A simulation test method and system for instruments and meters under multi-physical field coupling

Abstract

The application discloses a kind of simulation test method and system of instrument and meter under the action of multiple physical field coupling, and it is related to electric digital data processing technical field.It includes: multiple physical field is divided into fast variable field sub-domain and slow variable field sub-domain based on time scale threshold, fitting is carried out to coupling region, and multiple physical field coupling model is constructed;Its control equation is solved by asynchronous coupling, and full-field state data set is obtained;Data assimilation is carried out to hidden state variable, and optimal estimate value is obtained;Optimal estimate value is decomposed into fast and slow, and drift compensation is carried out to original output signal, and compensation output signal is obtained;The comparison deviation of compensation output signal and reference measurement value is used as new information to carry out inversion to boundary condition parameter, and optimal estimate is obtained;Based on optimal estimate, multiple physical field coupling model is carried out on-line closed loop correction, and updated full-field state data set is obtained, and dynamic response characteristic index is determined.The application significantly improves the inversion accuracy of real dynamic response characteristic under complex environment.

Description

Technical Field

[0001] This invention relates to the field of electronic digital data processing technology, and in particular to a simulation testing method and system for instruments under multi-physics coupling. Background Technology

[0002] With the development of electronic digital data processing technology, multi-physics coupling simulation technology has emerged. Multi-physics coupling simulation technology can integrate multiple physical field information in the working environment of instruments and meters, and provide a new path for dynamic response simulation testing of instruments and meters by modeling and simulating the interaction of various physical fields.

[0003] In traditional technologies, simulation testing methods for the dynamic response characteristics of instruments are mostly based on single-physics field modeling to simulate the influence of a single physical field on the instrument. Simultaneously, a synchronous coupling solution method is used to handle multi-physics field problems to obtain the dynamic response characteristics of the instrument. Patent No. CN119004751A discloses a multi-physics field accelerated failure simulation method for substation sensors. This method proposes to simulate the physical geometric model corresponding to the instrument separately for temperature and random vibration physical fields, obtaining individual temperature simulation results and random vibration simulation results. Then, the electromagnetic-temperature-vibration simulation results are compared with the single-physics field simulation results to obtain the dynamic response characteristics of the instrument.

[0004] However, the simulation in this method is based on static preset boundaries, and the nonlinear interaction of each physical field at the coupling interface is severed. It is impossible to make real-time dynamic estimation of the hidden state variables (such as internal cumulative damage and time-varying parameters) during the operation of the instrument, and it is difficult to reproduce the real dynamic response characteristics under complex environments with high precision. Summary of the Invention

[0005] Therefore, it is necessary to provide a simulation testing method and system for instruments under multi-physics coupling to address the above-mentioned technical problems.

[0006] This invention provides a simulation testing method for instruments under multi-physics coupling, comprising: To acquire the various physical fields that interact and influence each other in the working environment of the instrument under test; Based on a time scale threshold, multiple physical fields are divided into fast-changing field subdomains and slow-changing field subdomains. A pre-trained physical information neural network is used to fit the coupling region between the fast-changing field subdomains and the slow-changing field subdomains to obtain a multi-physics coupling model. The control equations of the multi-physics coupling model are solved asynchronously to obtain the full-field state dataset. Based on the multiphysics coupling model and the full-field state dataset, the hidden state variables of the instrument under test are assimilated by the reduced-order extended Kalman filter algorithm to obtain the optimal estimate of the hidden state variables. The optimal estimate of the hidden state variable is decomposed into fast and slow components to obtain fast and slow components; based on the fast and slow components, the original output signal acquired in real time by the instrument under test is drift compensated to obtain the compensated output signal. The compensated output signal is compared with the reference measurement value representing the real physical quantity to obtain the comparison deviation; the comparison deviation is used as information to invert the boundary condition parameters of the multiphysics coupling model through the ensemble Kalman filter algorithm to obtain the optimal estimate of the boundary condition parameters. The optimal estimation of boundary condition parameters is used to perform online closed-loop correction of the multiphysics coupling model, resulting in an updated full-field state dataset. Based on the updated full-field state dataset, the dynamic response characteristics of the instruments are determined.

