearing, respectively. and are respectively the first-order derivative corresponding to the horizontal translation displacement and the first-order derivative corresponding to the vertical translation displacement of the outer ring, and are respectively the second-order derivative corresponding to the horizontal translation displacement and the second-order derivative corresponding to the vertical translation displacement of the outer ring; , , , are the vertical translation displacement, mass, damping and stiffness of the unit resonator of the virtual bearing, respectively, and are respectively the first-order derivative and the second-order derivative of the vertical translation displacement of the unit resonator; is the radial load applied to the inner ring; and are the Hertzian contact forces in the horizontal and vertical directions of the virtual bearing, respectively; is the acceleration due to gravity.

[0057] Among them, the Hertz contact force in the horizontal and vertical directions can be expressed as: ; ; in, is the contact stiffness of the virtual bearing; The virtual bearing The contact deformation parameters between the roller and the raceway; The virtual bearing Angular position of the rollers; is the number of rollers in the virtual bearing.

[0058] No. The contact deformation parameters of roller and raceway It can be expressed as: ; in, is the radial clearance of the virtual bearing.

[0059] No. Angular position of the rollers It can be expressed as: ; in, is the rotation speed of the cage of the virtual bearing, which can be specifically expressed as: ; in, is the shaft speed of the virtual bearing, is the roller diameter of the virtual bearing, is the contact angle of the virtual bearing, is the pitch diameter of the virtual bearing.

[0060] From the above, it can be seen that the construction of a refined dynamic system by integrating the multi-component degree-of-freedom dynamic model is intended to ensure that the simulation process can accurately simulate the parameter changes in actual operation, thereby capturing the operating characteristics of the bearing, and then improving the accuracy and reliability of the simulated operating parameters, so that the fault diagnosis model can be trained more accurately and efficiently based on these parameters in the future.

[0061] Step 320 , marking the actual operating parameters and simulated operating parameters of the sample bearing with fault categories, so as to obtain fault category labels corresponding to the actual operating parameters and simulated operating parameters respectively.

[0062] Step 330, continuous wavelet transform processing is performed on the simulated operation parameters generated by the simulation and the actual operation parameters acquired by the actual collection, so as to convert the actual operation parameters and the simulated operation parameters from time series data into time-frequency diagrams, thereby obtaining first sample input information corresponding to the actual operation parameters and second sample input information corresponding to the simulated operation parameters. The specific conversion steps can be referred to the continuous wavelet transform processing steps of the target operation parameters, which will not be repeated here.

[0063] Step 340 , constructing a sample data set using the first sample input information and the second sample input information obtained through data processing as sample data, and using the fault category labels corresponding to the actual operating parameters and the simulated operating parameters as sample labels.

[0064] Step 350, constructing a converged attention visual perception model (or TransRepVit model), which is used to simulate human visual perception, so that the model can focus more on the key areas in the time-frequency diagram, thereby achieving global perception, and at the same time being able to perform fine-grained perception of each pixel, thereby improving the accuracy of bearing fault diagnosis.

[0065] Step 360, based on the sample data set, iteratively train the aggregated attention visual perception model, so as to use the trained aggregated attention visual perception model with the best model performance among the aggregated attention visual perception models after multiple iterative training as the fault diagnosis model.

[0066] The method provided in the embodiment of the present application drives the dynamic simulation of the sample bearing digital twin model through preset fault states and constraints, accurately obtains the simulated operating parameters under different health states, and performs fault category marking and continuous wavelet transform synchronously with the actual operating parameters to convert the operating parameters into time-frequency diagrams, and constructs a sample data set that comprehensively reflects the difference between reality and simulation based on the converted time-frequency diagrams and the marked fault category labels, and then iteratively trains the aggregated attention visual perception model so that the trained aggregated attention visual perception model can automatically learn important features as well as local and global features at different time and frequency scales, and finally obtain an efficient and accurate bearing fault diagnosis model, which significantly improves the accuracy and generalization ability of fault diagnosis.

