CN115809595A - Digital twin model construction method reflecting rolling bearing defect expansion - Google Patents

Abstract

The invention discloses a digital twin model construction method reflecting rolling bearing defect expansion, and relates to the field of bearing health management and operation and maintenance; the proposed digital twin model bridges a physical space and a virtual space by using a vibration signal of a rolling bearing, realizes dynamic update of a virtual entity by using a signal approximation method without measuring the actual defect size of the bearing, and solves the problem that the time-varying defect size of the bearing is difficult to obtain in the running state. The relation between the root mean square value of the actually measured vibration signal and the length of the local defect in the virtual space is established by using the LSTM, the physical space signal is mapped to the virtual space, the model can directly predict the size of the local defect of the actual bearing according to the collected vibration signal under the same working condition after training of historical data, and the virtual entity is helped to be rapidly updated so as to realize operation and maintenance and health monitoring of the rolling bearing.

Description

A digital twin model construction method reflecting the extension of rolling bearing defects

technical field

The invention relates to the fields of bearing health management and operation and maintenance, in particular to a method for constructing a digital twin model reflecting the expansion of rolling bearing defects.

Background technique

Rolling bearings are one of the most commonly used parts in rotating machinery equipment, and their working reliability directly affects the safety and operation stability of rotating machinery equipment, so it is very important for the operation and maintenance of rolling bearings. Many scholars have simulated the dynamic response of rolling bearings by establishing local fault dynamic models to operate and maintain large mechanical equipment containing rolling bearings. In view of the different degrees of idealization in the establishment process of the rolling bearing dynamic model, the model cannot completely restore the motion behavior of the internal components of the rolling bearing, resulting in differences between the simulated dynamic response output by the model and the real dynamic response of the rolling bearing. Therefore, how to reduce the difference between the simulated dynamic response and the real dynamic response of rolling bearings is a key issue to improve the accuracy of rolling bearing fault detection.

Digital twin technology can act as a bridge between the simulated dynamic response and the real dynamic response. Most traditional digital twin models are based on real-time sensor data, and their modeling process often requires the help of some simulation software, such as ICEM CFD, MATLAB, ANSYS, etc., or some mathematical mechanical models. They continuously update the parameters of virtual entities based on real-time sensor data to mimic the motion of physical entities and generate simulated dynamic responses. At present, this technology has been applied to the management and operation and maintenance of bearings. Farhat et al. used digital twin technology to provide data sets for the neural network of faulty bearing classification. Experiments show that the data sets generated by digital twin technology can still meet the classification accuracy requirements. In order to accurately predict the remaining service life of rolling bearings during the degradation process, Liu et al. established a digital twin model of bearings based on bearing vibration phenomena, and used Domain Adversarial Neural Network (DANN) to achieve domain adaptation between simulation data and real data. Based on the dynamic model and the historical measurement data of the bearing, Qin et al. proposed a digital twin model construction method driven by the joint data and model. The simulated vibration signal generated by the model was corrected with the measured vibration signal after being corrected by CycleGAN. highly consistent. However, the information provided by the sensor is limited, and some parameters cannot be directly obtained by the sensor, such as the local defect size of the rolling bearing, which makes the traditional digital twin model unable to display these physical quantities.

Contents of the invention

The technical problem to be solved by the present invention is to provide a digital twin model construction method that reflects the expansion of rolling bearing defects based on the shortcomings of traditional digital twin models, based on signal noise reduction methods, mathematical mechanism models, dynamic update methods and deep learning technology. Real-time prediction of time-varying local defect sizes and generation of simulated vibration signals that are highly consistent with measured vibration signals.

