Digital-analog combined drive rolling bearing remaining service life prediction system and method

Through the combined digital-to-analog driving method, combined with the damage-fracture coupling theory and dynamic model, the recursive feature modal decomposition algorithm and CWGAN-GP network model are used to generate and correct the full-life cycle vibration signal of rolling bearing inner ring faults, solving the traditional The problem of low accuracy of the model in the natural peeling state achieves high-precision life prediction.

CN119043720APending Publication Date: 2024-11-29SHENYANG JIANZHU UNIVERSITY
View PDF 0 Cites 5 Cited by

Patent Information

Application Number
CN202411233353.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-04
Publication Date
2024-11-29

AI Technical Summary

Technical Problem

Traditional digital twin models have low accuracy in the prediction of residual service life of rolling bearings, especially in the identification of defect size and simulation of full life cycle signals in natural peeling state.

Method used

Using a digital-to-analog combined driving method, combined with the damage-fracture coupling theory and dynamic model, the full life cycle vibration signal of rolling bearing inner ring fault is generated and corrected through the recursive feature modal decomposition algorithm and the CWGAN-GP network model, and is used to construct The gated cycle unit network performs the remaining service life prediction.

Benefits of technology

The accuracy of the full-life cycle vibration signal of rolling bearing inner ring fault is improved, and the dependence on the measured signal of the full-life life is reduced, and high-precision defect size identification and life prediction are achieved in the natural peeling state.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119043720A_ABST
    Figure CN119043720A_ABST
Patent Text Reader

Abstract

The invention provides a digital-analog combined drive rolling bearing remaining service life prediction system and method, and relates to the technical field of rolling bearing health prediction. According to the method, crack cracking characteristics are considered, and a damage variable concept is introduced to simulate an evolution process of an inner ring defect size, so that a full-life-cycle accurate vibration signal of a rolling bearing inner ring fault is obtained. The provided model only needs to monitor data in real time, and the requirement of a traditional twin model for a full-life-cycle actually-measured vibration signal under the current working condition is avoided. Meanwhile, under the condition of natural stripping, the defect edge of the rolling bearing is rough and smooth sometimes, so that the entering / exiting characteristic signals of the rolling body are inconsistent and are generally weak, and the defect size is difficult to accurately recognize. According to the method, fault features and noise of vibration signals of the rolling bearing are effectively decomposed by adopting an improved feature mode decomposition method, and high-precision identification of the defect size of the inner ring of the rolling bearing in a natural peeling state is realized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of rolling bearing health prediction, and in particular relates to a system and method for predicting the remaining service life of a rolling bearing driven by a combination of digital and analog. Background Art

[0002] As a key component of rotating machinery, rolling bearings support rotating bodies, reduce friction, and reduce energy loss. In rotating machinery and equipment, the number of rolling bearing failures accounts for 40-90% of the total number of failures of large and small equipment. Among them, cracks in the inner and outer rings are the most common failures of rolling bearings, accounting for 90% of all rolling bearing failures. Therefore, the health status of the inner and outer rings of rolling bearings is closely related to the safe operation of mechanical equipment. If rolling bearing failures can be discovered and predicted in time, it will not only improve production efficiency, but also reduce maintenance costs and ensure safe operation of the factory. Therefore, research on the prediction of the remaining service life of rolling bearings has become an urgent need.

[0003] At present, there are four major categories of remaining service life prediction methods for rolling bearings: mathematical statistics, physical models, deep learning, and digital twins. The remaining service life prediction model based on mathematical statistics can obtain better expected conclusions. This model uses the principles of probability statistics to quantify the probability distribution of the remaining service life through in-depth analysis of system operation data, but it requires a large amount of expert experience and experimental data to achieve. The remaining service life prediction method of rolling bearings based on physical models has the highest accuracy. This model uses physical rules to analyze the failure mechanism of rolling bearings to predict the remaining service life, but it needs to clarify the physical rules of rolling bearings. At the same time, the model is only applicable to the current working conditions and has poor adaptability. The remaining service life prediction method of rolling bearings based on deep learning is widely used because deep learning uses powerful universal approximation theorems to fit any functional relationship without considering the failure mechanism. However, the black box characteristics of deep learning cannot explain its decision-making process and require a large amount of training data. In actual industry, it is difficult and relatively expensive to obtain a large amount of training data for the entire life cycle of rolling bearings.

[0004] Fortunately, thanks to the development of simulation, sensors, and computers, the emergence and rapid rise of digital twin methods have provided a new solution to this challenge. Digital twin technology can accurately reflect the actual production process and the operating performance of the entire life cycle. Simulation through twin models can map the dynamic evolution behavior of physical processes. Jiang et al. proposed one of the most popular DT definitions and proposed a three-dimensional DT model, which consists of real-world entities, virtual entities, and data connections that link these two entities together. Zhao et al. used the simulation data set obtained by DT to expand the sample size and improve the accuracy of the rolling bearing life prediction results to address the small sample problem of mechanical equipment such as bearings. In order to reduce the error between the simulated signal and the measured signal, an improved CycleGAN method was proposed to achieve the mapping of simulated data in the virtual space and measured data in the physical space. Zhang et al. established a high-fidelity digital twin model and used a partial domain adaptive algorithm to transfer the diagnostic knowledge learned from the digital twin model to the measured data of the physical test bench. This method can accurately diagnose the health status of actual rolling bearings using measured data. Wang et al. used a mature dynamic model to construct a digital twin model and corrected it through the Pearson correlation coefficient. This is a model online learning that synchronizes different states of the bearing into the normal operation model to obtain normal and faulty simulation numerical data. The above model has the ability to simulate quickly, and the real-time interaction effect is significant, while maintaining good prediction accuracy. However, the above models have common problems: the traditional digital twin requires the full life measured vibration signal of the rolling bearing to identify the defect size, so that the defect size at different degradation stages can be applied to the model to simulate the full life cycle signal; the digital twin model requires the full life cycle fatigue defect size under the current working conditions. The traditional digital twin model does not take into account the defect size identification problem under the natural spalling state, resulting in large defect size errors and low accuracy of model prediction results.

[0005] The traditional twin model uses full-life measured signals to simulate full-life cycle signals, and the model accuracy requires a high amount of full-life measured signal data. Summary of the invention

[0006] In view of the shortcomings of the prior art, the present invention proposes a system and method for predicting the remaining useful life of a rolling bearing driven jointly by digital and analog, which generates a full life cycle vibration signal of the inner ring fault of the rolling bearing based on damage mechanics coupled dynamics and uses an update module to correct the vibration signal to solve the problem of low accuracy of the model prediction results.

[0007] The technical solution of the present invention is:

[0008] On the one hand, the present invention provides a rolling bearing remaining service life prediction system driven by digital and analog co-operation, which is a digital twin model, including a physical entity module, a high-fidelity model module, an overall update module, a local update module and a remaining service life prediction module;

[0009] The physical entity module is used to obtain the vibration signal of the inner ring fault of the rolling bearing monitored in real time;

[0010] The high-fidelity model module is used to characterize the failure mechanism of rolling bearings using damage-fracture coupling theory, establish a rolling bearing dynamics model with defect expansion by combining damage mechanics and dynamics, and generate vibration signals of the entire life cycle of the inner ring fault of the rolling bearing simulated at different degradation stages of the rolling bearing using the rolling bearing dynamics model;

[0011] The overall update module is used to decompose the inner ring fault vibration signal of the rolling bearing monitored in real time by using a recursive eigenmode decomposition algorithm to obtain the defect size of the inner ring of the rolling bearing, and update the high-fidelity model module by using the obtained defect size of the inner ring of the rolling bearing to obtain an updated vibration signal of the inner ring fault of the rolling bearing over the entire life cycle;

[0012] The local update module is a CWGAN-GP network model, which is used to use the updated simulated vibration signal of the rolling bearing inner ring fault as conditional information, and obtain the corrected vibration signal of the rolling bearing inner ring fault throughout the life cycle through continuous game between the generator and the discriminator;

[0013] The remaining service life prediction module is used to construct a training set using the corrected vibration signal of the entire life cycle of the rolling bearing inner ring fault to train the gated recurrent unit network GRU, and use the trained gated recurrent unit network GRU to realize the prediction of the remaining service life.

[0014] On the other hand, the present invention also provides a method for predicting the remaining service life of a rolling bearing driven by digital-analog combination, comprising the following steps:

[0015] Step 1: Simulate the dynamic evolution process of the inner ring defect of the rolling bearing, and introduce the dynamic evolution process of the inner ring defect of the rolling bearing into the dynamic model to obtain the vibration signal of the simulated inner ring fault of the rolling bearing throughout its life cycle;

[0016] Step 1.1: Construct a rolling bearing dynamics model;

[0017] The rolling bearing dynamics model is shown in formula (1):

[0018]

[0019] Where X(t) is the displacement of the rolling bearing along the X direction, is the speed of the rolling bearing along the X direction, is the acceleration of the rolling bearing along the X direction, F is the radial force, M is the mass of the rolling bearing, Q is the total stiffness of the rolling bearing, C is the total damping of the rolling bearing, and t is the time; the X direction is the radial or axial direction of the rolling bearing;

[0020] Step 1.2: Use the extended finite element method of ABAQUS to construct the inner ring fatigue spalling evolution model;

[0021] Step 1.2.1: Calculate the contact stress of the rolling bearing according to Hertz contact theory;

[0022] Step 1.2.2: Simplify the rolling element and inner ring structure of the rolling bearing into a rolling element with an equivalent radius that contacts the inner ring surface at a relative speed and rolls on a semi-infinite plane. Use ABAQUS to create a simplified component to simulate the actual state of the rolling element contacting the inner ring of the rolling bearing.