[0067] In some embodiments, the step of obtaining the fault diagnosis model in step 360 further includes: Inputting the first sample input information and the second sample input information in the sample data set into the aggregated attention visual perception model to obtain the fault diagnosis prediction information corresponding to the actual operating parameters and the simulated operating parameters respectively; Acquire a first loss function according to a maximum mean difference between the first sample input information and the second sample input information; Obtaining a second loss function according to the deviation between the fault diagnosis prediction information corresponding to the actual operating parameters and the simulated operating parameters and the fault category labels corresponding to the actual operating parameters and the simulated operating parameters; The first loss function and the second loss function are combined to iteratively train the aggregated attention visual perception model to obtain the fault diagnosis model.

[0068] Optionally, after obtaining the sample data set based on step 340, the first sample input information and the second sample input information in the sample data set may be input into the aggregated attention visual perception model so that the aggregated attention visual perception model can perform fault feature extraction and fault category diagnosis on the first sample input information and the second sample input information, thereby outputting fault diagnosis prediction information corresponding to the actual operating parameters and the simulated operating parameters.

[0069] Subsequently, a cross entropy loss calculation is performed based on the deviation between the fault diagnosis prediction information corresponding to the actual operating parameters and the simulated operating parameters and the fault category labels corresponding to the actual operating parameters and the simulated operating parameters to obtain a second loss function.

[0070] In addition, in order to improve the accuracy and real-time performance of the overall diagnosis, the maximum mean difference technology is introduced in the model training process to obtain the first loss function by minimizing the distribution difference between the simulated operating parameters and the actual operating parameters in the feature space, so as to encourage the aggregated attention visual perception model to learn more general and robust feature representations during the model training process and reduce overfitting of specific training data, that is, to minimize the distribution difference between the simulated operating parameters and the actual operating parameters, so that the model trained on the simulated operating parameters can better adapt to the actual operating parameters, thereby improving the performance of the model in fault diagnosis on unseen data (that is, applying the actual operating parameters for fault diagnosis).

[0071] Among them, the maximum mean difference, that is, the first loss function The calculation formula can be expressed as: ; in, For the The spectrum power entropy characteristics corresponding to the sample input information of the simulation operation parameters; No. The spectrum power entropy characteristics corresponding to the sample input information of the actual operating parameters; , , represents the sample data volume, It means that the characteristic kernel is The regenerating Hilbert space norm of .

[0072] Subsequently, the first loss function and the second loss function are combined to iteratively train the aggregated attention visual perception model to obtain a fault diagnosis model that can accurately output fault diagnosis prediction information based on actual operating parameters.

[0073] The combination here may be weighted addition, direct addition, model fusion, etc., which is not specifically limited in the embodiments of the present application.

[0074] Exemplarily, the combining the first loss function and the second loss function to iteratively train the aggregated attention visual perception model to obtain the fault diagnosis model includes: Performing weighted addition on the first loss function and the second loss function to obtain a target loss function; According to the target loss function, the aggregated attention visual perception model is iteratively trained to obtain the fault diagnosis model.

[0075] Optionally, the first loss function and the second loss function can be weightedly added to obtain the target loss function, so that during the model training process, both the deviation of the model prediction results and the distribution differences between the model operating parameters and the actual operating parameters in the feature space can be considered, thereby improving the accuracy and robustness of the model training.

[0076] Then, based on the target loss function, the aggregated attention visual perception model is iteratively trained to obtain a fault diagnosis model that can accurately output fault diagnosis prediction information according to actual operating parameters.

[0077] The method provided in the embodiment of the present application inputs the first sample input information and the second sample input information into the aggregated attention visual perception model, constructs a first loss function based on the maximum mean difference between the two, and constructs a second loss function based on the deviation between the fault diagnosis prediction information of the actual and simulated operating parameters and the corresponding fault category label. The two types of loss functions are combined for iterative training, which effectively improves the accuracy and generalization ability of the aggregated attention visual perception model in fault diagnosis, and finally obtains a high-performance fault diagnosis model.

