Structural dynamics digital twin model evolution method based on Bayesian estimation

Abstract

According to the Bayesian estimation-based structural dynamics digital twin model evolution method provided by the invention, the to-be-estimated parameter vectors including the mass, damping and / or rigidity of a measured system are introduced, and priori knowledge and actual measurement response are both considered in evolution calculation of the to-be-estimated parameters, so that the accuracy of the to-be-estimated parameters is ensured; and comparing the estimated parameter change rate with a set parameter change rate safety threshold, judging whether the system performance is safe or not, and substituting the estimated parameter into the constructed dual-time scale structure dynamics digital twinborn model to obtain the required real response data of the system. The method provided by the invention effectively solves the problem of inaccurate estimation of model parameter problems in the prior art, only needs to take actual measurement response data as a target, does not need to obtain accurate modal parameters of the measured system, reduces the difficulty of data acquisition, and improves the enforceability of digital twinning of structural dynamics.

Description

Technical Field

[0001] The present invention belongs to the field of digital twin technology, and specifically relates to a structural dynamics digital twin model evolution method based on Bayesian estimation. Background Art

[0002] In the fields of aerospace and civil engineering, the safety and reliability of structures are of paramount importance. Whether it's large-scale civil structures like bridges and buildings, or aerospace equipment, these structural systems are constantly experiencing the dual impact of "slow-time" long-term service effects and "fast-time" dynamic response. Long-term service effects involve changes in system parameters caused by fatigue, wear, and aging that may occur during long-term operation, while short-term dynamic response focuses on the dynamic behavior of the system when subjected to various external stimuli.

[0003] Traditional structural health monitoring and maintenance methods often rely on periodic inspections and offline analysis, failing to achieve real-time monitoring and prediction. When dealing with complex systems (such as aircraft structures, bridges, and high-rise buildings), existing modeling and analysis methods and structural health management technologies fail to fully establish real-time interactive connections with the physical entities. They ignore the complex dynamic factors that evolve over time during actual operation, resulting in biased simulation results and difficulty accurately simulating and predicting the system's true behavior.

[0004] With the rapid development of sensor technology, the Internet of Things, and big data, digital twins of structural dynamics have emerged. These digital twins can account for changes on two different timescales. By utilizing data collected by deployed sensors, they accurately simulate the behavior of real physical systems in a digital space. They dynamically track the structure's vibration, displacement, stress, and other responses, predict its fatigue life and potential failures, optimize design and maintenance strategies, and significantly reduce experimental and maintenance costs. Digital twin technology not only systematically addresses uncertainties in sensor data and model parameters but also provides data-driven decision support for engineers and managers, driving the aerospace and civil engineering fields toward intelligent and automated development, ensuring the safety and efficient operation of structures throughout their lifecycles.

[0005] At present, the evolutionary methods in the field of structural dynamics digital twins are mainly based on dual-time scale models. By separating the system dynamic time (fast time) and the system property evolution time (slow time), the long-term changes of system properties (such as mass, stiffness, etc.) can be tracked. The evolution of the digital twin model depends on sensor data. The method currently proposed is to use sensors to collect the natural frequency and damping factor of the measured system, and directly use the collected modal parameters to estimate the stiffness and mass of the measured system. However, the inventors of this application found the following technical problems in the process of implementing the existing method: the existing method only relies on a single sensor data for parameter identification, fails to fully consider all the known information of the structure, resulting in unreliable identification results; and when processing sensor data errors and uncertainty factors, it mainly relies on statistical average data to reduce the impact of errors, lacks systematic processing of complex uncertainty factors, and the constructed model lacks adaptability, is sensitive to noise, and is difficult to cope with system parameter changes of complex multi-degree-of-freedom systems. Summary of the Invention

[0006] The purpose of the present invention is to solve the problem that the existing digital twin model evolution method is only applicable to single-degree-of-freedom systems, not to multi-degree-of-freedom systems, and the evolution results are unreliable, and to provide a structural dynamics digital twin model evolution method based on Bayesian estimation.