[0007] Optionally, training a physical information neural network specifically includes: The system acquires various historical physical fields that interact and influence each other in the working environment of the instrument, and divides these historical physical fields into historical fast-changing field subdomains and historical slow-changing field subdomains based on time scale thresholds. The historical rapidly changing field subdomain is solved recursively time-by-time using the explicit time integration method to record the historical rapidly changing field state variables at each time point, thus obtaining the historical rapidly changing field time series. The historical slowly changing field subdomain is solved using the implicit time integration method to obtain the historical slowly changing field state variables at each time point. A physical information neural network is used to fit the fast-changing field time series in the coupling region. During the fitting process, the network loss function is constructed based on the following formula: ; ); ; in, The output of the physical information neural network, These are the training parameters for the physical information neural network. For time, The number of solution steps for a rapidly varying field subdomain within a single slowly varying field subdomain. Let this be the current solution time point for the slowly varying field subdomain. To determine the solution time step for the rapidly varying field subdomain, For fast-changing field time series, To balance the coefficients of the data fitting term and the physical constraint term, For the physical governing equations satisfied by the rapidly varying field subdomain, Indicated by The physical information of the network parameters is mapped to the neural network. Time is the independent variable. The vector of parameters to be trained consists of network weights and biases. For fast-changing field diffusion terms, The diffusion coefficient of the medium, For gradient operators, These are terms representing the rapidly changing field source term, external excitation, and multiphysics coupling interaction term. The physical information neural network is trained with the goal of minimizing the network loss function, resulting in the trained physical information neural network.

[0008] Optionally, a multi-physics coupling model is obtained by fitting the coupling region between the rapidly varying field subdomain and the slowly varying field subdomain using a pre-trained physical information neural network; the governing equations of the multi-physics coupling model are then solved asynchronously to obtain the full-field state dataset. The fast-changing field subdomain is solved recursively time-by-time using the explicit time integration method to record the fast-changing field state variables at each time step, thus obtaining the fast-changing field time series. The slow-changing field subdomain is solved using the implicit time integration method to obtain the slow-changing field state variables at each time step. The precise value of the slowly varying field subdomain at each time step is determined by a pre-trained physical information neural network, and the precise value is passed as an equivalent load to the control equation of the slowly varying field subdomain to obtain the updated slowly varying field state quantity. At each time step, the state variables of the slowly varying field are decomposed into eigenorthogonal components to extract time coefficients and interpolate them to obtain a continuously varying function. Based on the continuously varying function, the boundary conditions of the slowly varying field corresponding to each time step of the rapidly varying field subdomain are determined and passed to the control equations of the rapidly varying field subdomain to obtain the updated rapidly varying field time series. The updated fast-changing field time series and the updated slow-changing field state variables at all times are integrated to obtain the full-field state dataset; The governing equations for the slowly varying field subdomains include second-order partial differential equations with constant coefficients that describe the variation of physical quantities in the slowly varying field subdomains. The governing equations for the rapidly varying field subdomains include first-order partial differential equations with constant coefficients that describe the variation of physical quantities in the rapidly varying field subdomains.

[0009] Optionally, based on the following formula, the rapidly changing field subdomain is recursively solved time-by-time using an explicit time integration method to record the rapidly changing field state variables at each time point, thereby obtaining the rapidly changing field time series: ; ; Based on the following formula, the state variables of the slowly varying field at each time step are obtained by solving the slowly varying field subdomain using the implicit time integration method: ; ; in, For the first The physical field state quantities of a rapidly changing field subdomain. For the first The physical field state quantities of a rapidly changing field subdomain. To determine the solution time step for the rapidly varying field subdomain, For the first The time value at any given moment For the first Evolution operator of physical field state quantities at time step For the first The physical field state quantities of a time-varying slow field subdomain. For the first The physical field state variables to be solved in a time-varying slow field subdomain. Let be the solution time step for the slowly varying field subdomain. For the first The time value at any given moment For the first Evolution operator of physical field state quantities at time step For the physical field diffusion term, The diffusion coefficient of the medium, For gradient operators, This is the physical field source term.