[0078] In some embodiments, the aggregated attention visual perception model constructed in step 350 includes a preprocessing layer, a first feature layer, a downsampling layer, an aggregated attention layer, a second feature layer, a pooling layer, and a fully connected layer; Wherein, the preprocessing layer is constructed based on the convolutional layer and the residual layer; Each of the first feature layer and the second feature layer is constructed based on a first feature block and a second feature block, the first feature block is constructed based on a plurality of convolutional layers with different convolutional kernel sizes, an SE attention layer, and a feedforward network layer; the second feature block is constructed based on a plurality of convolutional layers with different convolutional kernel sizes and a feedforward network layer; The downsampling layer is constructed based on the second feature block, a plurality of convolutional layers with different convolutional kernel sizes, and a feedforward network layer.

[0079] Figure 5 is a schematic diagram of the structure of the aggregated attention visual perception model provided by the present invention; Figure 5As shown in the figure, the aggregated attention visual perception model (or TransRepVit model) is constructed by sequentially connected preprocessing layers (or Stem layers), first feature layers (or Stage1 layers), down sampling layers (or Down sample), aggregated attention layers (or AggregatedAttention layers), second feature layers (or Stage2 layers), pooling layers (or Pooling) and fully connected layers (FC).

[0080] The dimension of the input information input to the aggregated attention visual perception model is B×3×H×W, where B usually represents the batch size, 3 represents the number of channels (RGB), and X and W represent the height and width of the input data, respectively.

[0081] Among them, the Stem layer is the first module layer of the TransRepVit model, which includes a convolution layer, a multi-layer activation function layer and a residual layer. For example, the Stem layer can be based on a deep convolution layer (or DW convolution layer) with a convolution kernel size of 3x3 and a step size of 2, two activation function layers (ACT layers), a convolution layer with a convolution kernel size of 1x1, and a residual layer. The activation function layer can use the ReLu activation function to enable the model to learn more complex patterns, and the residual layer is used to jump connect the first activation layer and the second activation function layer. Correspondingly, after the input information of dimension B×3×H×W is input into the Stem layer, the Stem layer first feeds the input information to the DW convolution layer with a stride of 2 and a convolution kernel size of 3x3 to extract the feature map of the corresponding scale of the input information to obtain the first feature map. Subsequently, the first feature map is subjected to nonlinear processing through the first activation layer so that the model can learn more complex patterns, thereby obtaining the second feature map. Subsequently, the second feature map is subjected to the convolution layer with a convolution kernel size of 1x1 and the second activation layer in turn to obtain the third feature map. Finally, the third feature map and the second feature map are residually connected through the residual layer to reduce the size of the input image and extract the basic features of the image, thereby obtaining the fourth feature map.

[0082] The Stage1 layer and the Stage2 layer are repeated modules with the same structure in the model, which are constructed by the first feature block (or RepVitSEBlock) and the second feature block (or RepVitBlock); RepVitBlock is composed of a 3x3 DW convolution layer, a 1x1 DW convolution, and a feedforward neural network (FFN), which are used to extract multi-scale features from the input feature map through a 3x3 DW convolution layer and a 1x1 DW convolution, and then perform multi-scale feature fusion through a feedforward neural network to output the high-level features of the feature map. RepVitSEBlock is composed of a 3x3 DW convolution layer, a 1x1 DW convolution, an SE attention mechanism, and a feedforward neural network (FFN); it adds the SE attention mechanism on the basis of RepVitBlock to learn the dependencies between different channels and assign corresponding weights to each channel, thereby obtaining the high-level features of the input feature map. Therefore, the role of Stage is to further extract high-level features of the feature map.