[0023] Furthermore, when creating simplified components using ABAQUS, it is assumed that the materials of the rolling bearing components are in a lossless state, and the mesh unit damage is set to 0;

[0024] Step 1.2.3: Use the pre-calculated contact stress of the rolling bearing to equalize the rolling process of the rolling element by moving along the inner ring surface of the rolling bearing to achieve the setting of the cyclic load;

[0025] Step 1.2.4: Introduce the damage variable into the linear elastic stress-strain constitutive equation, use the linear elastic stress-strain constitutive equation to simulate the mechanical properties of the material during fatigue failure, and calculate the shear stress of the mesh unit under the current cyclic load;

[0026] The linear elastic stress-strain constitutive equation with the introduction of damage variables is shown in formula (2):

[0027] σ=J(1-D)ε(2)

[0028] Where σ is the shear stress, J is the material stiffness, ε is the strain, and D is the damage variable;

[0029] Step 1.2.5: Calculate the damage evolution rate of each grid unit in the rolling contact domain using the damage evolution equation; the rolling contact domain is the rolling contact range of the rolling element of the rolling bearing;

[0030] The damage evolution equation is shown in formula (3):

[0031]

[0032] Where N is the total number of cycles of cyclic load, A is the correction factor, τ, m is the parameter determined by the fatigue crack growth rate test of bearing steel material, and Δτ is the stress amplitude;

[0033] Step 1.2.6: According to the material properties of the inner ring of the rolling bearing, the damage of the inner ring of the rolling bearing is set as the accumulation of linear elastic damage, and the damage value of each grid cell is calculated according to the damage evolution rate of each grid cell;

[0034] The calculation method of the damage value is shown in formula (4):

[0035]

[0036] in, is the damage value of the jth grid unit at the bth cycle, is the damage value of the jth grid unit at the b-1th cycle, is the damage evolution rate of the jth grid unit at the bth cycle, ΔN b is the current number of cycles;

[0037] Step 1.2.7: Use the extended finite element XFEM in ABAQUS to perform numerical simulation on the fatigue damage accumulation that increases with the number of cycles of cyclic load, and simulate the process of crack initiation and crack propagation. When the damage value of the mesh unit reaches the critical damage value, the mesh unit is considered to be failed. The number of cycles of cyclic load and the position and size of the failed mesh unit are recorded at this time, and the mesh units that are spalled due to the failed mesh units are removed. The cyclic load is continuously applied until the failed mesh units extend to the surface of the inner ring of the rolling bearing. The rolling bearing is considered to be completely failed, and the fatigue spalling evolution model is obtained.

[0038] Step 1.3: Extract the position and size of the failed mesh unit in the fatigue spalling evolution model and calculate the total deformation of the rolling element; the method for calculating the total deformation of the rolling element is:

[0039] ψ=xcosα z +ysinα z -γ-s(t)(5)

[0040] Where: ψ is the total deformation of the rolling element, s(t) is the displacement of the rolling element when it passes through the defect, x is the displacement of the rolling element in the circumferential direction when it passes through the defect, y is the displacement of the rolling element in the radial direction when it passes through the defect, and α z is the angle between the zth rolling element and the radial direction, and γ is the bearing clearance;

[0041] Step 1.4: Calculate the total deformation of the rolling element in each cycle, and introduce the total deformation of the rolling element in each cycle into the rolling bearing dynamics model. Calculate the vibration signal based on the fourth-order Runge-Kutta method, obtain the vibration signals of different degradation stages, and combine the vibration signals of different degradation stages to obtain the vibration signal of the simulated rolling bearing inner ring fault throughout the life cycle.

[0042] Step 2: Obtain the vibration signal of the inner ring fault of the rolling bearing monitored in real time;

[0043] Step 3: Decompose the vibration signal of the inner ring fault of the rolling bearing monitored in real time by using a recursive eigenmode decomposition algorithm to identify the defect size of the inner ring of the rolling bearing;

[0044] Step 3.1: Load the vibration signal of the inner ring fault of the rolling bearing monitored in real time;

[0045] Step 3.2: Initialize the number of iterations k = 1, and set the initial signal of the kth iteration to x k (t);

[0046] Further, when k=1, the initial signal is the vibration signal of the inner ring fault of the rolling bearing monitored in real time loaded in step 3.1;

[0047] Step 3.3: Initializing a finite impulse response filter bank through a Hanning window; the finite impulse response filter bank includes a plurality of finite impulse response filters;

[0048] Step 3.4: Filter the initial signal using a finite impulse response filter bank to obtain several decomposed modes;

[0049] The filtering method is:

[0050] u k (t) = x k (t)*f k (6)

[0051] Among them, u k (t) is the decomposition mode, * is the convolution operation, f k is a finite impulse response filter;

[0052] Step 3.5: Based on the initial signal x k (t), decomposition mode u k and the initial estimation period T k updating coefficients of finite impulse response filters in a finite impulse response filter bank;

[0053] Step 3.6: Calculate the correlation coefficients between the decomposed modes according to the decomposed modes and then construct a correlation matrix;

[0054] The calculation method of the correlation coefficient is:

[0055]

[0056] In the formula, CC is the correlation coefficient, u p 、u q There are two groups of modes, each of which includes several decomposed modes, u p (r) is u p The rth decomposition mode in q (r) is u q The rth decomposition mode in for u p The average value of (r), for u q (r) is the total number of average modes, R is the total number of decomposed modes;

[0057] Step 3.7: Select the decomposition modes corresponding to the two largest correlation coefficients from the correlation matrix and calculate their correlation kurtosis, and select the largest correlation kurtosis as the modal component of the recursive eigenmode decomposition algorithm;

[0058] Step 3.8: Calculate the residual signal and use it as the new initial signal:

[0059] The residual signal is:

[0060] x k+1 (t) = x k (t)-u k (t)(8)

[0061] Among them, x k+1 (t) is the residual signal;

[0062] Step 3.9: Use the stopping criteria to determine whether the stopping condition is reached according to the new initial signal. If so, output the current modal component. Otherwise, increase the current iteration number by 1 and return to step 3.4.

[0063] The stopping criteria are:

[0064]

[0065] in, is the modal component at the k-1th iteration, is the k-th modal component, is the set threshold;

[0066] Step 3.10: Use the waveform characteristics of the modal component to obtain the time interval between the rolling element of the rolling bearing entering the defect and exiting the defect, and calculate the defect size of the inner ring of the rolling bearing;

[0067]

[0068] Where, Υ is the defect size of the inner ring of the rolling bearing, T is the time interval between the rolling element of the rolling bearing entering the defect and exiting the defect, n is the rotation speed of the rolling bearing, is the inner diameter of the inner ring of the rolling bearing;

[0069] Step 3.11: Repeat steps 3.1 to 3.10 until the defect sizes of the inner ring of the rolling bearing corresponding to all the vibration signals of the inner ring faults of the rolling bearing monitored in real time are obtained;

[0070] Step 4: Use the radial basis function network RBF to update the inner ring defect size of the high-fidelity model to obtain the updated simulated vibration signal of the rolling bearing inner ring fault over the entire life cycle;

[0071] Step 4.1: Obtain a sample set of rolling bearing inner ring defect sizes in Indicates the vibration signal time of the rolling bearing's full life cycle corresponding to the acquired defect size;

[0072] Step 4.2: Use the rolling bearing inner ring defect size sample set to train the radial basis function network RBF to obtain a trained radial basis function network RBF;

[0073] Step 4.3: Input the defect size of the inner ring of the rolling bearing obtained in step 3.11 into the trained radial basis function network RBF to obtain the predicted value of the defect size of the inner ring of the rolling bearing in the future period of time;

[0074] Step 4.4: construct a rolling bearing inner ring defect size trend function using binomial fitting according to the predicted values ​​of the rolling bearing inner ring defect size at different times, and obtain the rolling bearing inner ring defect size of the entire cycle using the rolling bearing inner ring defect size trend function;

[0075] Step 4.5: Substitute the obtained full-cycle rolling bearing inner ring defect size into formula (5) and formula (1) to update the high-fidelity model, and obtain the updated simulated vibration signal of the rolling bearing inner ring fault over the full life cycle;

[0076] Step 5: Use the CWGAN-GP network model to correct the updated simulated vibration signal of the rolling bearing inner ring fault over the entire life cycle to obtain a corrected vibration signal of the rolling bearing inner ring fault over the entire life cycle;

[0077] Step 5.1: using the cosine similarity method, find the vibration signal of the inner race fault of the rolling bearing monitored in real time corresponding to the vibration signal of the simulated rolling bearing at different degradation stages in the updated vibration signal of the inner race fault of the rolling bearing over the whole life cycle;

[0078] Step 5.2: Set the parameters of the CWGAN-GP network model, use the updated simulated vibration signal of the rolling bearing inner ring fault as the condition information, and the real-time monitored vibration signal of the rolling bearing inner ring fault as the real sample, train the CWGAN-GP network model until the set training rounds are reached, and obtain the trained CWGAN-GP network model and output the corrected vibration signal of the rolling bearing inner ring fault throughout the life cycle;

[0079] Step 6: Use the corrected vibration signal of the rolling bearing inner ring fault over the entire life cycle to construct a training set to train the gated recurrent unit network GRU, and use the trained gated recurrent unit network GRU to predict the remaining service life;

[0080] Step 6.1: Use the corrected vibration signal of the rolling bearing inner race fault over the entire life cycle to construct the labeled data {xt, yt} as the training data in the training set, where x t is the vibration signal of the rolling bearing after correction, y t is the time label;

[0081] Step 6.2: Use the convolutional layer to extract the characteristics of the vibration signal of the corrected rolling bearing inner ring fault over the entire life cycle, and obtain a value of The eigenvalue vector of

[0082] Step 6.3: Use the eigenvalue vector to train the gated recurrent unit network GRU to obtain a trained gated recurrent unit network GRU;

[0083] Step 6.4: Obtain the vibration signal of the inner ring fault of the rolling bearing currently monitored in real time and input it into the trained gated recurrent unit network GRU to obtain the predicted health value, and then obtain the fitting value of the remaining service life;

[0084] The remaining useful life of a bearing is defined as the time interval between the predicted future time when the degradation state first reaches the failure threshold and the current time;

[0085] The remaining service life of the bearing is predicted by using the health curve output by the polynomial fitting prediction model. The calculation formula is as follows:

[0086]

[0087] In the formula, HI is the health value, t u To record the time, ω u and ξ are fitting parameters, U is the total number of fitting items, and u is the number of the fitting item.

