A component state estimation method based on digital twin model

Abstract

The present invention belongs to the field of digital twin data-driven technology, and specifically relates to a component state estimation method based on a digital twin model; the method comprises: acquiring component parameters in real time, inputting the component parameters into an improved digital twin model, and obtaining a state estimation result of a prototype component; compared with the traditional single model simplification technology, the present invention can be effectively expanded to large and complex systems, and provides a flexible expression framework for rapid model adaptation; it avoids the rapid increase in computational cost with the increase in parameters in a single model reduction method; the use of a reduced-order model greatly speeds up the Bayesian reasoning speed, and real-time solution is easier, and the resulting probability classification provides a basis for updating the digital twin, while quantifying the uncertainty in the model and improving the accuracy of the model.

Description

Technical Field

[0001] The present invention belongs to the field of digital twin data-driven technology, and specifically relates to a component state estimation method based on a digital twin model. Background Art

[0002] Since a high-order complex system model is simplified to a low-order approximate model, it is simpler in calculation and analysis than the original high-order system model, and can also provide enough information about the original system. It is usually necessary to simplify the high-order complex system model to a low-order approximate model. This model simplification method can be called a reduction method based on mathematical models. However, due to the complex relationship between the physical model and the mathematical model parameters, it is difficult to make a physical explanation for the dynamic behavior of the system.

[0003] A single static model is a common model currently available. However, the single static model ignores the differences between similar projects and fundamentally limits the ability to model any specific factor. The digital twin paradigm aims to overcome this limitation and provide an adaptive, comprehensive and authoritative digital twin for each unique physical factor. Therefore, the digital twin model has attracted widespread attention in a range of engineering applications.

[0004] The so-called digital twin refers to the integration of multidisciplinary and multi-scale simulation processes by making full use of physical models, sensors, operation history and other data. It serves as a mirror image of the physical product in the virtual space, reflecting the entire life cycle of the corresponding physical product. The physical model of the digital twin is a computational model based on discrete partial differential equations (PDE), which provides a higher degree of interpretability, reliability and predictive ability. The computational models in digital twins are often used throughout the project, such as simulation-based vehicle certification and fleet management, structural health monitoring, and aircraft maintenance procedures. Although the prospects of digital twins are recognized, it is a considerable challenge to create, calibrate and maintain the models required for a single digital twin.

[0005] Accurately modeling a complete engineering system usually requires very large computational models and requires the evaluation of a large amount of computing resources. Such models are usually solved using methods such as finite element analysis (FEA). However, in a real-time, multi-query environment, traditional large-scale physics-based models are usually difficult to solve, which requires digital twins to adapt the computational model when the model parameters change, and quickly evaluate the model to provide analysis and predictions for effective use in operational decisions. Summary of the invention

[0006] In view of this, the present invention proposes a component state estimation method based on a digital twin model, the method comprising: acquiring component parameters in real time, inputting the component parameters into an improved digital twin model, and obtaining a state estimation result of a prototype component;

[0007] The process of processing prototype component parameters using the improved digital twin model includes:

[0008] S1: Get prototype component parameters;

[0009] S2: Decompose the prototype component parameters using a static compression reduced-order primitive algorithm to obtain sub-parameters; aggregate the sub-parameters to obtain a data set; process the data in the data set using a reduced benchmark to obtain a reduced-order data set;

[0010] S3: Input the data in the reduced-order dataset into the digital twin model to obtain the digital twin asset status of the input data;

[0011] S4: Use the hidden Markov model to perform statistical analysis on the status of the digital twin asset and obtain the probability classification result of the digital twin asset status of the current data;

[0012] S5: Obtain the probability mass function of the component based on the probability classification result of the digital twin asset status of the current data, and predict the component status based on the probability mass function;

[0013] S6: Use the Bayesian assimilation algorithm to update the status of the digital twin asset;

[0014] S7: Quantifying uncertainty and error in digital twins based on probability mass functions;

[0015] S8: Repeat steps S3 to S7 to obtain the optimal digital twin model.