To achieve the above object, the present invention provides the following technical solutions:

A digital twin model construction method reflecting the expansion of rolling bearing defects, including the following steps:

S1: Obtain the full life cycle vibration signal of the rolling bearing, and use the singular value decomposition (SVD) technology to perform noise reduction preprocessing on the collected vibration signal;

S2: Construct a dynamic model describing the propagation behavior of local defects in rolling bearings based on the Hertzian contact theory;

S3: According to the signal approximation method, update the dynamic model parameters and estimate the time-varying defect size of the rolling bearing with the actual measurement data;

S4: Use the long short-term memory network (LSTM) to establish the relationship between the actual measured vibration signal and the time-varying defect, establish a digital twin model, and realize the real-time mapping from the physical entity to the virtual entity;

S5: Monitor the operating status and health of rolling bearings according to the extent of local defect expansion, estimate the current operating stages of rolling bearings in their entire life cycle, and make timely decisions and feedback before bearing failure.

The noise reduction preprocessing of the vibration signal collected by using singular value decomposition is specifically: based on the phase space reconstruction theory, constructing a Hankel matrix of r×s order, and setting the vibration signal collected by the sensor as Y=[y ₁ y ₂ …y _n ], then the Hankel matrix is shown in formula (1):

Among them, H is an r×s matrix, n is the total length of the signal, n=r+s-1, and r≤s, H can be split into H=D+W; D is the matrix of the smooth signal in the reconstruction space , W is the matrix of the noise interference signal, and the H, D, and W are all r×s order matrices, and the singular value decomposition of H can be obtained:

H = USV ^T (2)

Among them, U and V ^T are matrices of order r×r and s×s respectively, S is a diagonal matrix of order r×s, and the main diagonal is λ _i (i=1,2,…,m), where m=min(r,s), that is:

S=diag(λ ₁ ,λ ₂ ,…,λ _m ) (3)

Among them, λ ₁ , λ ₂ ,...,λ _m are the singular values of H, and λ ₁ ≥ _{λ 2} ≥... ≥ λ _m , U and V ^T are expressed as the left and right singular matrices of H; after the inverse operation of singular value decomposition , get the reconstructed matrix H _n , re-expand H _n into a new vibration signal Y _n , and get the noise-reduced vibration signal;

The singular value difference value is defined as b _i =λ _i -λ _i+1 (i=1,2,…,m-1), and the singular value difference sequence B=(b ₁ +b ₂ +…+b _m-1 ).

The dynamic model describing the expansion behavior of local defects in rolling bearings is constructed based on the Hertzian contact theory, specifically:

S2.1: Set the local defect of the rolling bearing as a cuboid defect with a depth of H, a length of L and a width equal to the width of the outer raceway of the bearing, and analyze the fault stage according to the contact between the rolling body and the outer raceway;

S2.2: Using a piecewise function as the displacement excitation, establish a two-degree-of-freedom dynamic model.

The stage of analyzing the fault according to the contact condition between the rolling element and the outer raceway includes:

According to the geometric relationship, calculate the maximum radial displacement H _d when the rolling element is in contact with the two edges of the local defect at the same time, as shown in formula (4):

Among them, r is the radius of the rolling element;

If H _d < H, the rolling element is only in contact with the edge of the defect, which is the early stage of failure;

If H _d >H, the rolling element is in contact with the edge of the defect and the bottom of the defect, which is the late stage of the fault.

The use of piecewise functions as displacement excitations to establish a two-degree-of-freedom dynamics model specifically includes:

Introduce a piecewise displacement activation function:

For the early stage of failure, the displacement excitation function is shown in formula (5):

For the later stage of the fault, the displacement excitation function is shown in formula (6):

Among them, R is the radius of the raceway of the outer ring, _β0 is the initial angular position of the local defect relative to the positive direction of the X-axis, β is the range angle of the local defect, and _θi is the angle of the i-th rolling element relative to the positive direction of the X-axis at time t β _q is the arc length when the rolling body contacts the bottom edge of the local defect, and H _max is the actual maximum radial displacement of the rolling body when the rolling body passes through the local defect, as shown in formula (7):

The contact stiffness between the rolling elements and the raceways of the inner and outer rings of the bearing satisfies the Hertz contact theory, and its load-deformation relationship is shown in formula (8):

F= ^Kδn (8)

