Small sample rolling bearing fault diagnosis method driven by mechanism and data fusion

By combining dynamic mechanism and deep convolution to generate an adversarial network, data that conforms to physical laws is generated, which solves the problem of unstable data quality in the fault diagnosis of small sample rolling bearings, and improves the accuracy and accuracy of fault diagnosis.

CN118886291APending Publication Date: 2024-11-01BEIJING UNIV OF TECH
View PDF 0 Cites 6 Cited by

Patent Information

Application Number
CN202410113797.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-01-26
Publication Date
2024-11-01

AI Technical Summary

Technical Problem

Under small sample conditions, it is difficult to establish accuracy in digital twin models based on dynamic mechanisms, resulting in differences in the distribution of simulation data and real data. The generation adversarial network uses random noise inputs to cause data quality to be unstable, affecting the accuracy of rolling bearing fault diagnosis.

Method used

Build a digital twin model based on dynamic mechanism to obtain simulation data, combine deep convolution with conditional information constraints to generate adversarial networks, use simulation signals instead of random noise to generate data that conforms to physical laws, and establish a fault diagnosis model through convolutional neural network to perform Data enhancement to improve diagnostic accuracy.

Benefits of technology

The accuracy of rolling bearing fault diagnosis under small sample conditions is improved, and the generated data is more in line with physical characteristics, reducing the error and inaccuracy of diagnostic results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118886291A_ABST
    Figure CN118886291A_ABST
Patent Text Reader

Abstract

The invention discloses a mechanism and data fusion driven small sample rolling bearing fault diagnosis method, which comprises the steps of fault simulation data acquisition: constructing a rolling bearing digital twin model based on a dynamics mechanism to acquire simulation data under a set fault working condition; data generation model construction: constructing a generator model and a discriminator model of a deep convolutional generative adversarial network structure under the constraint of conditional information; small sample data enhancement: using a simulation signal obtained by the digital twin model to replace random noise to serve as input of a generator model so as to generate data conforming to a physical mechanism, and using the data to expand small sample data; building a fault diagnosis model: building the fault diagnosis model based on the convolutional neural network; and fault diagnosis based on data enhancement: taking the enhanced data as a training sample of a fault diagnosis model, so that the fault diagnosis model can fully learn fault feature information contained in the training sample, thereby improving the rolling bearing fault diagnosis precision under a small sample condition.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of engineering machinery fault diagnosis and health management, and specifically relates to a small sample rolling bearing fault diagnosis method driven by mechanism and data fusion. Background Art

[0002] Rotating machinery is one of the most widely used equipment in modern industry, and effective fault diagnosis technology is needed to ensure its reliability and stable operation. As the core component of rotating machinery, the failure of rolling bearings directly affects the performance of the equipment. Therefore, it is of great significance to diagnose rolling bearing faults. With the development of sensor technology and big data mining, data-driven fault diagnosis methods have been widely used in industrial equipment. However, data-driven methods are often implemented on the basis of sufficient data, and the performance of the model is greatly affected by the amount of data in the training process. However, in actual industrial scenarios, it is difficult to collect sufficient and available data (which can characterize the internal system state of the physical object). At present, data enhancement is usually performed using simulation data based on mechanism models or generated data based on data-driven models, but this method of small sample data enhancement still has the limitation of low data quality.

[0003] Since the mechanism model can effectively reflect the physical characteristics inside the physical object system, it is possible to use simulation data to supplement the lack of real data. For example, by using a fault diagnosis network based on finite element model simulation, a large amount of fault data under different types and conditions can be obtained to make up for the lack of fault samples, so as to improve the accuracy of the diagnosis model. As a high-fidelity mapping model of physical entities, digital twins have carried out a lot of research and application exploration in the field of fault diagnosis and prediction. For example, by constructing a digital twin model containing multiple attributes (geometry-physics-behavior and rules), fault data under different working conditions can be simulated.