[0088] Compared with the prior art, the beneficial effects of the present invention are:

[0089] (1) The rolling bearing fatigue spalling evolution model based on the damage-fracture coupling theory can better describe the evolution law of the bearing inner ring fatigue damage during the entire life cycle. On this basis, the evolution defects in the fatigue spalling evolution model are introduced into the rolling bearing dynamics model to establish a high-fidelity model of bearing inner ring spalling failure.

[0090] (2) In the case of natural spalling of the inner ring of the bearing, the real-time vibration signal of the rolling bearing is decomposed based on the recursive eigenmode decomposition algorithm to identify the size of the defect, solving the problem that the traditional digital twin model can only identify the defect size under artificial defects of the rolling bearing.

[0091] (3) The CWGAN-GP network model is used to interact with real-time information and simulation information, and the inner ring defect size throughout the life cycle is used to update the displacement excitation of the high-fidelity model, thereby realizing the dynamic "update" of the digital twin from local to global.

[0092] (4) The gated recurrent unit network (GRU) is used to track the bearing degradation status and predict the remaining service life.

[0093] In summary, the present invention considers the crack initiation characteristics and introduces the concept of damage variables to simulate the evolution of the inner ring defect size, so as to obtain the accurate vibration signal of the inner ring fault of the rolling bearing over the entire life cycle. The proposed model only requires real-time monitoring data, avoiding the traditional twin model's requirement for the measured vibration signal of the entire life cycle under the current working conditions. At the same time, under the condition of natural spalling, the edge of the rolling bearing defect is sometimes rough and sometimes smooth, which will cause the rolling element entry / exit characteristic signal to be inconsistent and usually weak, which makes it difficult to accurately identify the defect size. The present invention uses an improved characteristic mode decomposition method to effectively decompose the fault characteristics and noise of the vibration signal of the rolling bearing, so as to achieve high-precision identification of the defect size of the inner ring of the rolling bearing under the natural spalling state. BRIEF DESCRIPTION OF THE DRAWINGS

[0094] Figure 1 It is a flow chart of the method for predicting the remaining service life of a rolling bearing driven by digital-analog combination in an embodiment of the present invention;

[0095] Figure 2 This is a stress analysis diagram of fatigue failure of inner and outer rings in an embodiment of the present invention;

[0096] Figure 3 Schematic diagram of fatigue spalling evolution model in an embodiment of the present invention;

[0097] Among them, (a) is a schematic diagram of the crack initiation stage; (b) is a schematic diagram of the secondary or multiple spalling stage;

[0098] Figure 4It is a double pulse response caused by a local defect of the inner ring under an artificial defect in an embodiment of the present invention;

[0099] Figure 5 This is a vibration signal diagram of an inner ring defect under natural spalling in an embodiment of the present invention;

[0100] Figure 6 Flow chart of the recursive eigenmode decomposition algorithm in an embodiment of the present invention;

[0101] Figure 7 It is the CWGAN-GP network model in the embodiment of the present invention;

[0102] Figure 8 The architecture of the CWGAN-GP network model in the embodiment of the present invention;

[0103] Fig. 9 A comparison diagram of the full life cycle vibration signal of the inner race fault in an embodiment of the present invention;

[0104] Fig.10 A comparison diagram of a high-fidelity model simulation signal and a measured signal before updating in an embodiment of the present invention;

[0105] Fig.11 : is a spectrum comparison diagram of the high-fidelity model simulation signal and the measured signal of the bearing 2_1 in an embodiment of the present invention;

[0106] Fig.12 It is a simulation signal diagram in an embodiment of the present invention;

[0107] Fig.13 The modal components and spectrum diagram of the characteristic mode decomposition algorithm in the embodiment of the present invention;

[0108] Among them, (a) is the simulation signal modal component diagram of the characteristic mode decomposition algorithm; (b) is the spectrum diagram of the simulation signal modal component;

[0109] Fig.14 1 is a diagram showing the modal components and spectrum of the recursive eigenmode decomposition algorithm in an embodiment of the present invention;

[0110] Among them, (a) is the modal component diagram of the simulation signal of the recursive eigenmode decomposition algorithm; (b) is the frequency spectrum diagram of each modal component of the simulation signal;

[0111] Fig.15 is a defect size diagram identified by using a recursive eigenmode decomposition algorithm in an embodiment of the present invention;

[0112] Fig.16 A defect size diagram identified by VMD in an embodiment of the present invention;

[0113] Fig.17It is a simulation signal diagram of defect size simulation identified by using recursive eigenmode decomposition algorithm in an embodiment of the present invention;

[0114] Fig.18 It is a simulation signal diagram of defect size simulation using VMD in an embodiment of the present invention;

[0115] Fig.19 It is a RUL prediction diagram of the bearing 2_1 at a 100% ratio in an embodiment of the present invention. DETAILED DESCRIPTION

[0116] The present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0117] The rolling bearing remaining service life prediction system driven by digital and analog co-operation is a digital twin model. This model is based on the dynamic model and introduces the damage-fracture coupling theory to solve the problem that the accuracy of the traditional twin model depends on the amount of measured signal data over the entire life cycle. At the same time, the recursive eigenmode decomposition algorithm is used to solve the problem that traditional algorithms are difficult to accurately identify the defect size of rolling bearings in the natural spalling state.

[0118] The digital twin model includes a physical entity module, a high-fidelity model module, an overall update module, a local update module and a remaining service life prediction module;

[0119] The physical entity module is used to obtain the vibration signal of the inner ring fault of the rolling bearing monitored in real time;

[0120] In this implementation, the rolling bearing accelerated life test data set of Xi'an Jiaotong University is used;

[0121] The high-fidelity model module is used to characterize the failure mechanism of rolling bearings using damage-fracture coupling theory, establish a rolling bearing dynamics model with defect expansion by combining damage mechanics and dynamics, and generate vibration signals of the entire life cycle of the inner ring fault of the rolling bearing simulated at different degradation stages of the rolling bearing using the rolling bearing dynamics model;

[0122] The overall update module is used to decompose the inner ring fault vibration signal of the rolling bearing monitored in real time by using a recursive eigenmode decomposition algorithm to obtain the defect size of the inner ring of the rolling bearing, and to update the vibration signal of the whole life cycle of the inner ring fault of the rolling bearing simulated by the high-fidelity model module by using the obtained defect size of the inner ring of the rolling bearing, so as to obtain the updated vibration signal of the whole life cycle of the inner ring fault of the rolling bearing simulated;

[0123] The local update module is a CWGAN-GP network model, which is used to use the updated simulated vibration signal of the rolling bearing inner ring fault as conditional information, and obtain the corrected vibration signal of the rolling bearing inner ring fault throughout the life cycle through continuous game between the generator and the discriminator;

[0124] The remaining service life prediction module is used to construct a training set using the corrected vibration signal of the entire life cycle of the rolling bearing inner ring fault to train the gated recurrent unit network GRU, and use the trained gated recurrent unit network GRU to realize the prediction of the remaining service life.

[0125] The remaining service life prediction method of rolling bearing driven by digital and analog co-operation, such as Figure 1 As shown, the following steps are included:

[0126] Step 1: Analyze the fatigue failure mechanism of the inner ring of the rolling bearing, explore the influence of the cracking mode of the inner ring of the rolling bearing on the damage cracking criterion, simulate the dynamic evolution process of the inner ring defect of the rolling bearing, and introduce the dynamic evolution process of the inner ring defect of the rolling bearing into the dynamic model to obtain the vibration signal of the simulated inner ring fault of the rolling bearing throughout its life cycle;

[0127] The main modeling process of the traditional digital twin model of the dynamic response of rolling bearings under fatigue failure is as follows: (1) using the full-life measured vibration signal to identify the defect size within the full life cycle; (2) introducing the identified defect size within the full life cycle into the dynamic model, and then simulating the full-life cycle vibration signal of the rolling bearing. The accuracy of the traditional digital twin model mainly depends on the quality of the full-life measured signal data, but there is often a problem of insufficient full-life measured signal data during the service life of the bearing.

[0128] Starting from the perspective of crack initiation characteristics, the present invention analyzes the damage-fracture coupling mechanism of cracks to simulate the fatigue spalling generation and growth process of the inner ring of the rolling bearing from a healthy state to complete failure. On this basis, the inner ring fatigue spalling generation and growth characteristics are substituted into the dynamic model, thereby solving the problem of the traditional model digital twin's demand for measured signals over the entire life cycle.