[0010] Optionally, based on the multiphysics coupling model and the full-field state dataset, the latent state variables of the instrument under test are assimilated using a reduced-order extended Kalman filter algorithm to obtain the optimal estimates of the latent state variables. Specifically, this includes: Based on the control equations of the multiphysics coupling model, state transition equations are constructed to describe the recursive evolution of hidden state variables over time, and observation equations are constructed to describe the mapping relationship between hidden state variables and the raw output signals acquired in real time by instruments. The state vectors in the full-field state dataset are subjected to intrinsic orthogonal decomposition, and the basis matrix composed of orthogonal eigenvectors whose cumulative contribution rate satisfies a preset threshold during the intrinsic orthogonal decomposition is used as the reduced-order basis. Based on the following equation, the state transition equation and observation equation are projected onto a low-dimensional linear space composed of column vectors of the reduced-order basis to obtain the reduced-order state transition equation and reduced-order observation equation: ; in, This is the reduced-order state vector at the current moment. for The reduced-order state vector at time t, This is a reduced-order basis obtained from eigenorthogonal decomposition. Here is the state transition matrix. For the observation vector, For the observation matrix, For process noise, To observe noise; The projection of the state vector onto the reduced basis is used as the initial modal coefficient vector. Based on the reduced state transition equation, the reduced observation equation, and the original output signal collected in real time by the instrument, the posterior estimate of the modal coefficient vector at the current time is obtained by Kalman filtering recursively. By using a reduced basis, the posterior estimate of the modal coefficient vector at the current moment is mapped to the original high-dimensional space of the full-field state dataset, thus obtaining the optimal estimate of the latent state variable at the current moment.

[0011] Alternatively, the reduced-order extended Kalman filter algorithm can be expressed by the following equation: ; in, for The prior estimate of the state at time step [time]. Here is the state transition matrix. It is the transpose of the state transition matrix. for The posterior estimate of the state at time step [time]. To control the input matrix, for Constantly control the amount of input. for Prior estimation of the covariance matrix of the state at each time step. for The covariance matrix of the posterior estimate of the state at time step is... The process noise covariance matrix is... for Moment Kalman gain, For the observation matrix, Let be the transpose of the observation matrix. To observe the noise covariance matrix, for k The posterior estimate of the state at time step [time]. for Real-time observation data collected by instruments and meters. It is an identity matrix.

[0012] Optionally, the optimal estimate of the hidden state variable is decomposed into fast and slow components to obtain fast and slow components; based on the fast and slow components, drift compensation is performed on the raw output signal acquired in real time by the instrument under test to obtain a compensated output signal, specifically including: Based on the optimal estimate of the hidden state variables, the fast-changing component of the optimal estimate is extracted by a high-pass filter, and the slow-changing component of the optimal estimate is extracted by a low-pass filter. Based on the material thermophysical properties of the sensitive element of the instrument under test, a mapping relationship between the slowly varying component and the slowly varying drift is constructed to obtain the slowly varying mapping relationship; based on the dynamic response characteristics of the sensitive element of the instrument under test, a mapping relationship between the rapidly varying component and the rapidly varying drift is constructed to obtain the rapidly varying mapping relationship. Based on the fast-changing mapping relationship, the fast-changing component drift value at the current moment is obtained by mapping the fast-changing components; based on the slow-changing mapping relationship, the slow-changing component drift value at the current moment is obtained by mapping the slow-changing components. The fast-changing component drift value and the slow-changing component drift value are superimposed to obtain the total zero-point drift and the total sensitivity drift. The zero-point reference value and sensitivity reference value of the instrument under test are obtained under standard conditions. The total zero-point drift and the zero-point reference value are added together to obtain the zero-point compensation value. The total sensitivity drift and the sensitivity reference value are added together to obtain the sensitivity compensation value. Based on the zero-point compensation value and the sensitivity compensation value, the original output signal of the instrument under test is compensated in real time to obtain the compensated output signal.

[0013] Optionally, the comparison deviation is used as innovation to invert the boundary condition parameters of the multiphysics coupling model using an ensemble Kalman filter algorithm, thereby obtaining the optimal estimate of the boundary condition parameters, specifically including: The current compensated output signal is compared with the original output signal to determine the comparison deviation between the compensated output signal and the original output signal, and the comparison deviation is used as the information. Vectorize the boundary condition parameters of the multiphysics coupling model to obtain the initial set; Based on the new information, each member in the initial set is updated using the ensemble Kalman filter algorithm to obtain the updated boundary condition parameter set, and the average value of the updated boundary condition parameter set is used as the optimal estimate of the boundary condition parameters at the current time.

[0014] Optionally, the multiphysics coupling model is corrected online using the optimal estimate of the boundary condition parameters to obtain an updated full-field state dataset. Based on this updated full-field state dataset, the dynamic response characteristics of the instruments are determined, specifically including: The optimal estimate based on the boundary condition parameters is used as the new boundary condition input to the multiphysics coupling model to resolve the control equations of the multiphysics coupling model and obtain an updated full-field state dataset. Based on the updated full-field state dataset, the optimal estimate and the compensation output signal are updated until the deviation between the compensation output signal and the original output signal meets the preset threshold, and the updated compensation output signal is obtained. The dynamic response characteristics of the instrument under test are determined based on the updated compensated output signal.