[0007] To achieve the above objectives, the technical solutions provided by the present invention are:

[0008] A structural dynamics digital twin model evolution method based on Bayesian estimation includes the following steps:

[0009] Step 1: First, build a simplified structural dynamic model of the system based on the initial parameters and structural characteristics of the system's real physical structure And obtain the initial mass M0, initial damping C0 and initial stiffness K0;

[0010] Then, a slow time scale is introduced into the constructed simplified structural dynamics model of the system to establish a dual-time-scale structural dynamics digital twin model:

[0011]

[0012] Where, M(t s )、C(t s ), K(t s ) are the mass, damping and stiffness to be estimated respectively; t represents the fastest time of the system vibration response, t s Indicates the slow time in the entire life cycle of the system; set t s = 0, the system is in the initial state, and in the initial state M(t s )=M0,C(t s)=C0,K(t s )=K0;

[0013] Step 2: Establish the parameter vector β(t s ), taking the system acceleration response as the output and introducing Gaussian noise v, the system acceleration observation equation is obtained

[0014]

[0015] Among them, the parameter vector β(t s ) includes the mass M(t s ), damping C(t s ) and / or stiffness K(t s ); y is the system acceleration response, Represents the actual acceleration response and the parameter vector β(t s ) between the nonlinear functional relationship; define Gaussian noise v to have a mean of 0 and a variance of σ 2 Normal distribution;

[0016] Step 3: Based on the system acceleration observation equation constructed in step 2, the parameter vector β(t s )'s logarithmic posterior probability density function

[0017]

[0018] Where N is the number of acceleration response data sampling points;

[0019] Step 4: Set the evolution period T and assume that M(t s )=M,C(t s )=C,K(t s )=K; Combined with the dual-time-scale structural dynamics digital twin model established in step 1, the state space equation of the dual-time-scale structural dynamics digital twin model is obtained:

[0020]

[0021] Where, is the state vector, is the system acceleration response; F(t) is the external load matrix, matrix matrix Matrix C = [-M -1 KM -1 C],D=M -1 ; O is the zero matrix, I is the unit matrix;

[0022] Step 5: After the system runs for a full evolution cycle, the actual acceleration response data of the system is collected, and the system acceleration response is obtained by solving the state space equation of the dual-time-scale structural dynamics digital twin model; combined with the parameter vector β(t s ) is used to update the logarithmic posterior probability density function and obtain the optimal estimate of the parameter vector to be estimated using the maximum a posteriori estimation method.

[0023] Step 6: Set safety thresholds for the rate of change of mass, damping, and / or stiffness; Obtain estimated mass, damping, and / or stiffness, and determine whether the estimated rate of change of the mass, damping, and / or stiffness satisfies the corresponding rate of change safety threshold. If so, proceed to step 7. If the rate of change of any parameter does not meet the set rate of change safety threshold, the system structure is in a dangerous state and the evolution ends.

[0024] Among them, the parameter change rates of mass, damping and stiffness are calculated based on the estimated mass, damping, stiffness and the initial mass, damping and stiffness respectively;

[0025] Step 7: The optimal estimate obtained in step 5 Substitute into the dual-time-scale structural dynamics digital twin model established in step 1 and output the system response.

[0026] Furthermore, in step 1, the slow time t s The unit is determined according to the system structure characteristics, the slow time t s Hours, days, months or years.

[0027] Furthermore, in step 3, the parameter vector β(t s ) includes:

[0028] Step 3.1: Based on the system acceleration observation equation constructed in step 2, establish the likelihood function of the system acceleration response;

[0029]

[0030] Step 3.2: Make a priori assumptions: Use non-informative priors for the parameter vector to be estimated, assuming that the parameter vector to be estimated satisfies a uniform distribution within the set interval and has a variance σ 2 The inverse gamma distribution is used as the conjugate prior, i.e.

[0031] Step 3.3: Take the product of the likelihood function of the acceleration response and the prior distribution as the posterior distribution and normalize it to obtain P(β,σ 2 |y)∝L(β,σ2 |y)P(β)P(σ 2 ), then the logarithmic equation of the posterior distribution is:

[0032] lnP(β,σ 2 |y)=lnL(β,σ 2 |y)+lnP(β)+lnP(σ 2 );

[0033] Step 3.4: Substitute the likelihood function of the acceleration response established in step 3.1 and the prior hypothesis in step 3.2 into the logarithmic equation of the posterior distribution to obtain the parameter vector β(t s ) is the logarithmic posterior probability density function of .