[0083] The downsampling layer is constructed based on RepVitBlock, multiple convolutional layers with different convolution kernel sizes, and feedforward network layers. For example, the downsampling layer is constructed based on the sequentially connected RepVitBlock, 3x3 DW convolutional layer, 1x1 DW convolutional layer, and FNN; the 3x3 DW convolutional layer and 1x1 DW convolutional layer are used to perform spatial downsampling and modulate channel dimensions, respectively. RepVitBlock is used to further deepen the downsampling layer to sample high-level features. The FFN module is placed after the 1×1 convolution to memorize more potential information, thereby reducing the amount of calculation of subsequent layers while improving the effectiveness of the output features.

[0084] Aggregate Attention (or AA) is a pixel-level attention mechanism inspired by biological vision systems, which allows fine-grained perception of each query while maintaining global information perception; the aggregate attention layer can achieve feature extraction through the following formula: ; in, is a set of learnable labels, is the attention matrix after the sliding window starting from the learnable marker. It's a pixel The query matrix, In pixels The matrix of values ​​of the sliding window centered at . is the attention matrix after feature aggregation. is the value matrix after feature aggregation.

[0085] In summary, the aggregated attention visual perception model achieves effective extraction and fusion from input information to high-level features through sequentially integrated preprocessing layers (Stem layers), feature layers (Stage1 and Stage2), downsampling layers, and aggregated attention layers (AA). The Stem layer uses deep convolution and residual connections to preliminarily extract image features and reduce the size; the Stage layer further deepens feature extraction through repeated feature blocks, among which RepVitSEBlock introduces the SE attention mechanism to enhance the inter-channel dependency; the downsampling layer achieves spatial downsampling and feature modulation through RepVitBlock combined with multi-scale convolution and FFN; and the aggregated attention layer (AA) as the core, inspired by the biological visual system, significantly improves the model's ability to extract and understand image features through the attention mechanism of fine-grained perception and global information retention, and ultimately achieves efficient and accurate visual perception effects.

[0086] In order to verify the effectiveness of the bearing fault diagnosis method provided in the embodiment of the present application, the embodiment of the present application will compare the performance of the fault diagnosis model trained by the improved aggregated attention visual perception model provided in the embodiment of the present application with the existing models and methods under the same parameter configuration conditions.

[0087] Figure 6 A confusion matrix diagram comparing the bearing fault diagnosis method provided by the present invention with the existing method based on deep neural network. Figure 6 The horizontal axis is the fault diagnosis prediction result (predict), and the vertical axis is the fault diagnosis true result (True); Figure 6 (a) is the confusion matrix diagram of SCIE in the prior art; Figure 6 (b) is the confusion matrix diagram of BOTCL in the prior art; Figure 6 (c) is the confusion matrix diagram of GlideNet in the prior art; Figure 6 (d) is a confusion matrix diagram of the method provided in the embodiment of the present application (or Proposed).

[0088] like Figure 6 As shown, by comparing the method proposed in the embodiment of the present application with three popular methods based on deep neural networks, namely, a deep visual representation method based on partial correlation (SCIE for short), a bottleneck concept learner method that uses self-supervision and customized regularizers to learn bottleneck concepts (BOtCL for short), and a dense embedding network based on global, local and intrinsic features (GlideNet for short), it can be seen that the fault diagnosis model proposed in the embodiment of the present application is trained with a simulated bearing data set generated by a digital twin (or DT) model, so that it has good diagnostic performance.

[0089] In summary, the embodiment of the present application aims at the problem that the bearing operating parameters may not be collected in a specific industrial environment, and proposes a bearing fault diagnosis method based on digital twin and aggregated attention visual perception. By using artificial intelligence technology for automatic fault diagnosis, the results are fast and accurate. An operator only needs a short time to perform fault diagnosis on multiple rotating mechanical equipment at the same time. It greatly saves labor costs. And while improving the detection recall rate, it greatly speeds up the detection speed. The method provided in the embodiment of the present application achieves a high accuracy in bearing fault detection, greatly speeds up the detection speed while saving labor costs, and achieves high social and economic benefits. In addition, through simulation experiments with relevant data, the experimental results show that the method based on digital twins has achieved a high accuracy rate in bearing fault diagnosis tasks, and can accurately simulate and identify different bearing faults, further proving the effectiveness of the method provided in the embodiment of the present application.