[0016] Preferably, the process of decomposing the prototype component parameters by using the static compression reduction primitive algorithm includes: the process of affine decomposition of the prototype component parameters by following the reduction basis practice includes:

[0017] Define X as Hilbert space, P = R ρ For a closed parameter space, the prototype component parameters a and f can be decomposed into parameter-related parts and parameter-independent parts. The parameter-related parts and parameter-independent parts obtained by decomposition are combined to obtain the component parameter data set; the decomposition formula is:

[0018]

[0019]

[0020] Among them, R ρ represents a closed scope, u is a bounded linear function, v is the Dirichlet frontier, μ is a system-level component parameter, a(u,v;μ) is a bilinear form, f(v;u) is a linear form, Q a is the number of relevant parameter parts, Q fis the number of irrelevant parameter parts, represents an algebraic operation to calculate related parameters, represents an algebraic operation with independent parameters, a q (u,v) represents the system matrix of the reduced basis finite element, f q (v) represents the load vector.

[0021] The process of processing the data in the data set using the reduced benchmark includes: instantiating the data in the data set as the model components of the digital twin model, and obtaining the external boundary conditions of the model components; processing the sub-parameters in the data set using the control discrete partial differential equation PDE according to the external boundary conditions to eliminate the imbalance between the sub-parameters of the prototype components; weakening the sub-parameters of the prototype components after the imbalance is eliminated to obtain the weakened sub-parameters of the prototype components, and collecting all the weakened sub-parameters of the prototype components to obtain a reduced-order data set.

[0022] Preferably, the process of inputting the data in the reduced-order data set into the digital twin model for processing includes:

[0023] Input the reduced-order data set into the certified reduced benchmark RB to generate relevant parameters; establish a reduced-order model with reduced degrees of freedom according to the relevant parameters; the relevant parameters include geometric parameters, non-geometric parameters, computational grid, damage parameters, component port parameters, and component internal parameters;

[0024] Obtain the observed sensor data, combine the observed sensor data with the state transition probability model using the Bayesian state estimation method, and obtain the probability classification of the current state of the asset; input the probability classification result into the reduced-order model and update the digital twin model;

[0025] Based on the updated digital twin model, the current digital twin asset status is obtained.

[0026] Preferably, the process of using the hidden Markov model to perform statistical analysis on the state of the digital twin asset includes: obtaining the hidden state of the previous moment in the hidden Markov model; inputting the parameters of the digital twin model into the hidden Markov model, and obtaining the hidden state set at the current moment according to the hidden state sequence at the previous moment; obtaining the observation value at the current moment according to the hidden state at the current moment, and combining all the observation values ​​and hidden states to obtain the classification result of the digital twin asset state of the input data; the formula for obtaining the hidden state at the current moment is:

[0027] P(h t |h t-1 ,…,h2,h1,o t-1 ,…,o2,o1)=P(h t |h t-1 ),t=1,2,…,T

[0028] The formula for obtaining the observation at the current moment is:

[0029] P(o t |h T ,h T-1 ,…,h t+1 ,h t ,h t-1 ,…,h1,o T ,o T-1 ,…,o t+1 ,o t-1 ,…,o1)=P(o t |h t )

[0030] Among them, P(h t |h t-1 ) represents the state transfer function, P(o t |h t ) represents the likelihood function, T represents the time T, h t represents the hidden state at time t, N represents the number of hidden states, o t represents the observed value of the model at time t.

[0031] Preferably, the probability mass function of the component is obtained according to the probability classification result of the digital twin asset state of the current data, and the prediction of the digital twin asset state according to the probability mass function includes: t As a discrete random variable, the probability mass function is obtained using the posterior distribution, and the probability mass function is:

[0032]

[0033] Among them, b t (M j ) represents the model M j The belief state, u t represents the physical assets at time t, o t Represents the observed value of the model at time t.