[0004] Unlike the simulation data obtained by simulating the mechanism model, data generation usually uses generative adversarial networks (GANs) and their improved methods to generate fake data with similar distribution to real data. However, the original GANs model may not be able to obtain high-quality generated data, and many scholars use its improved methods to improve model performance. For example, by combining convolutional neural networks (CNNs) with GANs, a deep convolutional generative adversarial network is constructed to achieve high-quality expansion of the original data. In addition, some scholars use the conditional generative adversarial network architecture to add fault label information to the input samples of the original GANs model to improve the quality of generated data. At the same time, more improved models are used to generate fault data to make up for the lack of data volume and improve the quality of generated data, such as WG-CNNs, WGANs, integrated GANs, etc. In general, GANs and their improved models have strong applicability in the field of small sample fault diagnosis.

[0005] However, both the simulated data based on the mechanism model and the generated data based on GANs and its derivative models have certain differences from the distribution of real data. Therefore, when it comes to the research on rolling bearing fault diagnosis with small samples, the following problems mainly exist.

[0006] (1) In the process of building a digital twin model based on dynamic mechanisms, it is often necessary to simplify and approximate the model, making it difficult to establish an accurate digital twin model. This results in differences in the distribution of simulation data and real data, which will lead to errors and inaccuracies in the diagnostic results of the rolling bearing fault diagnosis model.

[0007] (2) For generative adversarial networks, random noise is used as the input of the generator model. However, since random noise has no physical meaning, using random noise as the input of the generator model may lead to unstable quality of generated data and easily generate data that does not meet expectations. Summary of the invention

[0008] To solve the above problems, the present invention discloses a small sample rolling bearing fault diagnosis method driven by mechanism and data fusion to overcome the problem that it is difficult to accurately identify fault modes under insufficient fault data. The method mainly includes: step 1, fault simulation data acquisition: constructing a rolling bearing digital twin model based on dynamic mechanism to obtain simulation data under set fault conditions; step 2, data generation model construction: constructing a generator model and a discriminator model of a deep convolutional generative adversarial network structure under conditional information constraints; step 3, small sample data enhancement: replacing random noise with the simulation signal obtained by the digital twin model as the input of the generator model to generate data that conforms to the physical mechanism and is used to expand the small sample data; step 4, fault diagnosis model construction: establishing a fault diagnosis model based on a convolutional neural network; step 5, fault diagnosis based on data enhancement: based on the enhanced data as the training sample of the fault diagnosis model, the fault diagnosis model can fully learn the fault feature information contained in the training sample, thereby improving the rolling bearing fault diagnosis accuracy under small sample conditions.

[0009] The present invention is implemented by the following technical solution: a small sample rolling bearing fault diagnosis method driven by mechanism and data fusion, comprising the following steps:

[0010] Step 1, fault simulation data acquisition: construct a rolling bearing digital twin model based on dynamic mechanism to obtain simulation data under set fault conditions;

[0011] Step 2: Data generation model construction: construct the generator model and discriminator model of the deep convolutional generative adversarial network structure under conditional information constraints;

[0012] Step 3, small sample data enhancement: The simulated signal obtained by the digital twin model is used as the input of the generator model instead of random noise to generate generated data that conforms to physical laws and has a similar distribution to the real signal, and the generated data is used to expand the original training samples;

[0013] Step 4, fault diagnosis model construction: Establish a fault diagnosis model based on convolutional neural network for fault pattern recognition;

[0014] Step 5, fault diagnosis based on data enhancement: Use the training samples expanded based on the generated data in step 3 to train the fault diagnosis model constructed in step 4, and use the real data under small sample conditions as test samples to complete the identification of fault modes.

[0015] Furthermore, the fault simulation data acquisition in step 1 is as follows:

[0016] Step 2.1: Establish a digital twin model based on dynamic mechanism. The differential equation is as follows:

[0017]

[0018] Among them, m1, m2, m3 are the masses of the inner ring, outer ring and unit resonator respectively; c1, c2, c3 are the damping coefficients of the inner ring, outer ring and unit resonator of the rolling bearing respectively; k1, k2, k3 are the stiffness of the inner ring, outer ring and unit resonator of the rolling bearing respectively; x1, x2, x3 are the absolute displacements of the inner ring, outer ring and unit resonator of the rolling bearing in the horizontal direction respectively; y1, y2, y3 are the absolute displacements of the inner ring, outer ring and unit resonator of the rolling bearing in the vertical direction respectively; F is the external radial load applied to the inner ring; f x and f y are the components of the nonlinear contact force of the rolling bearing in the radial horizontal and vertical directions. The detailed calculation process is as follows:

[0019]

[0020] Where i represents the number of the i-th ball, i = 1, 2, ..., N; λ i It means that when the contact elastic deformation is less than or equal to 0, the rolling element and the raceway are not in contact, there is no contact elastic deformation, and the value is 0; when the contact elastic deformation is greater than 0, the contact elastic deformation value is 1. i is the total contact deformation of the ith ball, is the relative displacement of the inner and outer rings, and the angular position θ of the ith ball i and the function of the rolling bearing radial clearance η, specifically expressed as:

[0021] δ i =(x1-x2)cos θi +(y1-y2)sinθi -η (3)

[0022] Angular position θ of the i-th ball i It can be expressed as

[0023]

[0024] Where N is the total number of balls; d t is the time interval, θ0 is the initial angular position of the cage, and ω is the angular velocity of the cage, which is calculated as follows:

[0025]

[0026] Where, d is the ball diameter; D m represents the pitch circle diameter; v is the shaft speed.

[0027] Step 2.2, fault modeling, is as follows:

[0028] The radial deformation μ of the i-th rolling element angular position caused by the damage of the inner and outer rings is:

[0029]

[0030] Among them, the damage angle position θ d =ωt+θ0, θ0 is the initial angular position of the damage; for the outer ring fixed, θ d =θ0;θ L represents the outer ring damage width; while the inner ring damage θ h The position changes continuously as the shaft rotates, so the angular position of the damage is a function of time.

[0031] For each rotation of the rolling element, it contacts the inner and outer rings of the rolling bearing once, so the radial deformation of the angular position caused by the damage of the i-th rolling element is:

[0032]

[0033] Among them, the rolling element damage angle position θ d =mod(ω s t+θ0,2π), where ω s is the angular velocity of the rolling body:

[0034]

[0035] Here, α represents the contact angle.

[0036] Taking the damage of the inner and outer rings or rolling elements of the rolling bearing into equation (3), the total deformation of the angular position of the i-th rolling element can be obtained.

[0037] δ i=(x1-x2)cos θi +(y1-y2)sin θi -η-μ (9)

[0038] Substitute equation (9) into equation (2) to calculate the total nonlinear contact force of the rolling bearing, and finally substitute it into (1) to calculate the vibration response of the rolling bearing.

[0039] Step 2.3, according to the rolling bearing digital twin model constructed in steps 2.1 and 2.2, set the simulation conditions of inner ring, outer ring and rolling element faults to obtain fault simulation data under the set fault conditions.

[0040] Furthermore, the small sample data enhancement in step 3 is as follows:

[0041] Step 3.1, replace the random noise with the fault simulation data obtained in step 2.3 as the input of the generator model in step 3.1;

[0042] Step 3.2, through the adversarial training method of the generator model and the discriminator model, generate data that conforms to the laws of physics and has a similar distribution to the real signal;

[0043] Step 3.3, generate data under the set fault condition according to steps 3.1 and 3.2, and use the generated data to expand the original training samples to achieve small sample data enhancement.

[0044] Furthermore, the fault diagnosis implementation method based on data enhancement in step 5 is as follows:

[0045] Step 4.1, fault diagnosis model training: the fault diagnosis model constructed by training the enhanced training samples obtained in step 3.3;

[0046] Step 4.2, divide the test samples: divide the real data under the small sample condition according to the fault condition as the test samples;

[0047] Step 4.3, fault pattern recognition: Based on the fault diagnosis model trained in step 4.1, perform fault pattern recognition and classification on the test samples to achieve accurate fault diagnosis based on data enhancement.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] In order to improve the similarity between fake samples and real samples, the present invention combines the digital twin model based on dynamic mechanism with the conditional deep convolutional neural network, and uses ordered simulation data instead of random noise as the input of the conditional deep convolutional neural network generator to avoid excessive randomness in the data generation process and generate data that is more in line with the physical properties of the real mechanical system. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a small sample rolling bearing fault diagnosis flow chart driven by mechanism and data fusion.

[0051] Figure 2 This is a schematic diagram of the bearing test rig at Case Western Reserve University.

[0052] Figure 3 It is the rolling bearing dynamics model and macro-geometric parameters.