[0090] The following is a description of the bearing fault diagnosis device based on digital twin and aggregated attention visual perception provided by the present invention. The bearing fault diagnosis device based on digital twin and aggregated attention visual perception described below and the bearing fault diagnosis method based on digital twin and aggregated attention visual perception described above can be referenced to each other.

[0091] Figure 7 is a structural schematic diagram of a bearing fault diagnosis device based on digital twin and aggregated attention visual perception provided by the present invention; Figure 7 As shown, the device includes an acquisition unit 710 and a diagnosis prediction unit 720 .

[0092] The acquisition unit 710 is used to acquire actual operating parameters of a target bearing, and acquire target input information according to the actual operating parameters of the target bearing.

[0093] The diagnosis prediction unit 720 is used to input the target input information into the fault diagnosis model to obtain the fault diagnosis prediction information of the target bearing; wherein the fault diagnosis model is obtained by training the aggregated attention visual perception model based on the actual operating parameters and simulated operating parameters of the sample bearing, and the fault category labels corresponding to the actual operating parameters and the simulated operating parameters; the simulated operating parameters are obtained by simulating the dynamic model in the digital twin model corresponding to the sample bearing; the aggregated attention visual perception model is constructed based on the fusion of the aggregated attention model and the visual perception model.

[0094] The device provided in the embodiment of the present application iteratively trains the aggregated attention visual perception model constructed by fusing the aggregated attention model and the visual perception model based on the simulated operating parameters of the sample bearing obtained by simulating the dynamic model in the digital twin model, and the actual operating parameters of the sample bearing, so that the fault diagnosis model obtained by training can focus on the key areas in the time-frequency diagram for global perception while performing fine-grained perception of each pixel point, thereby automatically, real-time and accurately diagnosing bearing faults, effectively solving the problems of accuracy, reliability and low efficiency of diagnosis results caused by human subjective factors.

[0095] Figure 8 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 8 As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830 and a communication bus 840, wherein the processor 810, the communication interface 820 and the memory 830 communicate with each other through the communication bus 840. The processor 810 may call the logic instructions in the memory 830 to execute a bearing fault diagnosis method based on digital twin and aggregated attention visual perception, the method comprising: obtaining actual operating parameters of a target bearing, and obtaining target input information according to the actual operating parameters of the target bearing; inputting the target input information into a fault diagnosis model to obtain fault diagnosis prediction information of the target bearing; wherein the fault diagnosis model is obtained by training an aggregated attention visual perception model based on the actual operating parameters and simulated operating parameters of a sample bearing, and the fault category labels corresponding to the actual operating parameters and the simulated operating parameters; the simulated operating parameters are obtained by simulating the dynamic model in the digital twin model corresponding to the sample bearing; the aggregated attention visual perception model is constructed by fusion of the aggregated attention model and the visual perception model.

[0096] In addition, the logic instructions in the above-mentioned memory 830 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention is essentially 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, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0097] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the bearing fault diagnosis method based on digital twins and aggregated attention visual perception provided by the above methods, the method including: obtaining actual operating parameters of the target bearing, and obtaining target input information based on the actual operating parameters of the target bearing; inputting the target input information into a fault diagnosis model to obtain fault diagnosis prediction information of the target bearing; wherein the fault diagnosis model is obtained by training an aggregated attention visual perception model based on the actual operating parameters and simulated operating parameters of the sample bearing, and the fault category labels corresponding to the actual operating parameters and the simulated operating parameters; the simulated operating parameters are obtained by simulating the dynamic model in the digital twin model corresponding to the sample bearing; and the aggregated attention visual perception model is constructed based on the fusion of the aggregated attention model and the visual perception model.