Among them, K is the total contact stiffness between the rolling element and the inner and outer rings, F is the Hertzian contact force, δ is the contact deformation, n is the load-deformation index, n=1.5 for the ball bearing, K _i and K _{o are the rolling element and K o} respectively For the contact stiffness of the inner and outer rings, the bearing system is regarded as a concentrated spring-mass model, and a two-degree-of-freedom dynamic equation is established, as shown in formula (9):

分别为内圈在X和Y方向的振动加速度，

分别为内圈在X和Y方向的振动速度，F_x、F_y分别为水平和竖直方向的载荷分量，δ_i为第i个滚动体与滚道之间的接触变形，如式(10)所示：Among them, M is the mass sum of the inner ring and the shaft, C is the damping of the system, K is the contact stiffness between the rolling body and the inner and outer ring raceways,

are the vibration accelerations of the inner ring in the X and Y directions, respectively,

are the vibration velocities of the inner ring in the X and Y directions, F _x and F _y are the load components in the horizontal and vertical directions, respectively, and δ _i is the contact deformation between the i-th rolling element and the raceway, as shown in formula (10 ) as shown:

δ _i ＝xcosθ _i +ysinθ _i -cH′ (10)

Among them, x and y are the vibration displacement of the inner ring in the X and Y directions respectively, c is the radial travel of the rolling bearing, H' is the displacement excitation function, when H _d <H, H'=H ₁ , when H _d > When H, H'=H ₂ , λ represents the effective contact area parameter of the i-th rolling element, as shown in formula (11):

According to the signal approximation method, updating dynamic model parameters with actual measurement data and estimating the time-varying defect size of the rolling bearing specifically includes the following steps:

S3.1: Calculate the peak value pkm(i) of each group of target signals, i=1, 2,...,n, wherein, n is the serial number of the last group of target signals; the target signal is a full life cycle noise reduction signal;

S3.2: Set the initial defect length L', the step size Δ and the error δ of the stop approach, use the Runge-Kutta numerical integration method to solve the dynamic equation, and obtain the peak value pks of the vibration signal corresponding to the current defect length, if pks<pkm , L'=L'+Δ then compare pks and pkm again until pks≥pkm;

S3.3: Set L'=L'+Δ/100 to obtain the pks corresponding to the current defect length, if pks≥pkm, continue to L'=L'+Δ/100 until pks-pkm≤δ;

S3.4: Apply S3.1-S3.3 to each group of target signals to obtain the defect length L(i) of the virtual entity in the full life cycle and its corresponding vibration signal peak value pks(i);

S3.5: Substitute the obtained defect length L(i) of the virtual entity into the formula to obtain the simulated vibration signal of the whole life cycle.

The use of the long-short-term memory network (LSTM) to establish the relationship between the actual measurement vibration signal and the time-varying defect, to realize the real-time mapping from the physical entity to the virtual entity, comprises the following steps:

S4.1: Calculate the root mean square value rms(i) of the measured vibration signal, use rms(i) as the input of LSTM, and use the defect length L(i) of the virtual entity as the output to construct a sample;

S4.2: Divide the sample into a training set and a verification set in proportion to train the LSTM neural network;

S4.3: Input the vibration signal collected by the sensor into the trained LSTM neural network to obtain the estimated time-varying defect size.

Beneficial technical effect

1. Compared with the traditional digital twin model, the technical solution adopted by the present invention not only makes full use of the historical measurement data and dynamic model, but also uses the proposed dynamic update method according to the relationship between the fault mechanism and the dynamic response of the rolling bearing Dynamically update the virtual entity, so that the present invention can not only generate the simulated vibration signal consistent with the measured vibration signal, but also directly estimate the size of the local defect of the rolling bearing, and explore the expansion behavior of the local defect of the rolling bearing under the condition of continuous service. Peacekeeping health monitoring offers a new tool.

2. The present invention uses the vibration signal of the rolling bearing to bridge the physical space and the virtual space, and uses the signal approximation method to realize the dynamic update of the virtual entity without measuring the actual defect size of the bearing, which solves the problem that it is difficult to obtain the time-varying defect size in the running state of the bearing. problem. Use LSTM to establish the relationship between the root mean square value of the measured vibration signal and the length of the local defect in the virtual space, and map the physical space signal to the virtual space. The local defect size of the bearing, which helps to update the virtual entity quickly.