[0053] Figure 4 It is a schematic diagram of the inner ring, outer ring and rolling element failure of the rolling bearing.

[0054] Figure 5 It is a schematic diagram of the generator model.

[0055] Figure 6 It is a schematic diagram of the discriminator model.

[0056] Figure 7 This is a schematic diagram of a fault diagnosis model based on a convolutional neural network.

[0057] Figure 8 It is the rolling bearing fault diagnosis result based on 100 real samples.

[0058] Fig. 9 It is the rolling bearing fault diagnosis result enhanced based on 1200 simulation samples.

[0059] Fig.10 The rolling bearing fault diagnosis results are enhanced based on 1200 generated samples. DETAILED DESCRIPTION

[0060] The present invention is described in further detail below in conjunction with the accompanying drawings.

[0061] (1)Reference Figure 1, the present invention discloses a small sample rolling bearing fault diagnosis method driven by mechanism and data fusion to overcome the problem that it is difficult to accurately identify fault modes under the condition of insufficient fault data. The method mainly includes: step 1, fault simulation data acquisition: construct a rolling bearing digital twin model based on dynamic mechanism to obtain simulation data under set fault conditions; step 2, data generation model construction: construct a generator model and a discriminator model of a deep convolutional generative adversarial network structure under conditional information constraints; step 3, small sample data enhancement: replace random noise with the simulation signal obtained by the digital twin model as the input of the generator model to generate data that conforms to the physical mechanism and is used to expand the small sample data; step 4, fault diagnosis model construction: establish a fault diagnosis model based on a convolutional neural network; step 5, fault diagnosis based on data enhancement: based on the enhanced data as the training sample of the fault diagnosis model, the fault diagnosis model can fully learn the fault feature information contained in the training sample, thereby improving the rolling bearing fault diagnosis accuracy under small sample conditions.

[0062] (2) Measured rolling bearing test data. Take the rolling bearing test data published by Case Western Reserve University (CWRU) as an example. Figure 2 As shown, the test bench consists of a motor, a torque sensor / encoder, a power meter and control electronics. Single-point faults are introduced into the test bearing by electrical discharge machining, and accelerometers are installed at the drive end and fan end of the motor housing, respectively, to collect vibration data at sampling frequencies of 12kHz and 48kHz, respectively. The present invention uses a total of 10 types of data for analysis and experiments, and selects the normal data with a sampling frequency of 12kHz and an operating condition of 0hp / 1797rpm and the drive end rolling bearing (model SKF-6205) with fault diameters of 0.1778mm (including inner ring, outer ring and rolling element), 0.3556mm (including inner ring, outer ring and rolling element) and 0.5334mm (including inner ring, outer ring and rolling element) at the sampling frequency of 12kHz and the operating condition of 0hp / 1797rpm for analysis, corresponding to fault labels 1 to 10.

[0063] (3) Data sample division. All data samples used are divided by overlapping sampling. For the data samples used in the conditional deep convolutional generative adversarial network, the sampling step is 50, and each data sample contains 500 sample points. For the data samples used for fault diagnosis, the sampling step is 30, and each data sample contains 200 sample points.

[0064] (4) Construct a digital twin model based on dynamic mechanism, Figure 3 (a) is the digital twin model of a rolling bearing with 6 degrees of freedom, and (b) is the macroscopic geometric parameters. Figure 4(a), (b) and (c) are schematic diagrams of the inner ring, outer ring and rolling element surface faults of the rolling bearing, respectively. Based on the digital twin model of the rolling bearing, the dynamic equations are solved in the MATLAB platform to obtain the vibration response under normal and fault conditions. Taking the inner ring fault of the rolling bearing as an example, the contact angle is 0°, the number of rolling elements is 9, the rolling element diameter is 7.94mm, the inner raceway diameter is 25mm, the outer raceway diameter is 52mm, the radial clearance is 0.0035mm, the motor load is 0.7457KW, and the motor speed is 1772r / min. Other parameter settings: m1, m2, m3 are 1.2638kg, 12.638kg, 1kg respectively; k1, k2, k3 are 4.241e4 N / m, 15.1056e6N / m, 8.8826e7N / m respectively; c1, c2, c3 are 1376.8Ns / m, 2210.7Ns / m, 2424.8Ns / m respectively.