[0034] Furthermore, the Bayesian assimilation algorithm is used to update the status of the digital twin asset, including:

[0035] Using Bayes' rule, we can obtain the value from the model M k Transfer to Model M j The probability is:

[0036]

[0037] Where M represents the number of models in the model library, u t represents control inputs that affect the state or observed dynamics;

[0038] According to the state transition probability of the model, the observation quantity of the model is predicted; the formula is:

[0039] o t =Z(M j ,u t )+v t , v t ~N(0,σ 2 )

[0040] Among them, t represents the observed quantity of the model, σ 2 represents the covariance of the observation noise, Z(M j ,u t ) indicates that the model is in M j state and the input is u t The response of the physical asset when t A discrete random variable representing the digital twin asset;

[0041] Update the status of digital twin assets based on the model’s observations.

[0042] Preferably, quantifying the uncertainty in the digital twin according to the probability mass function includes: taking the expected distribution of the observed quantity as an estimate and quantifying the uncertainty, the formula is:

[0043]

[0044] in, represents the expected value of the distribution, Q(M j ) represents the estimated effective output of physical assets, b t (M j ) represents the model M j state of belief.

[0045] Furthermore, quantifying the error in the digital twin according to the probability mass function includes: defining the error amount using the observed quantity, the formula is:

[0046]

[0047] Among them, b t represents the belief state, o t represents the observed quantity, M represents the model library, Z(M j ,u t ) indicates that the model is in M j state and the input is u t The response of the physical asset.

[0048] The beneficial effects of the present invention are as follows: compared with the traditional single model simplification technology, the component-based method is effectively extended to large and complex systems, and provides a flexible expression framework for rapid model adaptation; the static compressed reduced order element method (SCRBE) is introduced to avoid the rapid increase in computational cost with the increase of parameters in the single model reduction method; a component-based reduced order model library is proposed to be combined with Bayesian state estimation to create a data-driven physical data twin; the use of reduced order models greatly speeds up the Bayesian reasoning speed, and real-time solutions are easier. The resulting probability classification provides a basis for updating digital twins, and improves the accuracy of the model by quantifying the uncertainty in the model; through the present invention, users can obtain the state estimation results of prototype components, which can be used in multiple fields such as industrial Internet, smart cities, etc., and have good economic benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a component-based reduced-order modeling process in the present invention;

[0050] Figure 2 The state transition and observation sequence generation of the hidden Markov model in the present invention;

[0051] Figure 3 This is the recursive process of the Bayesian state estimation method in the present invention. DETAILED DESCRIPTION

[0052] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] The present invention proposes a component state estimation method based on a digital twin model, such as Figure 1 As shown, the method includes: acquiring component parameters in real time, inputting the component parameters into the improved digital twin model, and obtaining state estimation results of the prototype component;

[0054] The process of processing prototype component parameters using the improved digital twin model includes:

[0055] S1: Get prototype component parameters;

[0056] S2: Decompose the prototype component parameters using a static compression reduced-order primitive algorithm to obtain sub-parameters; aggregate the sub-parameters to obtain a data set; process the data in the data set using a reduced benchmark to obtain a reduced-order data set;

[0057] S3: Input the data in the reduced-order dataset into the digital twin model to obtain the digital twin asset status of the input data;

[0058] S4: Use the hidden Markov model to perform statistical analysis on the status of the digital twin asset and obtain the probability classification result of the digital twin asset status of the current data;

[0059] S5: Obtain the probability mass function of the component based on the probability classification result of the digital twin asset status of the current data, and predict the component status based on the probability mass function;

[0060] S6: Use the Bayesian assimilation algorithm to update the status of the digital twin asset;

[0061] S7: Quantifying uncertainty and error in digital twins based on probability mass functions;

[0062] S8: Repeat steps S3 to S7 to obtain the optimal digital twin model.