Description of drawings

Fig. 1 is a flow chart of a digital twin model construction method reflecting the expansion of rolling bearing defects proposed by the embodiment of the present invention;

Fig. 2 is the technical roadmap of the digital twin model construction method that the embodiment of the present invention proposes;

Fig. 3 is a schematic diagram of the expansion of local defects of the rolling bearing proposed by the embodiment of the present invention;

Fig. 4 is a schematic diagram of the geometric relationship of the rolling bearing displacement excitation function proposed by the embodiment of the present invention;

FIG. 5 is a flowchart of a signal approximation method proposed by an embodiment of the present invention.

Detailed ways

The specific implementation manners of the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments.

The present invention proposes a digital twin model construction method that reflects the expansion of rolling bearing defects. Most traditional digital twin models continuously update the parameters of virtual entities based on real-time sensor data to imitate the movement of physical entities and generate simulated dynamic responses. However, The information provided by the sensor is limited, and some parameters cannot be directly obtained by the sensor, such as the local defect size of the rolling bearing, which makes the traditional digital twin model unable to display these physical quantities. Problem, this method is based on the digital twin five-dimensional model theory, including signal processing, dynamic modeling and deep learning technology, estimates the size of the local defect size of the rolling bearing, and uses the signal approximation method to realize the dynamic update of the digital twin model to simulate In the whole process of defect expansion, LSTM is finally used to establish the relationship between the measured vibration signal of the rolling bearing and the estimated time-varying defect size to realize real-time mapping. The change of the virtual entity can represent the change of the physical entity; a digital twin model construction method for dynamically updating and real-time mapping of the local defect expansion behavior of rolling bearings, as shown in Figure 1, includes the following steps:

To achieve the above object, the present invention provides the following technical solutions:

A digital twin model construction method for dynamically updating and real-time mapping of the local defect expansion behavior of rolling bearings, as shown in Figure 1, includes the following steps:

S1: Obtain the full life cycle vibration signal of the rolling bearing, and use the singular value decomposition (SVD) technology to perform noise reduction preprocessing on the collected vibration signal;

The noise reduction preprocessing of the vibration signal collected by using singular value decomposition is specifically: based on the phase space reconstruction theory, constructing a Hankel matrix of r×s order, and setting the vibration signal collected by the sensor as Y=[y ₁ y ₂ …y _n ], then the Hankel matrix is shown in formula (1):

Among them, H is an r×s matrix, n is the total length of the signal, n=r+s-1, and r≤s, H can be split into H=D+W; D is the matrix of the smooth signal in the reconstruction space , W is the matrix of the noise interference signal, and the H, D, and W are all r×s order matrices, and the singular value decomposition of H can be obtained:

H = USV ^T (2)

Among them, U and V ^T are matrices of order r×r and s×s respectively, S is a diagonal matrix of order r×s, and the main diagonal is λ _i (i=1,2,…,m), where m=min(r,s), that is:

S=diag(λ ₁ ,λ ₂ ,…,λ _m ) (3)

Among them, λ ₁ , λ ₂ ,...,λ _m are the singular values of H, and λ ₁ ≥ _{λ 2} ≥... ≥ λ _m , U and V ^T are expressed as the left and right singular matrices of H; after the inverse operation of singular value decomposition , get the reconstructed matrix H _n , re-expand H _n into a new vibration signal Y _n , and get the noise-reduced vibration signal;

The SVD reconstruction component is determined using the singular value difference spectrum method. The difference value can reflect the degree of difference between two singular values. The larger the difference value is, the larger the difference between the singular values is. The appearance of these large difference values is due to the effective signal and noise. Due to the large irrelevance of the signal, the singular value difference spectrum sequence can describe the change and trend of the singular value sequence.