[0129] According to the stress and deformation mode of the crack surface, the crack extension type is divided into the opening type (type I) crack mode, the sliding type (type II) crack mode and the tearing type (type III) crack mode. The difference in the contact position between the inner and outer rings of the rolling bearing and the rolling element (the inner surface raceway of the outer ring and the outer surface raceway of the inner ring are in contact with the rolling element respectively) leads to the fact that the stress and deformation mode of the inner and outer ring cracks are very different (the inner surface crack of the outer ring is subjected to the force in the tangential direction of the rolling element movement, and the outer surface crack of the inner ring is subjected to the force in the normal direction of the rolling element movement). Figure 2As shown. Therefore, the sliding type (Type II) cracking mode has the main contribution to the fatigue spalling of the inner ring, and the opening type (Type I) cracking mode has the main contribution to the fatigue spalling of the outer ring. At the same time, by comparing the contact stress between the inner ring and the outer ring of the rolling bearing, it is found that the stress on the inner ring of the rolling bearing is greater than that on the outer ring, and different stress magnitudes can lead to differences in the crack propagation mechanism. Considering the influence of these two factors on the bearing damage state, it is imperative to introduce reasonable damage accumulation calculation equation parameters to better fit the evolution law of fatigue spalling of the inner ring of the rolling bearing.

[0130] The full life cycle of a rolling bearing includes three stages: the bearing fatigue crack initiation stage, the bearing fatigue crack initiation to the crack extension to the raceway surface, and the bearing secondary or multiple fatigue spalling extension stage. The bearing spalling growth mechanism is described in detail below. During the operation of the bearing, microcracks form under the action of cyclic loads starting from the stress concentration point on the subsurface of the inner ring. As the load continues to act, the microcracks gradually extend to the outer surface of the raceway, eventually forming fatigue spalling visible to the naked eye. Since the initial fatigue spalling is small in size, the bearing can still withstand the cyclic load after the initial formation of fatigue spalling. As the cyclic load continues to act, the spalling size gradually grows along the rolling direction, eventually forming secondary spalling or multiple fatigue spalling extensions.

[0131] Step 1.1: Construct a rolling bearing dynamics model;

[0132] Considering that using the finite element method to build a dynamic model and perform calculations may take a lot of time. Taking the calculation of a 0.1 second vibration signal as an example, the time required using simulation software is about four hours. In order to save time and computing costs, the present invention decides to sacrifice a small amount of accuracy to a certain extent to maintain the real-time nature of the digital twin model. Therefore, MATLAB is chosen to build a rolling bearing dynamic model. This decision aims to significantly reduce time and computing costs while maintaining reasonable accuracy.

[0133] Due to the special structure of the rolling bearing (composed of inner and outer rings and rolling elements), a relatively complex nonlinear dynamic system is formed during the service movement. The present invention simplifies the rolling bearing into a mass spring damping system and establishes a rolling bearing dynamics formula, i.e., a rolling bearing dynamics model, as shown in formula (1):

[0134]

[0135] Where X(t) is the displacement of the rolling bearing along the X direction, is the speed of the rolling bearing along the X direction, is the acceleration of the rolling bearing along the X direction, F is the radial force, M is the mass of the rolling bearing, Q is the total stiffness of the rolling bearing, C is the total damping of the rolling bearing, and t is the time; the X direction is the radial or axial direction of the rolling bearing;

[0136] Step 1.2: In order to accurately simulate the fatigue spalling process of rolling bearings, the extended finite element method of ABAQUS is used to construct the inner ring fatigue spalling evolution model, such as Figure 3 As shown;

[0137] The present invention uses ABAQUS to simulate the crack propagation process and obtains the unit damage batches and positions (temporal and spatial characteristics) during the crack propagation process of the fatigue spalling evolution model. On this basis, the unit damage batches and positions during the fatigue spalling evolution process are brought into the rolling bearing dynamics model constructed by MATLAB to simulate the vibration signals at different degradation stages. That is, the quantitative defects at different degradation stages are replaced by failed units or spalling units under different damage amounts to accurately describe the vibration signals at each stage of the rolling bearing defect evolution.

[0138] In order to enable the rolling bearing dynamics model to generate accurate full-life cycle vibration signals, this paper introduces the damage-fracture theory and combines it with dynamics to achieve the construction of a high-fidelity model. The specific steps for deriving the high-fidelity model are as follows:

[0139] Step 1.2.1: Calculate the contact stress of the rolling bearing according to Hertz contact theory;

[0140] Step 1.2.2: To reduce the computational cost, the rolling element and inner ring structure of the rolling bearing is simplified to a rolling element with an equivalent radius that contacts the inner ring surface at a relative speed and rolls on a semi-infinite plane. A simplified component is created using ABAQUS to simulate the actual state of the rolling element of the rolling bearing contacting the inner ring.

[0141] Furthermore, when creating simplified parts using ABAQUS, it is assumed that the materials of the rolling bearing components are in a lossless state and the mesh unit damage is set to 0.

[0142] Step 1.2.3: Use the pre-calculated contact stress of the rolling bearing to equalize the rolling process of the rolling element by moving along the inner ring surface of the rolling bearing to achieve the setting of the cyclic load;

[0143] Step 1.2.4: Introduce the damage variable into the linear elastic stress-strain constitutive equation, use the linear elastic stress-strain constitutive equation to simulate the mechanical properties of the material during fatigue failure, and calculate the shear stress of the mesh unit under the current cyclic load;

[0144] The linear elastic stress-strain constitutive equation with the introduction of damage variables is shown in formula (2):

[0145] σ=J(1-D)ε(2)

[0146] Where σ is the shear stress, J is the material stiffness, ε is the strain, and D is the damage variable;

[0147] Step 1.2.5: Based on the analysis results of shear stress and strain under a certain number of cyclic loads, the damage evolution rate of each grid unit in the rolling contact domain is calculated using the damage evolution equation considering the crack propagation type of the inner ring of the rolling bearing; the rolling contact domain is the rolling contact range of the rolling element of the rolling bearing;

[0148] The damage evolution equation is shown in formula (3):

[0149]

[0150] Where N is the total number of cycles of cyclic load, A is the correction factor, τ, m is the parameter determined by the fatigue crack growth rate test of bearing steel material, Δτ is the stress amplitude, which is equivalent to the stress intensity factor variation range ΔK in the Paris formula;

[0151] Step 1.2.6: According to the material properties of the inner ring of the rolling bearing, the damage of the inner ring of the rolling bearing is set as the accumulation of linear elastic damage, and the damage value of each grid cell is calculated according to the damage evolution rate of each grid cell;

[0152] The calculation method of the damage value is shown in formula (4):

[0153]

[0154] in, is the damage value of the jth grid unit at the bth cycle, is the damage value of the jth grid unit at the b-1th cycle, is the damage evolution rate of the jth grid unit at the bth cycle, ΔN b is the current number of cycles;

[0155] Step 1.2.7: Use the extended finite element XFEM in ABAQUS to perform numerical simulation on the fatigue damage accumulation that increases with the number of cycles of cyclic load, and simulate the process of crack initiation and crack propagation. When the damage value of the mesh unit reaches the critical damage value, the mesh unit is considered to be failed. The number of cycles of cyclic load and the position and size of the failed mesh unit are recorded at this time, and the mesh units that are spalled due to the failed mesh units are removed. The cyclic load is continuously applied until the failed mesh units extend to the surface of the inner ring of the rolling bearing. The rolling bearing is considered to be completely failed, and the fatigue spalling evolution model is obtained.

[0156] Step 1.3: Under the action of cyclic load, the rolling element will undergo elastic deformation. Due to spalling, the elastic deformation of the rolling element decreases when it passes through the inner ring defect. The position and size of the failed mesh unit in the fatigue spalling evolution model are extracted to calculate the total deformation of the rolling element.

[0157] The method for calculating the total deformation of the rolling element is:

[0158] ψ=xcosα z +ysinα z -γ-s(t)(5)

[0159] Where: ψ is the total deformation of the rolling element, s(t) is the displacement of the rolling element when it passes through the defect, x is the displacement of the rolling element in the circumferential direction when it passes through the defect, y is the displacement of the rolling element in the radial direction when it passes through the defect, and α z is the angle between the zth rolling element and the radial direction, and γ is the bearing clearance;

[0160] Step 1.4: Calculate the total deformation of the rolling element in each cycle, and introduce the total deformation of the rolling element in each cycle into the rolling bearing dynamics model. Calculate the vibration signal based on the fourth-order Runge-Kutta method, obtain the vibration signals of different degradation stages, and combine the vibration signals of different degradation stages to obtain the vibration signal of the simulated rolling bearing inner ring fault throughout the life cycle.

[0161] Step 2: Obtain the vibration signal of the inner ring fault of the rolling bearing monitored in real time;

[0162] Step 3: Decompose the vibration signal of the inner ring fault of the rolling bearing monitored in real time by using a recursive eigenmode decomposition algorithm to identify the defect size of the inner ring of the rolling bearing;

[0163] Spalling caused by fatigue is the most common cause of rolling bearing failure, and its spalling size can well characterize the degree of failure of rolling bearings. Studies have reported that in the acceleration signal, contact stress redistribution occurs when the rolling element enters and exits the spalling zone, thereby causing a low-frequency step response. At the same time, when the rolling element exits the spalling zone, it hits the edge of the spalling zone, which often causes a high-frequency impact response. Traditional spalling size calculation methods attempt to capture the signal characteristics of the rolling element entering, hitting and exiting the defect, and use its waveform characteristics to estimate the spalling size. The present invention conducts an in-depth analysis of the waveform characteristics of the vibration signals of artificial defects and natural defects. In the case of artificial defects, the vibration signal is relatively clear, which can directly reflect the waveform characteristics when the ball enters, hits and exits the defect. In contrast, in the case of natural defects, the vibration signal fails to clearly reflect these waveform characteristics, such as Figure 4 and Figure 5 shown.