[0063] The process of decomposing the prototype component parameters using the static compression reduced-order primitive algorithm includes:

[0064] Define X as Hilbert space, P = R ρ For a closed parameter space, the prototype component parameters a and f can be decomposed into parameter-related parts and parameter-independent parts. The parameter-related parts and parameter-independent parts obtained by decomposition are combined to obtain the component parameter data set; the decomposition formula is:

[0065]

[0066]

[0067] Among them, R ρ represents a closed scope, u is a bounded linear function, v is the Dirichlet frontier, μ is a system-level component parameter, a(u,v;μ) is a bilinear form, f(v;u) is a linear form, Q a is the number of relevant parameter parts, Q f is the number of irrelevant parameter parts, represents an algebraic operation to calculate related parameters, represents an algebraic operation with independent parameters, a q (u,v) represents the system matrix of the reduced basis finite element, f q (v) represents the load vector; and Only related to the parameter μ∈P, and a q :X×X→R and f q :X→R has nothing to do with μ.

[0068] Optionally, the prototype component parameters may be a set of variables inherent to the generator simulation model that can characterize the generator model, such as: Poisson's ratio, wrinkle angle, Young's modulus, reduced stiffness, material loss, crack length, delamination size.

[0069] Processing the data in the data set using the reduced benchmark includes: instantiating the data in the data set as a model component of the digital twin model, and obtaining the external boundary conditions of the model component; preferably, the external boundary conditions are linear elastic, and the sub-parameters in the data set are processed using the control discrete partial differential equation PDE according to the external boundary conditions to eliminate the imbalance between the sub-parameters of the prototype component; weakening the sub-parameters of the prototype component after the imbalance is eliminated to obtain the weakened sub-parameters of the prototype component, and all the weakened sub-parameters of the prototype component are collected to obtain a reduced-order data set; the weakening formula is:

[0070]

[0071] Among them, X(μ) is the function space, which is used to place the components after weakening parameters.

[0072] The process of inputting data from the reduced-order dataset into the digital twin model for processing includes:

[0073] Inputting the reduced-order data set into a certified reduced benchmark (RB) to generate relevant parameters; establishing a reduced-order model with significantly reduced degrees of freedom according to the relevant parameters; preferably, the relevant parameters include: geometric parameters, non-geometric parameters, computational grids, damage parameters, component port parameters, and component internal parameters; optionally, for a generator simulation model, the relevant parameters include Poisson's ratio, wrinkle angle, Young's modulus, reduced stiffness, material loss, crack length, delamination size, rotation speed, frequency, voltage, and noise; wherein the rotation speed, frequency, voltage, and noise are component internal parameters;

[0074] Obtain the observed sensor data, and use the Bayesian state estimation method to combine the observed sensor data with the state transition probability model (usually derived from historical data or calculated through physics-based simulation) to obtain the probability classification of the current state of the asset; input the probability classification result into the reduced-order model, and assemble the system according to the model components of the reduced-order model; calculate the reduced basis approximation and solve the system; update the model library M of the digital twin model according to the solution result, M={M j |j=1,…,|M|},M j ,j=1,…,|M| represents each model in the model library, and the updated digital twin model is obtained;

[0075] Based on the updated digital twin model, the current digital twin asset status is obtained.

[0076] The hidden Markov model is used to perform statistical analysis on the state of the digital twin asset to obtain the probability classification result of the digital twin asset state of the current data, including: obtaining the hidden state of the previous moment in the hidden Markov model; inputting the parameters of the digital twin model into the hidden Markov model, and obtaining the hidden state set Q = {q1, q2, ..., q N}; Obtain the observed value at the current moment according to the hidden state at the current moment, collect all the observed values ​​and hidden states, and obtain the classification result of the digital twin asset state of the input data; the specific process is:

[0077] The state of the physical asset is defined as the parameters of the digital twin, and the evolution of the physical asset is modeled as a hidden Markov model, such as Figure 2 As shown in Figure 1, the hidden Markov model consists of two stages: the generation process of random state sequences that cannot be directly observed and the generation process of random sequences that can be directly observed. The hidden state is a set of parameters that accurately describe the physical assets. Each hidden state sequence is represented by h = {h1,h2,…,h t}, the hidden state set at time t is represented by Q = {q1,q2,…,q N}, the set of all observable observation sequences is V = {v1,v2,…,v m}; The observation sequence observed at each moment is O = {o1, o2, …, o t}, its state h at any time t t It is only related to the state at the previous moment t-1. The formula for the hidden state at the current moment is:

[0078] P(h t |h t-1 ,…,h2,h1,o t-1 ,…,o2,o1)=P(h t |h t-1 ),t=1,2,…,T

[0079] The observation independence assumption means that for the observation sequence, the observation o at any time t t Only depends on the hidden state h at that time t t , and does not depend on the hidden state and observation at other times except time t. The formula for the observation at the current time is:

[0080] P(o t |h T ,h T-1 ,…,h t+1 ,h t ,h t-1 ,…,h1,o T ,o T-1 ,…,ot+1 ,o t-1 ,…,o1)=P(o t |h t )

[0081] Among them, P(h t |h t-1 ) represents the state transfer function, P(o t |h t ) represents the likelihood function, T represents the time T, h t represents the hidden state at time t, q represents the possible hidden state, N represents the number of hidden states, and o t represents the observed value of the model at time t, v m Represents the observation quantity that may be observed at time m.

[0082] The probability mass function of the component is obtained based on the probability classification result of the digital twin asset state of the current data. The prediction of the digital twin asset state based on the probability mass function includes: t As a discrete random variable, the probability mass function is obtained using the posterior distribution, and the probability mass function is:

[0083]

[0084] Among them, b t (M j ) represents the model M j The belief state, which represents the state of the digital twin asset, u t represents the physical assets at time t, o t represents the observed quantity of the model, d t A discrete random variable representing a digital twin asset.

[0085] The Bayesian assimilation algorithm is used to update the state of the digital twin asset, including: using the Bayesian rule to obtain the k Transfer to Model M j The probability of the state transition of the model is used to predict the observation of the model; the state of the digital twin asset is updated according to the observation of the model; the specific process is as follows:

[0086] Belief state b t Recursively update at each time step, monitor the initial belief state b0 at t=0, and if t>0, update according to the previous time step b t-1 The belief state input is used to obtain the state of the digital twin asset from the model M using the Bayesian rule. k Transfer to Model M j Probability:

[0087]

[0088] Where M represents the number of models in the model library, u t Represents a control input that affects the state or observation dynamics.

[0089] Observe the digital twin asset with input u t When in state M j The probability of adding a zero-mean Gaussian error term to quickly evaluate the model and predict the observed value of the model is:

[0090] o t =Z(M j ,u t )+v t , v t ~N(0,σ 2 )

[0091] Among them, t represents the observed quantity of the model, σ 2 represents the covariance of the observation noise, Z(M j ,u t ) indicates that the model is in M j state and the input is u t The response of the physical asset.

[0092] like Figure 3 As shown in Figure 2, the specific process of state estimation using the Bayesian method is as follows:

[0093] The system model M(·) is the state transfer function P(h t |h t-1 ),t=1,2,…,T, the likelihood function between the output and the state is P(o t |h t ), the observation sequence is O = {o1,o2,…,o t}, each hidden state sequence h={h1,h2,…,h t} is known, based on Bayes' theorem, the conditional probability density is:

[0094]

[0095] Simplified to:

[0096]

[0097] The prediction process is:

[0098] p(h t |o 0:t-1 )=∫p(h t |h t-1 )p(h t-1 |o 0:t-1 )dht-1

[0099] Update process:

[0100]

[0101] Among them, p(o t ) is the prior probability of the observation sequence, p(h t ) is the prior probability of the hidden state sequence, p(o 0:t |h t ) is the posterior probability of the observed sequence, p(h t |o 0:t ) is the posterior probability of the hidden state sequence, p(o t |o 0:t-1 ) is the normalization constant, p(h t |h t-1 ) is the state transition probability.