[0065] (5) Design the generator model, discriminator model and fault diagnosis model. The design of the three types of models all use CNN as the model architecture. Figure 6 It is a schematic diagram of the generator model. The network structure of the generator model consists of a self-attention mechanism (BN) + CNN, which uses 2 convolutional layers and 2 transposed convolutional layers. Figure 7 is a schematic diagram of the discriminator model. The network structure of the discriminator model consists of spectral normalization (BN) + CNN, in which 2 convolutional layers and 3 transposed convolutional layers are used. Based on the constructed generator model and discriminator model, the 10 types of data in (2) are generated. Taking the inner circle fault as an example, 100 real samples and 100 simulated samples are used as training sets, and 20 real samples are used as test sets. Figure 7 This is a schematic diagram of a fault diagnosis model based on a convolutional neural network. The network structure of the fault diagnosis model is designed as a 2-layer CNN. The first layer of convolution kernels uses a 64×1 structure with a step size of 2×1. The second layer of convolution kernels uses a 32×1 structure with a step size of 2×1, and the last layer is a fully connected layer. In addition, batch normalization is used after each layer to avoid the gradient vanishing problem, and the SoftMax function is combined to achieve the output of the classification results.

[0066] (6) Design fault diagnosis experiments. The present invention sets up 3 experiments in total: Experiment 1: 100 real samples are used as training sets, and a total of 6 experiments are conducted; Experiment 2: On the basis of Experiment 1, 0, 100, 300, 500, 800 and 1200 simulated samples are additionally added as training sets; Experiment 3: On the basis of Experiment 1, 0, 100, 300, 500, 800 and 1200 generated samples are additionally added as training sets. The training set is used in the training process of the fault diagnosis model. In addition, the test set used in each experiment is 100 real samples. Finally, the fault diagnosis results of each experiment are shown in Table 1, and the results of the 6th experiment are selected for visual display: Fig.10 It is a rolling bearing fault diagnosis result based on 100 real samples; the experimental results show that the feasibility of a small sample rolling bearing fault diagnosis method driven by mechanism and data fusion disclosed by the present invention is verified.

[0067] The contents not described in detail in the specification of the present invention belong to the prior art known to the professional and technical personnel in this field.

[0068] The above description is only a preferred embodiment of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by technicians in this technical field within the essential scope of the present invention should also fall within the protection scope of the present invention.

[0069] Table 1 Comparison of diagnostic accuracy (%) using simulated data and generated data for training sample enhancement and original data

Claims

1. A small sample rolling bearing fault diagnosis method driven by mechanism and data fusion, It is characterized in that The following steps are involved: Step 1, fault simulation data acquisition: construct a rolling bearing digital twin model based on dynamic mechanism to obtain simulation data under set fault conditions; Step 2: Data generation model construction: construct the generator model and discriminator model of the deep convolutional generative adversarial network structure under conditional information constraints; Step 3, small sample data enhancement: The simulated signal obtained by the digital twin model is used as the input of the generator model instead of random noise to generate generated data that conforms to physical laws and has a similar distribution to the real signal, and the generated data is used to expand the original training samples; Step 4, fault diagnosis model construction: Establish a fault diagnosis model based on convolutional neural network for fault pattern recognition; Step 5, fault diagnosis based on data enhancement: Use the training samples expanded based on the generated data in step 3 to train the fault diagnosis model constructed in step 4, and use the real data under small sample conditions as test samples to complete the identification of fault modes.