The singular value difference value is defined as b _i =λ _i -λ _i+1 (i=1,2,…,m-1), and the singular value difference sequence B=(b ₁ +b ₂ +…+b _m-1 );

In this embodiment, the purpose of performing noise reduction preprocessing on the measured vibration signal is to reduce the influence of white noise on the amplitude of the vibration signal, so that the subsequent dynamic update steps are more accurate, and the estimated size of the local defect of the rolling bearing is more accurate;

S2: Construct a dynamic model describing the propagation behavior of local defects in rolling bearings based on the Hertzian contact theory;

As shown in Figure 2, a digital twin model construction method for dynamically updating and real-time mapping of the local defect expansion behavior of rolling bearings is based on the digital twin five-dimensional model theory, which integrates physical space, virtual space, twin space, service and data connection; the model The physical space mainly includes physical platform construction, sensor arrangement, and data uploading; the virtual space mainly includes geometric models, physical models, behavior models, and rule models; the geometric model is a description of the geometric parameters of physical entities, such as shape and size; The physical model describes the physical properties, stress conditions and constraints of physical entities; the behavior model in this framework mainly represents the process of the local defects of rolling bearings expanding over time; the rule model includes regular rules based on historical associated data and implicit The experience of knowledge summary corresponds to using LSTM to establish the relationship between the measured vibration signal and the virtual entity in this framework;

The purpose is to use the mathematical mechanism model to simulate the internal motion logic of the rolling bearing, and present the influence of the local defect expansion of the rolling bearing on its dynamic response from the perspective of mechanism and mathematics; as shown in Figure 3, and then generate simulation data similar to the actual measurement data, Include the following steps:

S2.1: Set the local defect of the rolling bearing as a cuboid defect with a depth of H, a length of L and a width equal to the width of the outer raceway of the bearing, and analyze the fault stage according to the contact between the rolling body and the outer raceway;

The stage of analyzing the fault according to the contact condition between the rolling element and the outer raceway includes:

According to the geometric relationship, calculate the maximum radial displacement H _d when the rolling element is in contact with the two edges of the local defect at the same time, as shown in formula (4):

Among them, r is the radius of the rolling element;

If H _d < H, the rolling element is only in contact with the edge of the defect, which is the early stage of failure;

If H _d > H, the rolling element is in contact with the edge of the defect and the bottom of the defect, which is the late stage of the fault;

S2.2: Using a piecewise function as the displacement excitation, establish a two-degree-of-freedom dynamic model;

The use of piecewise functions as displacement excitations to establish a two-degree-of-freedom dynamics model specifically includes:

As shown in Figure 4, the segmented displacement excitation function is introduced:

For the early stage of failure, the displacement excitation function is shown in formula (5):

For the later stage of the fault, the displacement excitation function is shown in formula (6):

Among them, R is the radius of the raceway of the outer ring, _β0 is the initial angular position of the local defect relative to the positive direction of the X-axis, β is the range angle of the local defect, and _θi is the angle of the i-th rolling element relative to the positive direction of the X-axis at time t β _q is the arc length when the rolling body contacts the bottom edge of the local defect, and H _max is the actual maximum radial displacement of the rolling body when the rolling body passes through the local defect, as shown in formula (7):

The contact stiffness between the rolling elements and the raceways of the inner and outer rings of the bearing satisfies the Hertz contact theory, and its load-deformation relationship is shown in formula (8):

F= ^Kδn (8)

Among them, K is the total contact stiffness between the rolling element and the inner and outer rings, F is the Hertzian contact force, δ is the contact deformation, n is the load-deformation index, n=1.5 for ball bearings, K _i and K _o are the rolling elements and For the contact stiffness of the inner and outer rings, the bearing system is regarded as a concentrated spring-mass model, and a two-degree-of-freedom dynamic equation is established, as shown in formula (9):

分别为内圈在X和Y方向的振动加速度，

分别为内圈在X和Y方向的振动速度，F_x、F_y分别为水平和竖直方向的载荷分量，δ_i为第i个滚动体与滚道之间的接触变形，如式(10)所示：Among them, M is the mass sum of the inner ring and the shaft, C is the damping of the system, K is the contact stiffness between the rolling body and the inner and outer ring raceways,

are the vibration accelerations of the inner ring in the X and Y directions, respectively,

are the vibration velocities of the inner ring in the X and Y directions, F _x and F _y are the load components in the horizontal and vertical directions, respectively, and δ _i is the contact deformation between the i-th rolling element and the raceway, as shown in formula (10 ) as shown:

δ _i ＝xcosθ _i +ysinθ _i -cH′ (10)

Among them, x and y are the vibration displacement of the inner ring in the X and Y directions respectively, c is the radial travel of the rolling bearing, H' is the displacement excitation function, when H _d <H, H'=H ₁ , when H _d > When H, H'=H ₂ , λ represents the effective contact area parameter of the i-th rolling element, as shown in formula (11):

S3: According to the signal approximation method, the dynamic model parameters are updated with the actual measurement data and the time-varying defect size of the rolling bearing is estimated; Approximate the measured data in the physical space to the greatest extent, so that the parameters of the virtual entity are continuously updated according to the data in the physical space, and use the data of the virtual entity to represent the process of the defect expansion of the physical entity in the physical space, so as to realize the digital twin,

In this embodiment, as shown in Figure 5, since the size of the time-varying defect cannot be measured during the operation of the rolling bearing, the expansion of the defect of the bearing can only be analyzed through the vibration signal collected by the sensor, so a signal approximation method is proposed, To establish the relationship between the virtual space simulation signal and the physical space measured signal; the purpose of the signal approximation method is to approximate the simulated data generated in the virtual space to the measured data in the physical space to the greatest extent, so that the virtual entity parameters are continuously updated according to the physical space data, and use the virtual entity The data is used to represent the process of the expansion of physical entity defects in the physical space to realize the digital twin; since the target signal of the signal approximation is the measured signal after SVD noise reduction preprocessing, its amplitude changes smoothly and the impact characteristics are more obvious, so the selection of the approximation algorithm is more The simple peak is used as an indicator to evaluate the approximation effect, which specifically includes the following steps:

S3.1: Calculate the peak value pkm(i) of each group of target signals, i=1, 2,...,n, wherein, n is the serial number of the last group of target signals; the target signal is a full life cycle noise reduction signal;

S3.2: Set the initial defect length L', the step size Δ and the error δ of the stop approach, use the Runge-Kutta numerical integration method to solve the dynamic equation, and obtain the peak value pks of the vibration signal corresponding to the current defect length, if pks<pkm , L'=L'+Δ then compare pks and pkm again until pks≥pkm;

S3.3: Set L'=L'+Δ/100 to obtain the pks corresponding to the current defect length, if pks≥pkm, continue to L'=L'+Δ/100 until pks-pkm≤δ;

S3.4: Apply S3.1-S3.3 to each group of target signals to obtain the defect length L(i) of the virtual entity in the full life cycle and its corresponding vibration signal peak value pks(i);

S3.5: Substitute the obtained defect length L(i) of the virtual entity into the formula to obtain the simulated vibration signal of the whole life cycle;

S4: Use the long short-term memory network (LSTM) to establish the relationship between the actual measured vibration signal and the time-varying defect, establish a digital twin model, and realize the real-time mapping from the physical entity to the virtual entity;

The LSTM neural network establishes the relationship between the physical space measured signal and the virtual entity, and the specific steps are as follows:

S4.1: Calculate the root mean square value rms(i) of the measured vibration signal, use rms(i) as the input of LSTM, and use the defect length L(i) of the virtual entity as the output to construct a sample;

S4.2: Divide the sample into a training set and a verification set in proportion to train the LSTM neural network; in this embodiment, 70% is used as a training set and 30% is used as a verification set;

S4.3: Input the vibration signal collected by the sensor into the trained LSTM neural network to obtain the estimated time-varying defect size;

As shown in Figure 5, the amplitude of the noise-reduced signal is generally smaller than that of the noise-containing signal. After de-noising, the amplitude of the signal is smoother. This de-noised signal is more suitable as the target function for signal approximation. In the next stage, use the signal approximation method to update the virtual entity to lay the foundation;

S5: Monitor the operating status and health of rolling bearings according to the extent of local defect expansion, estimate the current operating stages of rolling bearings in their entire life cycle, and make timely decisions and feedback before bearing failure.