[0102] The state prediction probability mass function is obtained using the posterior distribution, and then the output prediction function is encapsulated into the system using the Bayesian formula and the posterior likelihood to obtain the asset state update and the probability of the model transferring from one state to another.

[0103] Estimating the output of an asset based on a digital twin model includes: during the asset monitoring cycle, inputting the current state of the digital twin asset, and calculating the asset output of the current state of the digital twin asset using smoothing and prediction methods.

[0104] Quantifying uncertainty in digital twins based on probability mass functions involves taking the expected distribution of observed quantities as an estimate and quantifying uncertainty using the formula:

[0105]

[0106] in, represents the expected distribution of the observed quantity, Q(M j ) represents the estimated effective output of physical assets, b t (M j ) represents the model M j state of belief.

[0107] Quantifying the error in digital twins based on the probability mass function includes: defining the error amount using the observed quantity, the formula is:

[0108]

[0109] Among them, b t represents the belief state, o t represents the observed quantity, M represents the model library, Z(M j ,u t) indicates that the model is in M j state and the input is u t The response of the physical asset.

[0110] Compared with the traditional single model simplification technology, the component-based method of the present invention is effectively extended to large and complex systems, and provides a flexible and expressive framework for rapid model adaptation; the static compressed reduced order element method (SCRBE) is introduced to avoid the rapid increase in computational cost with the increase of parameters in the single model reduction method; a component-based reduced order model library is proposed to be combined with Bayesian state estimation to create a data-driven physical data twin; the use of reduced order models greatly speeds up the Bayesian reasoning speed and makes real-time solutions easier. The resulting probabilistic classification provides a basis for the update of digital twins, and the uncertainty in the quantified model is used to improve the accuracy of the model; through the present invention, users can obtain the state estimation results of prototype components, which can be used in multiple fields such as industrial Internet, smart cities, etc., and have good economic benefits.

[0111] The above embodiments further illustrate the purpose, technical solutions and advantages of the present invention in detail. It should be understood that the above embodiments are only preferred implementation modes of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made to the present invention within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A component state estimation method based on a digital twin model, characterized in that: include: Obtain component parameters in real time, input the component parameters into the improved digital twin model, and obtain the state estimation results of the prototype component; The process of processing prototype component parameters using the improved digital twin model includes: S1: Get prototype component parameters; S2: Decompose the prototype component parameters using a static compression reduced-order primitive algorithm to obtain sub-parameters; aggregate the sub-parameters to obtain a data set; process the data in the data set using a reduced benchmark to obtain a reduced-order data set; S3: Input the data in the reduced-order dataset into the digital twin model to obtain the digital twin asset status of the input data; S4: Use the hidden Markov model to perform statistical analysis on the state of the digital twin asset to obtain the probability classification result of the digital twin asset state of the current data; the process of using the hidden Markov model to perform statistical analysis on the state of the digital twin asset includes: obtaining the hidden state of the previous moment in the hidden Markov model; inputting the parameters of the digital twin model into the hidden Markov model, and obtaining the hidden state set of the current moment according to the hidden state sequence of the previous moment; obtaining the observation value of the current moment according to the hidden state of the current moment, and collecting all the observation values ​​and hidden states to obtain the classification result of the digital twin asset state of the input data; the formula for obtaining the hidden state of the current moment is: P(h t |h t-1 ,…,h2,h1,o t-1 ,…,o2,o1)=P(h t |h t-1 ),t=1,2,…,T The formula for obtaining the observation at the current moment is: P(o t |h T ,h T-1 ,…,h t+1 ,h t ,h t-1 ,…,h1,o T ,o T-1 ,…,o t+1 ,o t-1 ,…,o1)=P(o t |h t ) Among them, P(h t |h t-1 ) represents the state transfer function, P(o t |h t ) represents the likelihood function, T represents the time T, h t represents the hidden state at time t, N represents the number of hidden states, o t represents the observed value of the model at time t; S5: Obtain the probability mass function of the component based on the probability classification result of the digital twin asset status of the current data, and predict the digital twin asset status based on the probability mass function; the probability mass function is: Among them, b t (M j ) represents the model M j The belief state, u t represents the physical assets at time t, d t A discrete random variable representing the digital twin asset; S6: Using the Bayesian assimilation algorithm to update the state of the digital twin asset; Using the Bayesian assimilation algorithm to update the state of the digital twin asset includes: obtaining the model M according to the Bayesian rule k Transfer to Model M j The probability is: Where M represents the number of models in the model library; According to the state transition probability of the model, the observation quantity of the model is predicted; the formula is: o t =Z(M j ,u t )+v t ,v t ~N(0,σ 2 ) Among them, σ 2 represents the covariance of the observation noise, Z(M j ,u t ) indicates that the model is in M j state and the input is u t the response of physical assets; Update the status of digital twin assets based on the model’s observations; S7: Quantify uncertainty and error in digital twins based on probability mass function; Quantifying uncertainty in digital twins based on probability mass function includes: taking the expected distribution of observed quantities as an estimate and quantifying uncertainty, the formula is: in, represents the expected value of the distribution, Q(M j ) represents the estimated effective output of physical assets, b t (M j ) represents the model M j state of belief; Quantifying the error in digital twins based on the probability mass function includes: defining the error amount using the observed quantity, the formula is: Among them, b t Indicates a state of belief; S8: Repeat steps S3 to S7 to obtain the optimal digital twin model.