2. According to claim 1, a small sample rolling bearing fault diagnosis method driven by mechanism and data fusion, It is characterized in that The detailed process of acquiring the fault simulation data in step 1 is as follows: Step 2.1: Establish a digital twin model based on dynamic mechanism. The differential equation is as follows: Among them, m 1 , m 2 , m 3 Respectively represent the mass of the inner ring, outer ring and unit resonator; c 1 , c 2 , c 3 The damping coefficients of the inner ring, outer ring and unit resonator of the rolling bearing respectively; k 1 , k 2 , k 3 Respectively represent the stiffness of the inner ring, outer ring and unit resonator of the rolling bearing; x 1 、x 2 、x 3 are the absolute displacements of the inner ring, outer ring and unit resonator of the rolling bearing in the horizontal direction; y1, y2, y3 are the absolute displacements of the inner ring, outer ring and unit resonator of the rolling bearing in the vertical direction; F is the external radial load applied to the inner ring; f x and f y are the components of the nonlinear contact force of the rolling bearing in the radial horizontal and vertical directions. The detailed calculation process is as follows: Where i represents the number of the i-th ball, i = 1, 2, ..., N; λ i It means that when the contact elastic deformation is less than or equal to 0, the rolling element and the raceway are not in contact, there is no contact elastic deformation, and the value is 0; when the contact elastic deformation is greater than 0, the contact elastic deformation value is 1; δ i is the total contact deformation of the ith ball, is the relative displacement of the inner and outer rings, and the angular position θ of the ith ball i and the function of the rolling bearing radial clearance η, specifically expressed as: δ i (x) 1 -x 2 )cos θi +(and 1 -and 2 )without θi -η (3) Angular position θ of the i-th ball i It can be expressed as Where N is the total number of balls; d t is the time interval, θ 0 is the initial angular position of the cage, and ω is the angular velocity of the cage, calculated as follows: Where, d is the ball diameter; D m represents the pitch circle diameter; v is the shaft speed; Step 2.2, fault modeling, is as follows: The radial deformation μ of the i-th rolling element angular position caused by the damage of the inner and outer rings is: Among them, the damage angle position θ d =ωt+θ 0 ,θ 0 is the initial angular position of the damage; for a fixed outer ring, θ d =θ 0 θ L represents the outer ring damage width; while the inner ring damage θ h The position changes continuously as the shaft rotates, so the angular position of the damage is a function of time; For each rotation of the rolling element, it contacts the inner and outer rings of the rolling bearing once, so the radial deformation of the angular position caused by the damage of the i-th rolling element is: Among them, the rolling element damage angle position θ d =mod(ω s t+θ 0 ,2π), where ω s is the angular velocity of the rolling body: Where α represents the contact angle; Taking the damage of the inner and outer rings or rolling elements of the rolling bearing into equation (3), the total deformation of the angular position of the i-th rolling element can be obtained; d i =(x 1 -x 2 )cos θi +(y 1 -y 2 )sin θi -h-m (9) Substitute equation (9) into equation (2) to calculate the total nonlinear contact force of the rolling bearing, and finally substitute it into (1) to calculate the vibration response of the rolling bearing; Step 2.3, according to the rolling bearing digital twin model constructed in steps 2.1 and 2.2, set the simulation conditions of inner ring, outer ring and rolling element faults to obtain fault simulation data under the set fault conditions.

3. According to claim 1, a small sample rolling bearing fault diagnosis method driven by mechanism and data fusion, It is characterized in that The small sample data enhancement process in step 3 is as follows: Step 3.1, replace the random noise with the fault simulation data obtained in step 2.3 as the input of the generator model in step 3.1; Step 3.2, through the adversarial training method of the generator model and the discriminator model, generate data that conforms to the laws of physics and has a similar distribution to the real signal; Step 3.3, generate data under the set fault condition according to steps 3.1 and 3.2, and use the generated data to expand the original training samples to achieve small sample data enhancement.

4. According to claim 1, a small sample rolling bearing fault diagnosis method driven by mechanism and data fusion, It is characterized in that The fault diagnosis implementation method based on data enhancement in step 5 is as follows: Step 4.1, fault diagnosis model training: the fault diagnosis model constructed by training the enhanced training samples obtained in step 3.3; Step 4.2, divide the test samples: divide the real data under the small sample condition according to the fault condition as the test samples; Step 4.3, fault pattern recognition: Based on the fault diagnosis model trained in step 4.1, perform fault pattern recognition and classification on the test samples to achieve accurate fault diagnosis based on data enhancement.

Citation Information

Cited By

  • Fatigue life prediction method and related device

    CN119227471A

  • Bearing fault diagnosis method based on dynamical model under small sample condition

    CN120163070A

  • Fault positioning test method for civil aircraft environment control system based on data and real fusion

    CN121050414A

  • Rolling bearing diagnosis system and method based on dynamic correction and neural symbol reasoning

    CN121144970A

  • Rolling bearing diagnosis system and method based on dynamic correction and neural symbolic reasoning

    CN121144970B