[0164] Considering the weak change and easy masking of the bearing entry / exit characteristic signal caused by the uneven defect edge in the case of natural spalling, the traditional signal decomposition algorithm is not effective in natural defect identification. The advantage of eigenmode decomposition is that it utilizes the excellent properties of correlation kurtosis to remove redundant information and lock fault information in the mode selection process, and can adaptively and accurately decompose the fault mode and show excellent robustness to other interference and noise. Appropriate decomposition parameters are the key to the eigenmode decomposition algorithm to separate vibration signals. If the modal decomposition number is overestimated, the same component signal is divided into different modal components. If the modal decomposition number is underestimated, multiple component signals will be stuffed into one modal component, which means that the multi-component signal is not completely decomposed. However, the eigenmode decomposition algorithm requires manual setting of the modal number, and it is difficult to effectively decompose the vibration signal by manually setting the modal decomposition number.

[0165] In order to overcome the above problems, the present invention refers to the recursive idea of ​​the recursive decomposition algorithm and proposes a recursive eigenmode decomposition algorithm. The recursive algorithm is used to adaptively match the optimal mode number to overcome the limitation that the eigenmode decomposition requires manual setting of the mode number, thereby realizing accurate identification of defect size under natural spalling state.

[0166] The stopping criterion is an important part of the recursive algorithm. The traditional stopping criterion is not applicable to the eigenmode decomposition algorithm, so the stopping condition of EMD is used to set the stopping criterion to realize the recursive eigenmode decomposition algorithm. This algorithm has the advantages of removing redundant component signals and mixed component signals in the eigenmode decomposition algorithm, and can also realize adaptive decomposition without presetting the number of mode decompositions.

[0167] The recursive eigenmode decomposition algorithm provides an adaptive and effective method for the decomposition of rolling bearing vibration signals. The present invention uses it for the defect size identification of rolling bearings. First, a set of FIR filters are initialized and the measured vibration signal is decomposed. Secondly, the filter group is iteratively updated with the correlation kurtosis as the objective function, and the fault cycle is estimated in each filtering process; then, by comparing the correlation coefficient of each two modes, the mode with the smaller correlation kurtosis is abandoned from the two modes with the largest correlation coefficient. Then, when the stop condition is reached, the decomposition ends. Finally, the waveform characteristics of the decomposed signal are used to calculate the defect size of the inner ring of the rolling bearing.

[0168] Specific steps are as follows Figure 6 As shown, including:

[0169] Step 3.1: Load the vibration signal of the inner ring fault of the rolling bearing monitored in real time;

[0170] Step 3.2: Initialize the number of iterations k = 1, and set the initial signal of the kth iteration to x k (t);

[0171] Further, when k=1, the initial signal is the vibration signal of the inner ring fault of the rolling bearing monitored in real time loaded in step 3.1;

[0172] Step 3.3: Initializing a finite impulse response filter bank through a Hanning window; the finite impulse response filter bank includes a plurality of finite impulse response filters;

[0173] Step 3.4: Filter the initial signal using a finite impulse response filter bank to obtain several decomposed modes;

[0174] The filtering method is:

[0175] u k (t) = x k (t)*f k (6)

[0176] Among them, u k (t) is the decomposition mode, * is the convolution operation, f k is a finite impulse response filter;

[0177] Step 3.5: Based on the initial signal x k (t), decomposition mode u k and the initial estimation period T k updating coefficients of finite impulse response filters in a finite impulse response filter bank;

[0178] Step 3.6: Calculate the correlation coefficients between the decomposed modes according to the decomposed modes and then construct a correlation matrix;

[0179] The calculation method of the correlation coefficient is:

[0180]

[0181] In the formula, CC is the correlation coefficient, u p 、u q There are two groups of modes, each of which includes several decomposed modes, u p (r) is u p The rth decomposition mode in q (r) is u q The rth decomposition mode in for u p The average value of (r), for u q (r) is the total number of average modes, R is the total number of decomposed modes;

[0182] Step 3.7: Select the decomposition modes corresponding to the two largest correlation coefficients from the correlation matrix and calculate their correlation kurtosis, and select the largest correlation kurtosis as the modal component of the recursive eigenmode decomposition algorithm;

[0183] Step 3.8: Calculate the residual signal and use it as the new initial signal:

[0184] The residual signal is:

[0185] x k+1 (t) = x k (t)-u k (t)(8)

[0186] Among them, x k+1 (t) is the residual signal;

[0187] Step 3.9: Use the stopping criteria to determine whether the stopping condition is reached according to the new initial signal. If so, output the current modal component. Otherwise, increase the current iteration number by 1 and return to step 3.4.

[0188] The stopping criteria are:

[0189]

[0190] in, is the modal component at the k-1th iteration, is the k-th modal component, is the set threshold;

[0191] In this embodiment is 0.1;

[0192] Step 3.10: Use the waveform characteristics of the modal component to obtain the time interval between the rolling element of the rolling bearing entering the defect and exiting the defect, and calculate the defect size of the inner ring of the rolling bearing;

[0193]

[0194] Where, Υ is the defect size of the inner ring of the rolling bearing, T is the time interval between the rolling element of the rolling bearing entering the defect and exiting the defect, n is the rotation speed of the rolling bearing, is the inner diameter of the inner ring of the rolling bearing;

[0195] Step 3.11: Repeat steps 3.1 to 3.10 until the defect sizes of the inner ring of the rolling bearing corresponding to all the vibration signals of the inner ring faults of the rolling bearing monitored in real time are obtained;

[0196] According to the above algorithm, the inner ring defect size under the natural peeling state is calculated using real-time monitoring data. However, the above algorithm cannot predict the future size. Therefore, the present invention introduces a radial basis function network (RBF) to construct an inner ring defect size trend function.

[0197] Step 4: Use the radial basis function network RBF to update the inner ring defect size of the high-fidelity model to obtain the updated simulated vibration signal of the rolling bearing inner ring fault over the entire life cycle;

[0198] Step 4.1: Obtain a sample set of rolling bearing inner ring defect sizes in Indicates the vibration signal time of the rolling bearing's full life cycle corresponding to the acquired defect size;

[0199] Step 4.2: Use the rolling bearing inner ring defect size sample set to train the radial basis function network RBF to obtain a trained radial basis function network RBF;

[0200] Step 4.3: Input the defect size of the inner ring of the rolling bearing obtained in step 3.11 into the trained radial basis function network RBF to obtain the predicted value of the defect size of the inner ring of the rolling bearing in the future period of time;

[0201] Step 4.4: construct a rolling bearing inner ring defect size trend function using binomial fitting according to the predicted values ​​of the rolling bearing inner ring defect size at different times, and obtain the rolling bearing inner ring defect size of the entire cycle using the rolling bearing inner ring defect size trend function;

[0202] Step 4.5: Substitute the obtained full-cycle rolling bearing inner ring defect size into formula (5) and formula (1) to update the high-fidelity model, and obtain the updated simulated vibration signal of the full life cycle of the rolling bearing inner ring fault;

[0203] The present invention uses input and output data for "training" to build the corresponding relationship between independent variables and dependent variables. The trained neural network is used to predict the evolution trend of the inner ring defect size and use the binomial to fit the inner ring defect size-time functional relationship, thereby solving the problem of missing future data of the bearing under real-time monitoring of a certain working condition.

[0204] Step 5: Use the CWGAN-GP network model to correct the updated simulated vibration signal of the rolling bearing inner ring fault over the entire life cycle to obtain a corrected vibration signal of the rolling bearing inner ring fault over the entire life cycle;

[0205] In order to build a bridge for information exchange and sharing between the high-fidelity model and the physical entity, the present invention introduces the CWGAN-GP network model to analyze the real-time monitoring data characteristics to realize the simulated vibration signal correction, and uses the real-time monitoring data to provide "data nutrients" for the optimization of the high-fidelity model, thereby achieving the local update effect of the high-fidelity model data level. The detailed process is as follows: Figure 7 shown.

[0206] In this paper, the CWGAN-GP network model uses the real-time monitoring of the vibration signal of the inner ring fault of the rolling bearing to provide "data nutrients" for the optimization of the digital twin model. The specific implementation steps for correcting the vibration signal of the inner ring fault of the rolling bearing throughout its life cycle are as follows:

[0207] Step 5.1: using the cosine similarity method, find the vibration signal of the inner race fault of the rolling bearing monitored in real time corresponding to the vibration signal of the simulated rolling bearing at different degradation stages in the updated vibration signal of the inner race fault of the rolling bearing over the whole life cycle;

[0208] The present invention constructs a more realistic vibration signal by adding auxiliary condition information (updated vibration signal of the whole life cycle of the inner ring fault of the rolling bearing) in the CWGAN-GP network model, so a similarity measurement method is needed to measure the degradation stage of the inner ring fault vibration signal monitored in real time and the vibration signal of the whole life cycle of the inner ring fault of the rolling bearing simulated after the update. The cosine similarity method is introduced to calculate the similarity between the vibration signal of the whole life cycle of the inner ring fault of the rolling bearing simulated after the update and the vibration signal of the inner ring fault of the rolling bearing monitored in real time, so as to ensure that the degradation stage represented by the vibration signal of the inner ring fault of the rolling bearing monitored in real time corresponds to the degradation stage of the vibration signal of the whole life cycle of the inner ring fault of the rolling bearing simulated after the update, and the formula is as follows:

[0209]

[0210] Wherein: H is the updated simulated vibration signal of the rolling bearing inner ring fault, and Φ is the real-time monitored vibration signal of the rolling bearing inner ring fault.