[0034] Furthermore, in step 4, the evolution period T is set to a single slow time t s Or slow time t s multiples of .

[0035] Furthermore, in step 5, the maximum a posteriori estimation method is used to update and solve the optimal estimate of the parameter vector to be estimated. The process is:

[0036] According to the parameter vector β(t s ) logarithmic posterior probability density function, and establish the target optimization function of maximum a posteriori estimation And min(J(β,σ 2 )) is used as the optimization target to optimize the target optimization function and obtain the optimal estimate of the parameter vector to be estimated

[0037] Furthermore, in step 5, the L-BFGS quasi-Newton method is used to optimize and solve the target optimization function to obtain the optimal estimate of the parameter vector to be estimated.

[0038] Furthermore, in step 6, if the system is a multi-degree-of-freedom system, different structural position points are selected and different parameter change rate safety thresholds are set respectively.

[0039] Furthermore, in step 6, the safety threshold of the stiffness parameter change rate is no more than 5%.

[0040] The advantages of the present invention are:

[0041] 1. In the structural dynamics digital twin model evolution method proposed in the present invention, an estimated parameter vector including the mass matrix, damping matrix and / or stiffness matrix of the measured system is introduced, and both prior knowledge and measured responses are taken into account in the evolution calculation of the estimated parameters, ensuring the accuracy of the estimated parameters. Finally, the structural dynamics model of the measured system is evolved using the estimated parameters. The model evolution results are reliable, which effectively solves the problem of inaccurate estimation of model parameters in the existing technology.

[0042] 2. The method of the present invention only needs to target the measured response data, which can be acceleration, velocity or displacement. It is not necessary to obtain accurate modal parameters of the measured system, which reduces the difficulty of data acquisition and improves the feasibility of structural dynamics digital twins.

[0043] 3. In the method of the present invention, the response error is taken into account during the evolution of the structural dynamics model of the measured system using the estimated parameters, thereby improving the noise resistance of the structural dynamics digital twin evolution and achieving high evolution accuracy.

[0044] Additional aspects and advantages of the present invention will be set forth in part in the description which follows and, in part, will be obvious from the description which follows, or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments with reference to the accompanying drawings, in which:

[0046] Figure 1 This is a flow chart of the structural dynamics digital twin model evolution method based on Bayesian estimation of the present invention;

[0047] Figure 2 This is a comparison diagram of the simulated response and the emulation response of the single-degree-of-freedom system based on digital twins in Example 1;

[0048] Figure 3 This is a comparison chart of the accuracy of the system parameter evolution results obtained by the method of the present invention for the single-degree-of-freedom system in Example 1 and the accuracy of the parameter results obtained by the traditional calculation method;

[0049] Figure 4 It is a structural diagram of the three-layer shear frame in the second embodiment;

[0050] Figure 5 This is a comparison chart of the first-layer simulation response and the emulation response of the three-layer shear frame digital twin model in Example 2;

[0051] Figure 6 This is a comparison chart of the simulated response and the emulated response of the second layer of the three-layer shear frame digital twin model in Example 2;

[0052] Figure 7 This is a comparison chart of the simulated response and the emulated response of the third layer of the three-layer shear frame digital twin model in Example 2;

[0053] Figure 8 This is a comparison chart of the accuracy of the system parameter evolution results obtained by the method of the present invention for the three-layer shearing framework in Example 2 and the accuracy of the parameter results obtained by the traditional calculation method. DETAILED DESCRIPTION

[0054] Reference Figure 1 In order to overcome the problems of traditional structural dynamics digital twin model evolution methods, such as inaccurate identification results, inapplicability to complex multi-degree-of-freedom systems, and sensitivity to noise, this embodiment provides a structural dynamics digital twin model evolution method based on Bayesian estimation. In order to better understand the technical solution of the present invention, the technical solution of the present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments are exemplary and are intended to be used to explain the present invention, and should not be understood as limiting the present invention.