[0098] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the bearing fault diagnosis method based on digital twin and aggregated attention visual perception provided by the above-mentioned methods, the method comprising: obtaining actual operating parameters of a target bearing, and obtaining target input information based on the actual operating parameters of the target bearing; inputting the target input information into a fault diagnosis model to obtain fault diagnosis prediction information of the target bearing; wherein the fault diagnosis model is obtained by training an aggregated attention visual perception model based on the actual operating parameters and simulated operating parameters of a sample bearing, and the fault category labels corresponding to the actual operating parameters and the simulated operating parameters; the simulated operating parameters are obtained by simulating the dynamic model in the digital twin model corresponding to the sample bearing; and the aggregated attention visual perception model is constructed based on the fusion of the aggregated attention model and the visual perception model.

[0099] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0100] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0101] Finally, 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 aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. 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 bearing fault diagnosis method based on digital twin and aggregated attention visual perception, characterized in that: include: Acquire actual operating parameters of a target bearing, and acquire target input information according to the actual operating parameters of the target bearing; Inputting the target input information into a fault diagnosis model to obtain fault diagnosis prediction information of the target bearing; The fault diagnosis model is obtained by training the aggregated attention visual perception model based on the actual operating parameters and simulated operating parameters of the sample bearing, as well as the fault category labels corresponding to the actual operating parameters and the simulated operating parameters; The simulation operation parameters are obtained by simulating the dynamic model in the digital twin model corresponding to the sample bearing; the aggregated attention visual perception model is constructed by fusion of the aggregated attention model and the visual perception model.

2. The bearing fault diagnosis method based on digital twin and aggregated attention visual perception according to claim 1 is characterized in that: The fault diagnosis model is trained based on the following steps: Taking the preset fault state and the preset constraint conditions as input parameters, the dynamic model in the digital twin model corresponding to the sample bearing is simulated to obtain the simulated operating parameters of the sample bearing; the preset fault state includes a single fault state and a compound fault state, and the preset constraint conditions are conditions for constraining the operating conditions of the bearing; Marking the actual operating parameters and the simulated operating parameters of the sample bearing by fault category, and obtaining the fault category labels corresponding to the actual operating parameters and the simulated operating parameters respectively; Preprocessing the actual operation parameters and the simulated operation parameters respectively to obtain first sample input information corresponding to the actual operation parameters and second sample input information corresponding to the simulated operation parameters; the preprocessing includes continuous wavelet transform processing; Taking the first sample input information and the second sample input information as sample data, and taking the fault category labels corresponding to the actual operating parameters and the simulated operating parameters respectively as sample labels, a sample data set is constructed; The aggregated attention visual perception model is iteratively trained according to the sample data set to obtain the fault diagnosis model.

3. The bearing fault diagnosis method based on digital twin and aggregated attention visual perception according to claim 2 is characterized in that: The iterative training of the aggregated attention visual perception model according to the sample data set to obtain the fault diagnosis model includes: Inputting the first sample input information and the second sample input information in the sample data set into the aggregated attention visual perception model to obtain the fault diagnosis prediction information corresponding to the actual operating parameters and the simulated operating parameters respectively; Acquire a first loss function according to a maximum mean difference between the first sample input information and the second sample input information; Obtaining a second loss function according to the deviation between the fault diagnosis prediction information corresponding to the actual operating parameters and the simulated operating parameters and the fault category labels corresponding to the actual operating parameters and the simulated operating parameters; The first loss function and the second loss function are combined to iteratively train the aggregated attention visual perception model to obtain the fault diagnosis model.

4. The bearing fault diagnosis method based on digital twin and aggregated attention visual perception according to claim 3 is characterized in that: The combining the first loss function and the second loss function to iteratively train the aggregated attention visual perception model to obtain the fault diagnosis model includes: Performing weighted addition on the first loss function and the second loss function to obtain a target loss function; According to the target loss function, the aggregated attention visual perception model is iteratively trained to obtain the fault diagnosis model.