2. According to a component state estimation method based on a digital twin model according to claim 1, it is characterized in that: The process of decomposing the prototype component parameters using the static compression reduced-order primitive algorithm includes: Define X as Hilbert space, P = R ρ For a closed parameter space, the prototype component parameters a and f are decomposed into parameter-related parts and parameter-independent parts. The parameter-related parts and parameter-independent parts obtained by decomposition are combined to obtain the component parameter data set. The decomposition formula is: Among them, R ρ represents a closed scope, u is a bounded linear function, v is the Dirichlet boundary, μ is a system-level component parameter, a(u,v;μ) represents a bilinear form, f(v;u) represents a linear form, Q a is the number of relevant parameter parts, Q f is the number of irrelevant parameter parts, represents an algebraic operation to calculate related parameters, represents an algebraic operation with independent parameters, a q (u,v) represents the system matrix of the reduced basis finite element, f q (v) represents the load vector.

3. The component state estimation method based on the digital twin model according to claim 1 is characterized in that: The process of processing the data in the data set using the reduced benchmark includes: instantiating the data in the data set as the model components of the digital twin model to obtain the external boundary conditions of the model components; processing the sub-parameters in the data set using the control discrete partial differential equation PDE according to the external boundary conditions to eliminate the imbalance between the sub-parameters of the prototype components; weakening the sub-parameters of the prototype components after the imbalance is eliminated to obtain the weakened sub-parameters of the prototype components, and collecting all the weakened sub-parameters of the prototype components to obtain a reduced-order data set.

4. The component state estimation method based on the digital twin model according to claim 1 is characterized in that: The process of inputting data from the reduced-order dataset into the digital twin model for processing includes: Input the reduced-order data set into the certified reduced benchmark RB to generate relevant parameters; establish a reduced-order model with reduced degrees of freedom according to the relevant parameters; the relevant parameters include geometric parameters, non-geometric parameters, computational grid, damage parameters, component port parameters, and component internal parameters; Obtain the observed sensor data, combine the observed sensor data with the state transition probability model using the Bayesian state estimation method, and obtain the probability classification of the current state of the asset; input the probability classification result into the reduced-order model and update the digital twin model; Based on the updated digital twin model, the current digital twin asset status is obtained.

Images