[0055] Example 1: Single degree of freedom discrete system as the research object

[0056] In the study of structural dynamics, the single degree of freedom system is the model that researchers usually consider first, because it can simplify the complexity of the problem and reveal many basic characteristics of the vibration system. In order to facilitate the explanation of the principle of the method of the present invention, Example 1 first uses a single degree of freedom system as a simulation example to verify the feasibility and effectiveness of the structural dynamics digital twin model evolution method based on Bayesian estimation. Assume that the single degree of freedom system is in the initial state, with initial mass m0 = 1kg, initial damping k0 = 1N / m, and initial stiffness c0 = 0.1N·s 2 / m, and assuming that the stiffness of the single degree of freedom system changes with the slow time t s The following describes in detail the process of model evolution of the single degree of freedom described in Example 1 based on the method of the present invention.

[0057] Step 1: According to the dynamic differential equation of the single degree of freedom system The dual-time-scale structural dynamics digital twin model of the single-degree-of-freedom system is constructed as

[0058]

[0059] Among them, t s It represents the slow time of the single degree of freedom system in its entire life cycle, and the unit here is hours (h); t is the physical time of the vibration response of the single degree of freedom system, also known as the fast time, and the unit is seconds (s).

[0060] Step 2: Taking into account the mass, stiffness and damping parameters of the single degree of freedom system, establish the parameter vector to be estimated β = [m(t s ),c(t s ),k(t s )] T Taking the acceleration response as the output and introducing Gaussian noise v, the system acceleration observation equation is obtained

[0061]

[0062] Among them, the parameter vector β(t s ) includes the mass m(t s ), damping c(t s ) and / or stiffness k(t s ); y is the system acceleration response, Represents the actual acceleration response and the parameter vector β(t s ) between the nonlinear functional relationship; define Gaussian noise v to have a mean of 0 and a variance of σ 2 Normal distribution.

[0063] Step 3: For mass and stiffness parameters, since they are assumed to decrease on slow time scales, the mass m(t s ) and stiffness k(t s ) are all within the set range (β min ,β max ) satisfies uniform distribution, where β min is the minimum value that the corresponding parameter can take, β max The maximum value of the corresponding parameter can be set by those skilled in the art according to the characteristics of the structural system, which is common knowledge. In this embodiment, it is assumed that the mass m(t s ) and stiffness k(t s ) are all uniformly distributed within (0.5,1); at the same time, for σ 2 Using the inverse gamma distribution as the conjugate prior Then the logarithmic posterior probability density function of the parameter vector β to be estimated is:

[0064]

[0065] Where N is the number of acceleration response data sampling points.

[0066] Step 4: Set the evolution period to 1 hour. Freeze the slow time t s , that is, set m(t s )=m,c(t s )=c,k(t s)=k. Combining the differential equation of the single degree of freedom system in step 1, we can obtain its state space equation form:

[0067] Step 5: After the single-degree-of-freedom system runs for a full evolution cycle, the system's noisy acceleration response data y is collected using sensors, and the system's acceleration response is obtained by solving the state space equation of the dual-time-scale structural dynamics digital twin model in step 4; the collected and solved acceleration response is brought into step 3 to solve for the logarithmic posterior probability density function of the parameter vector β to be estimated, and the logarithmic posterior probability density function of the parameter vector β to be estimated obtained in step 3 is negatively signed to obtain the objective function of the maximum a posteriori estimation. Use L-BFGS method to solve min(J(β,σ 2 )) The optimal estimate of the parameter vector β to be estimated is

[0068] In order to verify the accuracy of the optimal estimate obtained by the method of the present invention, the quality of the single degree of freedom system after degradation is obtained based on simulation analysis in this embodiment. stiffness Damping The degraded mass, stiffness and damping of the system are brought into the state space equation of the single degree of freedom system to solve the system output response, such as Figure 2 As shown, the blue is the displacement response of the single-freedom system obtained based on simulation, and the green is the displacement response estimated based on the method of the present invention. Figure 2 It can be seen that the response estimated based on the method of the present invention is basically consistent with the acceleration response obtained by the simulation method. In order to illustrate the accuracy of the method of the present invention, the accuracy of parameter estimation is defined as To illustrate the accuracy of the maximum a posteriori estimate. Figure 3 As shown in the figure, the accuracy of the system parameter evolution results obtained based on the maximum a posteriori estimation method in this embodiment is compared with the accuracy of the system parameter results obtained by the traditional least squares method. The blue ones are the system stiffness, damping, stiffness and mass estimation accuracies obtained by the maximum a posteriori estimation method in the present invention, which are 0.92, 0.996 and 0.99 respectively, which are much better than the estimation results of the least squares method.