[0211] Step 5.2: Set the parameters of the CWGAN-GP network model, use the updated simulated vibration signal of the rolling bearing inner ring fault as the condition information, and the real-time monitored vibration signal of the rolling bearing inner ring fault as the real sample, train the CWGAN-GP network model until the set training rounds are reached, and obtain the trained CWGAN-GP network model and output the corrected vibration signal of the rolling bearing inner ring fault throughout the life cycle;

[0212] Fixed discriminator B, training generator G. Generator G continuously generates the corrected vibration signal of the rolling bearing, inputs it into discriminator B, and judges whether the corrected vibration signal of the rolling bearing is true or false. Fixed generator G, training discriminator B. Continuously train the discriminator to improve the discrimination ability of B. During the training process, both parties optimize their respective networks through alternating training, and use the game mechanism with each other until the two parties reach a dynamic balance, use the gradient descent method to minimize the loss function, and use the back propagation algorithm to update the network parameters.

[0213] The present invention uses the vibration signal of the rolling bearing monitored in real time as the judgment signal, and generates more accurate data based on the vibration signal of the inner ring fault of the rolling bearing after the update simulation. In this way, the generated data can be made more similar to the vibration signal of the inner ring fault of the rolling bearing monitored in real time while keeping the distribution characteristics of the vibration signal of the inner ring fault of the rolling bearing after the update simulation unchanged, thereby improving the training effect of the network. At the same time, the introduction of gradient penalty can train the network more stably and reduce the difficulty of training. The loss function formula of the CWGAN-GP network model is as follows:

[0214]

[0215] In the formula, G is the generator, B is the generator, VB,G is the objective function, E[] is the mathematical expectation, λ is the gradient penalty coefficient, B() is the probability of the real sample, is the gradient of the discriminator, x is random noise, is a random interpolation between the real data and the generated data, is the real data, P g To generate data distribution, P r is the actual data distribution, is a random interpolation distribution between the real data and the generated data;

[0216] Step 6: Use the corrected vibration signal of the rolling bearing inner ring fault over the entire life cycle to construct a training set to train the gated recurrent unit network GRU, and use the trained gated recurrent unit network GRU to predict the remaining service life;

[0217] In order to achieve accurate prediction of the remaining service life of rolling bearings based on deep learning, the full life cycle data under the same working conditions must be used as the training set of the neural network model. The lack of training set data under the same working conditions may lead to poor prediction of the remaining service life of rolling bearings under real-time monitoring. In response to the above problems, the present invention uses the corrected vibration signal of the full life cycle of the inner ring fault of the rolling bearing as the training set of the gated recurrent unit network GRU. The reasons are as follows: to maximize the use of data generated by the high-fidelity model of the full life cycle of the bearing to avoid the phenomenon of waste of simulation data; the vibration signal generated by the high-fidelity model can not only characterize the key characteristics of the rolling bearing life attenuation, but also predict the future degradation trend of the real-time monitoring physical entity.

[0218] Since the gated recurrent unit network GRU is optimized based on RNN, it not only overcomes the problem of gradient vanishing in traditional RNN, but also has a simpler structure, fewer parameters, and faster convergence speed. In order to explore the correlation between the real-time monitored vibration signal and life, the present invention introduces an encoder-decoder framework, in which the bidirectional gated recurrent unit network (BiGRU) is used as the encoder and the GRU is used as the decoder. In addition, this article also introduces an attention mechanism, which can adaptively weight features according to their importance, so that the model can autonomously pay attention to important features. The remaining useful life prediction model architecture is as follows Figure 8 As shown, the specific steps are as follows:

[0219] Step 6.1: Use the corrected vibration signal of the rolling bearing inner race fault over the entire life cycle to construct the labeled data {xt, yt} as the training data in the training set, where x t is the vibration signal of the rolling bearing after correction, y t is the time label;

[0220] Step 6.2: Use the convolutional layer to extract more realistic features of the corrected full life cycle vibration signal of the inner race fault of the rolling bearing, and obtain a size of The eigenvalue vector of

[0221] Step 6.3: Use the eigenvalue vector to train the gated recurrent unit network GRU to obtain a trained gated recurrent unit network GRU;

[0222] The GRU network is used to capture useful time information, and the attention mechanism is used to measure the importance of the output of each time step to the prediction result to focus on more important information features, thereby achieving efficient fitting of the regression relationship between vibration signals and remaining service life. The network is trained using a stochastic gradient descent optimizer to reduce the network training error and obtain the optimal model parameters.

[0223] Since the gated recurrent unit network GRU is a supervised model, yt ∈[0,1] is the remaining useful life label corresponding to time t. The mean square error is used as the loss function of the model, and the formula is as follows:

[0224]

[0225] Where: y s and are the true health value and the predicted health value, s is the number of the health value, S is the total number of health values, and L is the loss value;

[0226] Step 6.4: Obtain the vibration signal of the inner ring fault of the rolling bearing currently monitored in real time and input it into the trained gated recurrent unit network GRU to obtain the predicted health value, and then obtain the fitting value of the remaining service life;

[0227] The remaining useful life of a bearing can be defined as the time interval between the predicted future time when the degradation state first reaches the failure threshold and the current time;

[0228] The remaining service life of the bearing can be predicted by using the health curve output by the polynomial fitting prediction model. The calculation formula is as follows:

[0229]

[0230] In the formula, HI is the health value, t u To record the time, ω u and ξ are fitting parameters, U is the total number of fitting items, and u is the number of the fitting item.

[0231] Experimental verification:

[0232] Dataset introduction:

[0233] The validation data set for this paper comes from the XJTU-SY accelerated life test data set. The accelerated life test platform used in the rolling bearing accelerated life test conducted by Xi'an Jiaotong University was designed by the Joint Laboratory and manufactured by Suntech.

[0234] The test bearing is a LDKUER204 rolling bearing, and the main structural parameters of the rolling bearing are shown in Table 1. In order to obtain the vibration signal of the bearing throughout its life cycle, two PCB352C33 unidirectional acceleration sensors measure the horizontal and vertical directions of the bearing respectively. The DT9837 portable dynamic signal collector is used to collect vibration signals in the test. The sampling frequency is set to 25.6kHz, the sampling interval is 1min, and the sampling time is 1.28s each time. Under the working conditions shown in Table 2, an accelerated life test was carried out until complete failure to obtain a data set. This paper uses the Bearing2_1 data set under working condition 2 to verify the accuracy of the digital twin model.

[0235] Table 1 LDKUER204 bearing parameters

[0236]

[0237] Table 2 Bearing accelerated life test conditions

[0238]

[0239] High-fidelity model comparison and verification:

[0240] This paper refers to the parameters of rolling bearing condition 2 (load 11KN, speed 2250r / min) of Xi'an Jiaotong University and sets them as comparison model condition parameters and high-fidelity model condition parameters. Fig. 9 It was found that the method used by Deng et al. in a calibration-based hybrid transfer learning framework for RUL prediction of rolling bearing across different machines can simulate better vibration signals, but the method relies on a sufficient amount of full life cycle measured vibration signals as basic data to achieve better results.

[0241] The model proposed in this paper substitutes the dynamic evolution process of the inner ring defect of the rolling bearing into the dynamic model, and uses the fourth-order Runge-Kutta method of MATLAB to solve the dynamic response of different degradation stages, and obtains the full life cycle acceleration response simulated under working condition number 2, where Fig.10 This is a comparison chart of the high-fidelity model simulation vibration signal and the measured vibration signal before the update. By comparing the overall amplitude trend of the vibration signal, it is found that the overall trend of the vibration signal generated by the twin model before the update is basically consistent with the real vibration signal. To further verify the accuracy of the model, the frequency spectrum of the twin sample and the real sample is calculated by Fourier transform, as shown in Figure 1. Fig.11 As shown. It is worth noting that the frequency values ​​of the twin samples are very close to those of the real samples, which is consistent with the distribution characteristics of the real samples. In summary, the digital twin model proposed in this paper can effectively generate bearing life cycle vibration data with the same change trend fault characteristics as the actual data.

[0242] By comparison Fig. 9 , 10It is found that the model proposed in this paper can achieve similar results to the A calibration-based hybrid transfer learning framework for RUL prediction of rolling bearing across different machines model without the help of measured vibration signals throughout the life cycle. However, the traditional digital twin model can only obtain the defect size through measured vibration signals, resulting in its high dependence on measured data. The high-fidelity model proposed in this paper can accurately simulate the defect size, solving the dependence of the traditional digital twin model on the amount of data.

[0243] However, further analysis Fig.11 It was found that there was a large error between the simulated vibration signal and the actual vibration acceleration signal, and the amplitude of the simulated vibration signal in the early stage was significantly different from the real signal. Therefore, the high-fidelity model constructed by combining damage-fracture theory with dynamics alone cannot accurately simulate the vibration signal and needs further optimization and updating.

[0244] Update module comparison and verification

[0245] Simulation signal comparison verification:

[0246] References Simulated rolling bearing vibration signal FeatureModeDecomposition: NewDecomposition TheoryforRotatingMachineryFaultDiagnosis:

[0247] The sampling rate of the simulation signal is 20kHz, the time is 1s, and the specific parameters are shown in Table 3.