Claims (7)

1. A digital twin model construction method reflecting rolling bearing defect expansion, is characterized in that, comprises the following steps: S1: Obtain the vibration signal of the full life cycle of the rolling bearing, and use the singular value decomposition technology to perform noise reduction preprocessing on the collected vibration signal; S2: Construct a dynamic model describing the propagation behavior of local defects in rolling bearings based on the Hertzian contact theory; S3: According to the signal approximation method, update the dynamic model parameters and estimate the time-varying defect size of the rolling bearing with the actual measurement data; S4: Use the long-short-term memory network to establish the relationship between the actual measured vibration signal and the time-varying defect, establish a digital twin model, and realize the real-time mapping from the physical entity to the virtual entity; S5: Monitor the operating status and health of rolling bearings according to the extent of local defect expansion, estimate the current operating stages of rolling bearings in their entire life cycle, and make timely decisions and feedback before bearing failure. 2. A method for constructing a digital twin model that reflects the expansion of rolling bearing defects as claimed in claim 1, wherein said use of singular value decomposition to perform noise reduction preprocessing on the collected vibration signals, specifically: based on phase space weight Constructing theory, constructing a Hankel matrix of order r×s, setting the vibration signal collected by the sensor as Y=[y ₁ y ₂ ... y _n ], then the Hankel matrix is shown in formula (1):

Among them, H is an r×s matrix, n is the total length of the signal, n=r+s-1, and r≤s, H can be split into H=D+W; D is the matrix of the smooth signal in the reconstruction space , W is the matrix of the noise interference signal, and the H, D, and W are all r×s order matrices, and the singular value decomposition of H can be obtained: H = USV ^T (2) Among them, U and V ^T are matrices of order r×r and s×s respectively, S is a diagonal matrix of order r×s, and the main diagonal is λ _i (i=1,2,…,m), where m=min(r,s), that is: S=diag(λ ₁ ,λ ₂ ,…,λ _m ) (3) Among them, λ ₁ , λ ₂ ,...,λ _m are the singular values of H, and λ ₁ ≥ _{λ 2} ≥... ≥ λ _m , U and V ^T are expressed as the left and right singular matrices of H; after the inverse operation of singular value decomposition , get the reconstructed matrix H _n , re-expand H _n into a new vibration signal Y _n , and get the noise-reduced vibration signal; The singular value difference value is defined as b _i =λ _i -λ _i+1 (i=1,2,…,m-1), and the singular value difference sequence B=(b ₁ +b ₂ +…+b _m-1 ). 3. A method for constructing a digital twin model that reflects the expansion of rolling bearing defects as claimed in claim 1, wherein the dynamic model that describes the expansion behavior of rolling bearing local defects is constructed based on the Hertzian contact theory, specifically: S2.1: Set the local defect of the rolling bearing as a cuboid defect with a depth of H, a length of L and a width equal to the width of the outer raceway of the bearing, and analyze the fault stage according to the contact between the rolling body and the outer raceway; S2.2: Using a piecewise function as the displacement excitation, establish a two-degree-of-freedom dynamic model. 4. A method for constructing a digital twin model that reflects the expansion of rolling bearing defects as claimed in claim 3, wherein the stage of analyzing the fault according to the contact condition between the rolling element and the outer raceway includes: According to the geometric relationship, calculate the maximum radial displacement H _d when the rolling element is in contact with the two edges of the local defect at the same time, as shown in formula (4):

Among them, r is the radius of the rolling element; If H _d < H, the rolling element is only in contact with the edge of the defect, which is the early stage of failure; If H _d >H, the rolling element is in contact with the edge of the defect and the bottom of the defect, which is the late stage of the fault. 5. A kind of digital twin model construction method reflecting rolling bearing defect expansion as claimed in claim 4, is characterized in that, described adopting piecewise function as displacement excitation, establishes two degrees of freedom dynamic model, specifically comprises: Introduce a piecewise displacement activation function: For the early stage of failure, the displacement excitation function is shown in formula (5):