5. The bearing fault diagnosis method based on digital twin and aggregated attention visual perception according to claim 2 is characterized in that: The kinetic model is calculated based on the following formula: ; in, , , , , are the horizontal translation displacement, vertical translation displacement, mass, damping and stiffness of the inner ring of the virtual bearing in the digital twin model, respectively. and are the first-order derivative corresponding to the horizontal translation displacement and the first-order derivative corresponding to the vertical translation displacement of the inner ring, respectively. and are respectively the second-order derivative corresponding to the horizontal translation displacement and the second-order derivative corresponding to the vertical translation displacement of the inner ring; , , , , are the horizontal translation displacement, vertical translation displacement, mass, damping and stiffness of the outer ring of the virtual bearing, respectively. and are respectively the first-order derivative corresponding to the horizontal translation displacement and the first-order derivative corresponding to the vertical translation displacement of the outer ring, and are respectively the second-order derivative corresponding to the horizontal translation displacement and the second-order derivative corresponding to the vertical translation displacement of the outer ring; , , , are the vertical translation displacement, mass, damping and stiffness of the unit resonator of the virtual bearing, respectively, and are respectively the first-order derivative and the second-order derivative of the vertical translation displacement of the unit resonator; is the radial load applied to the inner ring; and are the Hertzian contact forces in the horizontal and vertical directions of the virtual bearing, respectively; is the acceleration due to gravity.

6. The bearing fault diagnosis method based on digital twin and aggregated attention visual perception according to any one of claims 1 to 5, characterized in that: The aggregate attention visual perception model includes a preprocessing layer, a first feature layer, a downsampling layer, an aggregate attention layer, a second feature layer, a pooling layer and a fully connected layer; Wherein, the preprocessing layer is constructed based on the convolutional layer and the residual layer; Each of the first feature layer and the second feature layer is constructed based on a first feature block and a second feature block, the first feature block is constructed based on a plurality of convolutional layers with different convolutional kernel sizes, an SE attention layer, and a feedforward network layer; the second feature block is constructed based on a plurality of convolutional layers with different convolutional kernel sizes and a feedforward network layer; The downsampling layer is constructed based on the second feature block, a plurality of convolutional layers with different convolutional kernel sizes, and a feedforward network layer.

7. The bearing fault diagnosis method based on digital twin and aggregated attention visual perception according to any one of claims 1 to 5, characterized in that: The step of acquiring target input information according to the actual operating parameters of the target bearing includes: Performing continuous wavelet transform processing on the actual operating parameters of the target bearing to obtain time-frequency features of multiple different scales; The target input information is obtained according to the time-frequency features of multiple different scales.

8. A bearing fault diagnosis device based on digital twin and aggregated attention visual perception, characterized in that: include: An acquisition unit, used to acquire actual operating parameters of a target bearing, and acquire target input information according to the actual operating parameters of the target bearing; A diagnosis prediction unit, used to input the target input information into a fault diagnosis model to obtain fault diagnosis prediction information of the target bearing; The fault diagnosis model is obtained by training the aggregated attention visual perception model based on the actual operating parameters and simulated operating parameters of the sample bearing, as well as the fault category labels corresponding to the actual operating parameters and the simulated operating parameters; The simulation operation parameters are obtained by simulating the dynamic model in the digital twin model corresponding to the sample bearing; the aggregated attention visual perception model is constructed by fusion of the aggregated attention model and the visual perception model.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, it implements the bearing fault diagnosis method based on digital twin and aggregated attention visual perception as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the bearing fault diagnosis method based on digital twin and aggregated attention visual perception as described in any one of claims 1 to 7.

Citation Information

Cited By

  • High-speed rail bearing fault diagnosis method and system

    CN119808600A