[0069] Step 6: Set the safety thresholds of the change rates of mass, damping and stiffness parameters to μ m =10%,μ c =20% and μ k =5%. Calculate the mass m(t s ), damping c(t s ) and stiffness k(t s), where the stiffness change rate λ k =9.9%>μ k , the mass and damping have not changed, which means that the single degree of freedom system has undergone significant changes within one evolution cycle, and the system stiffness may be in a dangerous state.

[0070] Example 2 takes a three-layer shear frame structure as the research object

[0071] In the field of civil engineering, multi-layer shear frame structure is a very common structural form. Multi-layer shear frame structure is relatively simple in structure and can better simulate the response characteristics of actual structure under dynamic loads such as earthquakes, and is often selected as the research object for building structure safety analysis. This model can accurately capture the vibration characteristics of the structure and provide key support for the construction and verification of digital twins; in addition, the shear frame model has the convenience of both experimental verification and numerical simulation, which effectively guarantees the accuracy and reliability verification of the digital twin method. Therefore, Example 2 takes the three-layer shear frame structure as the research object, and the specific structure is as follows: Figure 4 As shown, assuming that the mass and stiffness of each frame layer are equal in the initial state, mass m = 3.380 kg, stiffness k = 23128 N / m. The following describes in detail the process of model evolution of the three-layer frame structure described in Example 1 based on the method of the present invention.

[0072] Step 1: Establish a simplified dynamic model of the three-layer shear frame Where M, C, and K are the third-order mass matrix, damping matrix, and stiffness matrix, respectively. F(t) is the excitation force matrix, and x is the displacement response matrix of the system. Based on the initial conditions, the specific mass matrix and stiffness matrix can be given as follows:

[0073]

[0074] The damping adopts the Rayleigh damping assumption, C = αM + βK, where α is 0.01 and β is 0.001. Introduce the slow time t s Establish a dual-time-scale structural dynamics digital twin model of a three-degree-of-freedom system:

[0075]

[0076] Since the evolution of such structural parameters is relatively slow, the unit of slow time in this embodiment is year (y).

[0077] Step 2: For shear beam structures, the stiffness will slowly degrade, so only the stiffness parameter changes are considered. According to the specific form of the stiffness matrix, the parameter vector to be estimated is given as β = [k1(t s ),k2(t s ),k3(t s )]T Taking the acceleration response as the output and introducing Gaussian noise v, the acceleration observation equation of the shear beam structure is obtained: Define Gaussian noise v to have a mean of 0 and a variance of σ 2 Normal distribution.

[0078] Step 3: The prior assumption of stiffness is that k belongs to the uniform distribution of (20000, 24000), σ 2 Still using the inverse gamma distribution as the conjugate prior Then the logarithmic posterior probability density function of the parameter vector β to be estimated is:

[0079]

[0080] Where N is the number of acceleration response data sampling points.

[0081] Step 4: Set the evolution period to 1 year and freeze the slow time, that is, set M(t s )=M,C(t s )=C,K(t s ) = K. Write the digital twin model established in step 1 into state space form in is the state vector, with acceleration response As system output, where: matrix C=[-M -1 KM -1 C],D=M -1 ; Where F(t) is the external load matrix, O is the zero matrix, and I is the unit matrix;

[0082] Step 5. After the shear beam structure runs for a full evolution cycle, the system's noisy acceleration response data y is collected using sensors, and the system's acceleration response is obtained by solving the state space equation of the dual-time-scale structural dynamics digital twin model in step 4. The collected and solved acceleration response is brought into step 3 to solve for the logarithmic posterior probability density function of the parameter vector β to be estimated. The objective function of maximum a posteriori estimation is obtained by taking the negative sign of the logarithmic posterior probability density function of the parameter vector β to be estimated obtained in step 3:

[0083]

[0084] L-BFGS optimization method is used to solve min(J(β,σ 2 )) can get the optimal estimate of the parameter vector β

[0085] In order to verify the accuracy of the optimal estimation obtained by the method of the present invention, in Example 2, based on the simulation analysis, the stiffness of each layer of the three-layer shear frame structure after degradation is k1(t s1 )=22284(N / m),k2(t s1 )=22550(N / m),k3(t s1 )=23000(N / m), and the stiffness parameters of each layer are brought into the state space equation of the three-layer shear frame structure to directly calculate the system displacement response, such as Figure 5-Figure 7 As shown, the blue curve is the displacement response of the single-freedom system obtained based on simulation, and the green curve is the displacement response estimated based on the method of the present invention. In order to illustrate the accuracy of the method of the present invention, Figure 8 The accuracy of the system parameter evolution results obtained by the maximum a posteriori estimation method of the present invention and the accuracy of the system parameter results obtained by the traditional least squares method are shown. Figure 8 The medium blue represents the accuracy of each layer stiffness identified by the method of the present invention, which are 0.993, 0.994, and 0.995 respectively. The orange represents the recognition accuracy of the traditional least squares method. It can be seen from the results that the recognition accuracy based on the method of the present invention is better than that of the traditional least squares method.

[0086] Step 6: Set the safety thresholds of the change rates of the three stiffness parameters to μ k =5%. Calculate the stiffness parameter k1(t s )、k2(t s )、k3(t s ) rate of change where max(λ k )=3.6%<μ k , indicating that the change rate of each layer's stiffness is within the safe range. This means that the stiffness of the system meets the performance requirements after one evolution cycle.

[0087] Step 7: The optimal estimate obtained in step 5 Bring it into the dual-time-scale structural dynamics digital twin model established in step 1, complete the model evolution and output the system displacement response data. Figure 5-7 As shown in the figure, the blue represents the actual displacement response obtained by simulation, and the green represents the displacement response obtained using the evolution method of the present invention. As can be seen from the figure, the displacement response obtained using the digital twin model evolution method of the present invention is basically consistent with the real system, and the structural dynamics digital twin model of the three-layer shear beam has been successfully evolved.

[0088] It should be noted that, in the method of the present invention, the system response may be an acceleration response, a displacement response or a velocity response.

[0089] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present invention, and these modifications or replacements should all be included in the scope of protection of the present invention.

Claims

1. A structural dynamics digital twin model evolution method based on Bayesian estimation, characterized by: The following steps are involved: Step 1: First, build a simplified structural dynamic model of the system based on the initial parameters and structural characteristics of the system's real physical structure And obtain the initial mass M0, initial damping C0 and initial stiffness K0; Then, a slow time scale is introduced into the constructed simplified structural dynamics model of the system to establish a dual-time-scale structural dynamics digital twin model: Where, M(t s )、C(t s ), K(t s ) are the mass, damping and stiffness to be estimated respectively; t represents the fastest time of the system vibration response, t s Indicates the slow time in the entire life cycle of the system; Set t s = 0, the system is in the initial state, and in the initial state M(t s )=M0,C(t s )=C0,K(t s )=K0; Step 2: Establish the parameter vector β(t s ), taking the system acceleration response as the output and introducing Gaussian noise v, the system acceleration observation equation is obtained Among them, the parameter vector β(t s ) includes the mass M(t s ), damping C(t s ) and / or stiffness K(t s ); y is the system acceleration response, Represents the actual acceleration response and the parameter vector β(t s ) between the nonlinear functional relationship; define Gaussian noise v to have a mean of 0 and a variance of σ 2 Normal distribution; Step 3: Based on the system acceleration observation equation constructed in step 2, the parameter vector β(t s )'s logarithmic posterior probability density function Where N is the number of acceleration response data sampling points; Step 4: Set the evolution period T and assume that M(t s )=M,C(t s )=C,K(t s )=K; Combined with the dual-time-scale structural dynamics digital twin model established in step 1, the state space equation of the dual-time-scale structural dynamics digital twin model is obtained: Where, is the state vector, is the system acceleration response; F(t) is the external load matrix; matrix matrix Matrix C = [-M -1 KM -1 C],D=M -1 ; O is the zero matrix, I is the unit matrix; Step 5: After the system runs for a full evolution cycle, the actual acceleration response data of the system is collected, and the system acceleration response is obtained by solving the state space equation of the dual-time-scale structural dynamics digital twin model; combined with the parameter vector β(t s ) is used to update the logarithmic posterior probability density function and obtain the optimal estimate of the parameter vector to be estimated using the maximum a posteriori estimation method. Step 6: Set safety thresholds for the rate of change of mass, damping, and / or stiffness; Obtain estimated mass, damping, and / or stiffness, and determine whether the estimated rate of change of the mass, damping, and / or stiffness satisfies the corresponding rate of change safety threshold. If so, proceed to step 7. If the rate of change of any parameter does not meet the set rate of change safety threshold, the system structure is in a dangerous state and the evolution ends. Among them, the parameter change rates of mass, damping and stiffness are calculated based on the estimated mass, damping, stiffness and the initial mass, damping and stiffness respectively; Step 7: The optimal estimate obtained in step 5 Substitute into the dual-time-scale structural dynamics digital twin model established in step 1 and output the system response.