[0248] Table 3 Simulation signal parameters

[0249]

[0250] Based on the above parameters, a simulation signal is constructed, and the signal-to-noise ratio of the simulation signal is set to -10dB to simulate the noise environment that may be encountered in the actual environment. At a signal-to-noise ratio of -10dB, the useful information in the signal is often drowned out by the noise, making the signal analysis more challenging. Fig.12 As shown in the figure, it is difficult to find periodic pulses from the simulation signal. Due to the influence of noise, the characteristics of periodic pulses become less obvious or even completely obscured. Therefore, a recursive eigenmode decomposition algorithm is used to accurately extract useful information from the signal.

[0251] The eigenmode decomposition algorithm is applied to the simulation signal. Fig.13In (a), the modal components based on the eigenmode decomposition algorithm are shown. The characteristic information of periodic impact signals and random impacts can be clearly found in the modal components, which reflect the different frequency components and characteristics in the signal. Fig.13 As shown in (b), there are obvious spectrum peaks at the first ten harmonics, which enables the fault features of the periodic impulse signal to be clearly extracted. These significant spectrum peaks verify the effectiveness of the eigenmode decomposition algorithm in fault detection and feature extraction.

[0252] The recursive eigenmode decomposition algorithm shows significant advantages in the decomposition of composite signals. Fig.14 As shown in (a), the characteristic information of periodic shock signals and random shocks can be clearly represented in RIMF. Fig.14 As shown in (b), there are obvious peaks in the spectrum, which can accurately extract the fault characteristics of the periodic impulse signal. Figure 13-14 , it can be seen that the RIMF and spectrum based on the recursive eigenmode decomposition algorithm are basically the same as the IMF and spectrum based on the eigenmode decomposition algorithm. This shows that in terms of signal decomposition, the recursive eigenmode decomposition algorithm constructed in this paper has basically the same superiority as the eigenmode decomposition algorithm. In terms of adaptability, the recursive eigenmode decomposition algorithm can adaptively select the optimal modal decomposition number K to accurately decompose the vibration signal, while the modal decomposition algorithm requires manual continuous testing to screen out the optimal modal decomposition number K.

[0253] In summary, the recursive eigenmode decomposition algorithm proposed in this paper has good performance in feature extraction. At the same time, in terms of the selection of the modal decomposition number K, the recursive eigenmode decomposition algorithm proposed in this paper is better than the eigenmode decomposition algorithm. The recursive eigenmode decomposition algorithm not only performs well in extracting signal features, but also shows reliability when processing complex signals.

[0254] Defect size comparison verification:

[0255] In order to realize the interaction between the physical entity of the model and the twin model, it is necessary to use the recursive eigenmode decomposition algorithm to calculate the defect size under the natural spalling state. It is worth noting that only when there is a defect on the raceway surface of the rolling bearing will the vibration signal show an obvious double pulse phenomenon, and the recursive eigenmode decomposition algorithm can calculate the defect size. Analysis shows that in the bearing 2_1 data set, the defect size calculated suddenly changes from the middle and late stages, indicating that obvious local defects have occurred in the raceway at this time. This paper uses the recursive eigenmode decomposition algorithm to calculate the defect size at different degradation stages. The results are as follows: Fig.15 At the same time, this article uses the classic algorithm VMD for effect comparison, as shown in Fig.16As shown in Figure 2, the XJTU bearing data set does not publish the defect sizes at different degradation stages. Therefore, this paper substitutes the defect sizes obtained by different algorithms into the dynamic model and compares the vibration signal errors simulated by each model to verify the accuracy of the proposed algorithm. Figure 17-18 As shown. Figure 17-19 It is found that the curve solved by the recursive eigenmode decomposition algorithm is closer to the time domain characteristic curve and amplitude of the measured vibration signal, while the time domain characteristic curve and amplitude of the twin sample simulated based on VMD have large errors compared with the time domain characteristic curve and amplitude of the measured vibration signal.

[0256] RUL prediction and discussion:

[0257] This paper introduces the prediction root mean square error (RMSE) to evaluate the performance of the model prediction method. RMSE can be calculated according to the following formula.

[0258]

[0259] Where: is the predicted value, R is the true value, is the number of predictions;

[0260] In order to evaluate the performance of the proposed digital twin method in actual industrial scenarios, the XJTU-SY accelerated life test dataset is used to predict the RUL value using the Bearing2_1 data, as shown in Table 4. The main reason why the RMSE value is relatively small is that this paper uses polynomial fitting to reduce the volatility of the life prediction curve;

[0261] Table 4 RUL prediction comparison

[0262]

[0263] This paper uses the polynomial method to fit the predicted health index-life value curve. Fig.19 The prediction results show that the error between the predicted health index (HI) value and the actual value is small, and the predicted health index-life value curve is consistent with the actual health index-life value curve. In general, it is a feasible method to use the simulated data of the proposed digital twin model as a pre-training set to train the neural network, and the neural network can obtain relatively good prediction values, which proves the advancement and accuracy of the digital twin model.

Claims

1. A rolling bearing remaining service life prediction system driven by digital and analog co-operation, characterized in that: The system is a digital twin model, including a physical entity module, a high-fidelity model module, an overall update module, a local update module, and a remaining service life prediction module; The physical entity module is used to obtain the vibration signal of the inner ring fault of the rolling bearing monitored in real time; The high-fidelity model module is used to characterize the failure mechanism of rolling bearings using damage-fracture coupling theory, establish a rolling bearing dynamics model with defect expansion by combining damage mechanics and dynamics, and generate vibration signals of the entire life cycle of the inner ring fault of the rolling bearing simulated at different degradation stages of the rolling bearing using the rolling bearing dynamics model; The overall update module is used to decompose the inner ring fault vibration signal of the rolling bearing monitored in real time by using a recursive eigenmode decomposition algorithm to obtain the defect size of the inner ring of the rolling bearing, and update the high-fidelity model module by using the obtained defect size of the inner ring of the rolling bearing to obtain an updated vibration signal of the inner ring fault of the rolling bearing over the entire life cycle; The local update module is a CWGAN-GP network model, which is used to use the updated simulated vibration signal of the rolling bearing inner ring fault as conditional information, and obtain the corrected vibration signal of the rolling bearing inner ring fault throughout the life cycle through continuous game between the generator and the discriminator; The remaining service life prediction module is used to construct a training set using the corrected vibration signal of the entire life cycle of the rolling bearing inner ring fault to train the gated recurrent unit network GRU, and use the trained gated recurrent unit network GRU to realize the prediction of the remaining service life.

2. A method for predicting the remaining useful life of a rolling bearing driven by digital-analog combination, implemented by a rolling bearing remaining useful life prediction system driven by digital-analog combination according to claim 1, characterized in that: The steps include: Step 1: Simulate the dynamic evolution process of the inner ring defect of the rolling bearing, and introduce the dynamic evolution process of the inner ring defect of the rolling bearing into the dynamic model to obtain the vibration signal of the simulated inner ring fault of the rolling bearing throughout its life cycle; Step 2: Obtain the vibration signal of the inner ring fault of the rolling bearing monitored in real time; Step 3: Decompose the vibration signal of the inner ring fault of the rolling bearing monitored in real time by using a recursive eigenmode decomposition algorithm to identify the defect size of the inner ring of the rolling bearing; Step 4: Use the radial basis function network RBF to update the inner ring defect size of the high-fidelity model to obtain the updated vibration signal of the entire life cycle of the inner ring fault of the rolling bearing; Step 5: Use the CWGAN-GP network model to correct the updated simulated vibration signal of the rolling bearing inner ring fault over the entire life cycle to obtain a corrected vibration signal of the rolling bearing inner ring fault over the entire life cycle; Step 6: Use the corrected vibration signal of the rolling bearing inner ring fault throughout its life cycle to construct a training set to train the gated recurrent unit network GRU, and use the trained gated recurrent unit network GRU to predict the remaining service life.

3. The method for predicting the remaining service life of a rolling bearing driven by digital-analog combination according to claim 2 is characterized in that: The step 1 specifically includes: Step 1.1: Construct a rolling bearing dynamics model; The rolling bearing dynamics model is shown in formula (1): Where X(t) is the displacement of the rolling bearing along the X direction, is the speed of the rolling bearing along the X direction, is the acceleration of the rolling bearing along the X direction, F is the radial force, M is the mass of the rolling bearing, Q is the total stiffness of the rolling bearing, C is the total damping of the rolling bearing, and t is the time; the X direction is the radial or axial direction of the rolling bearing; Step 1.2: Use the extended finite element method of ABAQUS to construct the inner ring fatigue spalling evolution model; Step 1.3: Extract the location and size of the failed mesh element in the fatigue spalling evolution model and calculate the total deformation of the rolling element; The method for calculating the total deformation of the rolling element is: ψ=xcosα z +ysina z -γ-s(t) (5) Where: ψ is the total deformation of the rolling element, s(t) is the displacement of the rolling element when it passes through the defect, x is the displacement of the rolling element in the circumferential direction when it passes through the defect, y is the displacement of the rolling element in the radial direction when it passes through the defect, and α z is the angle between the zth rolling element and the radial direction, and γ is the bearing clearance; Step 1.4: Calculate the total deformation of the rolling element in each cycle, and introduce the total deformation of the rolling element in each cycle into the rolling bearing dynamics model. Calculate the vibration signal based on the fourth-order Runge-Kutta method, obtain the vibration signals of different degradation stages, and combine the vibration signals of different degradation stages to obtain the vibration signal of the simulated rolling bearing inner ring fault throughout the life cycle.