For the later stage of the fault, the displacement excitation function is shown in formula (6):

Among them, R is the radius of the raceway of the outer ring, _β0 is the initial angular position of the local defect relative to the positive direction of the X-axis, β is the range angle of the local defect, and _θi is the angle of the i-th rolling element relative to the positive direction of the X-axis at time t β _q is the arc length when the rolling body contacts the bottom edge of the local defect, and H _max is the actual maximum radial displacement of the rolling body when the rolling body passes through the local defect, as shown in formula (7):

The contact stiffness between the rolling elements and the raceways of the inner and outer rings of the bearing satisfies the Hertz contact theory, and its load-deformation relationship is shown in formula (8): F= ^Kδn (8) Among them, K is the total contact stiffness between the rolling element and the inner and outer rings, F is the Hertzian contact force, δ is the contact deformation, n is the load-deformation index, n=1.5 for ball bearings, K _i and K _o are the rolling elements and For the contact stiffness of the inner and outer rings, the bearing system is regarded as a concentrated spring-mass model, and a two-degree-of-freedom dynamic equation is established, as shown in formula (9):

Among them, M is the mass sum of the inner ring and the shaft, C is the damping of the system, K is the contact stiffness between the rolling body and the inner and outer ring raceways,

are the vibration accelerations of the inner ring in the X and Y directions, respectively,

are the vibration velocities of the inner ring in the X and Y directions, respectively, F _x and F _y are the load components in the horizontal and vertical directions, respectively, and δ _i is the contact deformation between the i-th rolling element and the raceway, as shown in formula (10 ) as shown: δ _i ＝xcosθ _i +ysinθ _i -cH′ (10) Among them, x and y are the vibration displacement of the inner ring in the X and Y directions respectively, c is the radial travel of the rolling bearing, H' is the displacement excitation function, when H _d <H, H'=H ₁ , when H _d > When H, H'=H ₂ , λ represents the effective contact area parameter of the i-th rolling element, as shown in formula (11):

6. A method for constructing a digital twin model reflecting the expansion of rolling bearing defects as claimed in claim 1, characterized in that, according to the signal approximation method, the parameters of the dynamic model are updated with actual measurement data and the time-varying defect size of the rolling bearing is estimated , including the following steps: S3.1: Calculate the peak value pkm(i) of each group of target signals, i=1, 2,...,n, wherein, n is the serial number of the last group of target signals; the target signal is a full life cycle noise reduction signal; S3.2: Set the initial defect length L', the step size Δ and the error δ of the stop approach, use the Runge-Kutta numerical integration method to solve the dynamic equation, and obtain the peak value pks of the vibration signal corresponding to the current defect length, if pks<pkm , L'=L'+Δ then compare pks and pkm again until pks≥pkm; S3.3: Set L'=L'+Δ/100 to obtain the pks corresponding to the current defect length, if pks≥pkm, continue to L'=L'+Δ/100 until pks-pkm≤δ; S3.4: Apply S3.1-S3.3 to each group of target signals to obtain the defect length L(i) of the virtual entity in the full life cycle and its corresponding vibration signal peak value pks(i); S3.5: Substitute the obtained defect length L(i) of the virtual entity into the formula to obtain the simulated vibration signal of the whole life cycle. 7. A kind of digital twin model construction method reflecting rolling bearing defect extension as claimed in claim 1, is characterized in that, described utilizes long-short-term memory network (LSTM) to establish the relation of actual measurement vibration signal and time-varying defect, realizes physical Real-time mapping of entities to virtual entities, including the following steps: S4.1: Calculate the root mean square value rms(i) of the measured vibration signal, use rms(i) as the input of LSTM, and use the defect length L(i) of the virtual entity as the output to construct a sample; S4.2: Divide the sample into a training set and a verification set in proportion to train the LSTM neural network; S4.3: Input the vibration signal collected by the sensor into the trained LSTM neural network to obtain the estimated time-varying defect size.

Images