2. The structural dynamics digital twin model evolution method based on Bayesian estimation according to claim 1 is characterized in that: In step 1, the slow time t s The unit is determined according to the system structure characteristics, the slow time t s Hours, days, months or years.

3. The structural dynamics digital twin model evolution method based on Bayesian estimation according to claim 1 or 2, characterized in that: In step 3, the parameter vector β(t s ) includes: Step 3.1: Based on the system acceleration observation equation constructed in step 2, establish the likelihood function of the system acceleration response; Step 3.2: Make a priori assumptions: Use non-informative priors for the parameter vector to be estimated, assuming that the parameter vector to be estimated satisfies a uniform distribution within the set interval and has a variance σ 2 The inverse gamma distribution is used as the conjugate prior, i.e. Step 3.3: Take the product of the likelihood function of the acceleration response and the prior distribution as the posterior distribution and normalize it to obtain P(β,σ 2 |y)∝L(β,σ 2 |y)P(β)P(σ 2 ), then the logarithmic equation of the posterior distribution is: lnP(β,σ 2 |y)=lnL(β,σ 2 |y)+lnP(β)+lnP(σ 2 ); Step 3.4: Substitute the likelihood function of the acceleration response established in step 3.1 and the prior hypothesis in step 3.2 into the logarithmic equation of the posterior distribution to obtain the parameter vector β(t s ) is the logarithmic posterior probability density function of .

4. The structural dynamics digital twin model evolution method based on Bayesian estimation according to claim 1 or 2 is characterized in that: In step 4, the evolution period T is set to a single slow time t s Or slow time t s multiples of .

5. The structural dynamics digital twin model evolution method based on Bayesian estimation according to claim 3 is characterized in that: In step 5, the maximum a posteriori estimation method is used to update and solve the optimal estimate of the parameter vector to be estimated. The process is: According to the parameter vector β(t s ) logarithmic posterior probability density function, and establish the target optimization function of maximum a posteriori estimation And min(J(β,σ 2 )) is used as the optimization target to optimize the target optimization function and obtain the optimal estimate of the parameter vector to be estimated 6. The structural dynamics digital twin model evolution method based on Bayesian estimation according to claim 5 is characterized in that: In step 5, the L-BFGS quasi-Newton method is used to optimize the target optimization function to obtain the optimal estimate of the parameter vector to be estimated.

7. The structural dynamics digital twin model evolution method based on Bayesian estimation according to claim 1 or 2, characterized in that: In step 6, if the system is a multi-degree-of-freedom system, different structural position points are selected and different parameter change rate safety thresholds are set respectively.

8. The structural dynamics digital twin model evolution method based on Bayesian estimation according to claim 7 is characterized in that: In step 6, the safety threshold of the stiffness parameter change rate is no more than 5%.

Images