4. The method for predicting the remaining service life of a rolling bearing driven by digital-analog combination according to claim 3 is characterized in that: The step 1.2 specifically includes: Step 1.2.1: Calculate the contact stress of the rolling bearing according to Hertz contact theory; Step 1.2.2: Simplify the rolling element and inner ring structure of the rolling bearing into a rolling element with an equivalent radius that contacts the inner ring surface at a relative speed and rolls on a semi-infinite plane. Use ABAQUS to create a simplified component to simulate the actual state of the rolling element contacting the inner ring of the rolling bearing. Furthermore, when creating simplified parts using ABAQUS, it is assumed that the materials of each component of the rolling bearing are in a lossless state, and the mesh unit damage is set to 0; Step 1.2.3: Use the pre-calculated contact stress of the rolling bearing to equalize the rolling process of the rolling element by moving along the inner ring surface of the rolling bearing to achieve the setting of the cyclic load; Step 1.2.4: Introduce the damage variable into the linear elastic stress-strain constitutive equation, use the linear elastic stress-strain constitutive equation to simulate the mechanical properties of the material during fatigue failure, and calculate the shear stress of the mesh unit under the current cyclic load; The linear elastic stress-strain constitutive equation with the introduction of damage variables is shown in formula (2): σ=J(1-D)ε (2) Where σ is shear stress, J is material stiffness, ε is strain, and D is damage variable; Step 1.2.5: Calculate the damage evolution rate of each grid unit in the rolling contact domain using the damage evolution equation; the rolling contact domain is the rolling contact range of the rolling element of the rolling bearing; The damage evolution equation is shown in formula (3): Where N is the total number of cycles of cyclic load, A is the correction factor, τ, m is the parameter determined by the fatigue crack growth rate test of bearing steel material, and Δτ is the stress amplitude; Step 1.2.6: According to the material properties of the inner ring of the rolling bearing, the damage of the inner ring of the rolling bearing is set as the accumulation of linear elastic damage, and the damage value of each grid cell is calculated according to the damage evolution rate of each grid cell; The calculation method of the damage value is shown in formula (4): in, is the damage value of the jth grid unit at the bth cycle, is the damage value of the jth grid unit at the b-1th cycle, is the damage evolution rate of the jth grid unit at the bth cycle, ΔN b is the current number of cycles; Step 1.2.7: Use the extended finite element XFEM in ABAQUS to perform numerical simulation on the fatigue damage accumulation that increases with the number of cycles of cyclic load, and simulate the process of crack initiation and crack propagation. When the damage value of the mesh unit reaches the critical damage value, the mesh unit is considered to have failed. The number of cycles of the cyclic load and the position and size of the failed mesh unit are recorded at this time, and the mesh units that have been peeled off due to the failed mesh units are removed. The cyclic load is continuously applied until the failed mesh units extend to the surface of the inner ring of the rolling bearing. The rolling bearing is considered to have completely failed, and a fatigue spalling evolution model is obtained.

5. The method for predicting the remaining service life of a rolling bearing driven by digital-analog combination according to claim 1 is characterized in that: The step 3 specifically includes: Step 3.1: Load the vibration signal of the inner ring fault of the rolling bearing monitored in real time; Step 3.2: Initialize the number of iterations k = 1, and set the initial signal of the kth iteration to x k (t); Further, when k=1, the initial signal is the vibration signal of the inner ring fault of the rolling bearing monitored in real time loaded in step 3.1; Step 3.3: Initializing a finite impulse response filter bank through a Hanning window; the finite impulse response filter bank includes a plurality of finite impulse response filters; Step 3.4: Filter the initial signal using a finite impulse response filter bank to obtain several decomposed modes; The filtering method is: u k (t)=x k (t)*f k (6) Among them, u k (t) is the decomposition mode, * is the convolution operation, f k is a finite impulse response filter; Step 3.5: Based on the initial signal x k (t), decomposition mode u k and the initial estimation period T k updating coefficients of finite impulse response filters in a finite impulse response filter bank; Step 3.6: Calculate the correlation coefficients between the decomposed modes according to the decomposed modes and then construct a correlation matrix; The calculation method of the correlation coefficient is: In the formula, CC is the correlation coefficient, u p 、u q There are two groups of modes, each of which includes several decomposed modes, u p (r) is u p The rth decomposition mode in q (r) is u q The rth decomposition mode in for u p The average value of (r), for u q (r) is the total number of average modes, R is the total number of decomposed modes; Step 3.7: Select the decomposition modes corresponding to the two largest correlation coefficients from the correlation matrix and calculate their correlation kurtosis, and select the largest correlation kurtosis as the modal component of the recursive eigenmode decomposition algorithm; Step 3.8: Calculate the residual signal and use it as the new initial signal: The residual signal is: x k+1 (t)=x k (t)-u k (t) (8) Among them, x k+1 (t) is the residual signal; Step 3.9: Use the stopping criteria to determine whether the stopping condition is reached according to the new initial signal. If so, output the current modal component. Otherwise, increase the current iteration number by 1 and return to step 3.

4. The stopping criteria are: in, is the modal component at the k-1th iteration, is the k-th modal component, is the set threshold; Step 3.10: Use the waveform characteristics of the modal component to obtain the time interval between the rolling element of the rolling bearing entering the defect and exiting the defect, and calculate the defect size of the inner ring of the rolling bearing; Where, Υ is the defect size of the inner ring of the rolling bearing, T is the time interval between the rolling element of the rolling bearing entering the defect and exiting the defect, n is the rotation speed of the rolling bearing, is the inner diameter of the inner ring of the rolling bearing; Step 3.11: Repeat steps 3.1 to 3.10 until the defect sizes of the inner ring of the rolling bearing corresponding to the vibration signals of all rolling bearing inner ring faults monitored in real time are obtained.

6. The method for predicting the remaining service life of a rolling bearing driven by digital-analog combination according to claim 1 is characterized in that: The step 4 specifically includes: Step 4.1: Obtain a sample set of rolling bearing inner ring defect sizes in Indicates the vibration signal time of the rolling bearing's full life cycle corresponding to the acquired defect size; Step 4.2: Use the rolling bearing inner ring defect size sample set to train the radial basis function network RBF to obtain a trained radial basis function network RBF; Step 4.3: Input the defect size of the inner ring of the rolling bearing obtained in step 3.11 into the trained radial basis function network RBF to obtain the predicted value of the defect size of the inner ring of the rolling bearing in the future period of time; Step 4.4: construct a rolling bearing inner ring defect size trend function using binomial fitting according to the predicted values ​​of the rolling bearing inner ring defect size at different times, and obtain the rolling bearing inner ring defect size of the entire cycle using the rolling bearing inner ring defect size trend function; Step 4.5: Substitute the obtained full-cycle rolling bearing inner ring defect size into formula (5) and formula (1) to update the high-fidelity model and obtain the updated simulated vibration signal of the rolling bearing inner ring fault over the entire life cycle.

7. The method for predicting the remaining service life of a rolling bearing driven by digital-analog combination according to claim 1 is characterized in that: The step 5 specifically includes: Step 5.1: using the cosine similarity method, find the vibration signal of the inner race fault of the rolling bearing monitored in real time corresponding to the vibration signal of the simulated rolling bearing at different degradation stages in the updated vibration signal of the inner race fault of the rolling bearing over the whole life cycle; Step 5.2: Set the parameters of the CWGAN-GP network model, use the updated simulated vibration signal of the rolling bearing inner ring fault as the conditional information, and the real-time monitored vibration signal of the rolling bearing inner ring fault as the real sample, train the CWGAN-GP network model until the set training rounds are reached, and obtain the trained CWGAN-GP network model and output the corrected vibration signal of the rolling bearing inner ring fault throughout the life cycle.

8. The method for predicting the remaining service life of a rolling bearing driven by digital-analog combination according to claim 1 is characterized in that: The step 6 specifically includes: Step 6.1: Use the corrected vibration signal of the rolling bearing inner ring fault over the entire life cycle to construct a labeled data set {x t ,y t } as the training data in the training set, where x t is the vibration signal of the rolling bearing after correction, y t is the time label; Step 6.2: Use the convolutional layer to extract the characteristics of the vibration signal of the corrected rolling bearing inner ring fault over the entire life cycle, and obtain a value of The eigenvalue vector of Step 6.3: Use the eigenvalue vector to train the gated recurrent unit network GRU to obtain a trained gated recurrent unit network GRU; Step 6.4: Obtain the vibration signal of the inner ring fault of the rolling bearing currently monitored in real time and input it into the trained gated recurrent unit network GRU to obtain the predicted health value, and then obtain the fitting value of the remaining service life; The remaining useful life of a bearing is defined as the time interval between the predicted future time when the degradation state first reaches the failure threshold and the current time; The remaining service life of the bearing is predicted by using the health curve output by the polynomial fitting prediction model. The calculation formula is as follows: In the formula, HI is the health value, t u To record the time, ω u and ξ are fitting parameters, U is the total number of fitting items, and u is the number of the fitting item.

Citation Information

Cited By

  • Conical floating sleeve drill bearing analysis method based on cloud computing

    CN120892767A

  • Rolling bearing fault diagnosis method based on mechanism and data fusion

    CN120951100A

  • Rolling bearing fault diagnosis method based on mechanism and data fusion

    CN120951100B

  • Physical mechanism-based part estimated life model construction method and system

    CN122286205A

  • A physical information driven rolling bearing residual life prediction method and device

    CN122819519A