[0015] This invention provides a simulation testing system for instruments under multi-physics coupling, comprising: The data acquisition module is used to acquire various physical fields that interact and influence each other in the working environment of the instrument under test. The multiphysics coupling simulation module is used to divide multiple physical fields into fast-changing field subdomains and slow-changing field subdomains based on a time scale threshold. The coupling region between the fast-changing field subdomains and the slow-changing field subdomains is fitted by a pre-trained physical information neural network to obtain a multiphysics coupling model. The control equations of the multiphysics coupling model are solved asynchronously to obtain a full-field state dataset. The hidden state assimilation module is used to assimilate the hidden state variables of the instrument under test based on the multi-physics coupling model and the full-field state dataset, and obtain the optimal estimate of the hidden state variables. The dynamic compensation module is used to decompose the optimal estimate of the hidden state variable into fast and slow components to obtain fast and slow components; based on the fast and slow components, the original output signal acquired in real time by the instrument under test is drift compensated to obtain the compensated output signal. The boundary condition correction module is used to compare the compensated output signal with the reference measurement value representing the real physical quantity to obtain the comparison deviation; the comparison deviation is used as information to invert the boundary condition parameters of the multiphysics coupling model through the ensemble Kalman filter algorithm to obtain the optimal estimate of the boundary condition parameters; The evaluation module is used to perform online closed-loop correction of the multiphysics coupling model based on the optimal estimation of boundary condition parameters, obtain an updated full-field state dataset, and determine the dynamic response characteristics of the instruments based on the updated full-field state dataset.

[0016] The simulation testing method and system for instruments under multi-physics coupling provided in this invention have the following advantages compared with the prior art: This invention utilizes a physical information neural network to fit the coupling region between fast-changing and slow-changing field subdomains. It automatically learns the strongly nonlinear, multi-scale interaction laws between fast and slow variables at the coupling interface from the data, thereby constructing a continuous and non-disjoint multiphysics coupling model. Compared to traditional methods based on fixed-boundary internal and external iteration strategies, this technique enables the multiphysics coupling model to dynamically characterize the transmission properties of the coupling interface. This provides a reliable model foundation for real-time dynamic estimation of hidden state variables during instrument operation, significantly improving the inversion accuracy of real dynamic response characteristics under complex environments. Attached Figure Description

[0017] Figure 1This is a schematic diagram illustrating the application environment of a simulation testing method for instruments under multi-physics coupling provided in one embodiment; Figure 2 This is a flowchart illustrating a simulation testing method for instruments under multi-physics coupling, as provided in one embodiment. Figure 3 This is a schematic diagram illustrating the process of generating a full-field state dataset for a simulation test method of instruments under multi-physics coupling, as provided in one embodiment. Figure 4 This is a schematic diagram of the compensation output signal generation process of a simulation test method for instruments under multi-physics coupling provided in one embodiment; Figure 5 This is a schematic diagram of the structure of a simulation test system for instruments under multi-physics coupling, provided in one embodiment.

[0018] Among them, 101 is a server; 102 is a terminal; 500 is a simulation test system for instruments under multi-physics coupling; 501 is a multi-physics coupling simulation module; 502 is a hidden state assimilation module; 503 is a dynamic compensation module; and 504 is a boundary condition correction and evaluation module. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0020] This invention provides a simulation testing method for instruments under multi-physics coupling, the method comprising: To obtain the various physical fields that interact and influence each other in the working environment of the instrument under test.

[0021] Based on a time-scale threshold, multiple physical fields are divided into fast-varying and slow-varying subdomains. A pre-trained physical information neural network is used to fit the coupling region between the fast-varying and slow-varying subdomains to obtain a multi-physics coupling model. The governing equations of the multi-physics coupling model are solved asynchronously to obtain a full-field state dataset.

[0022] Based on a multiphysics coupling model and a full-field state dataset, the hidden state variables of the instrument under test are assimilated using a reduced-order extended Kalman filter algorithm to obtain the optimal estimates of the hidden state variables.

[0023] The optimal estimate of the hidden state variable is decomposed into fast and slow components to obtain fast and slow components. Drift compensation is performed on the original output signal acquired in real time by the instrument under test based on the fast and slow components to obtain the compensated output signal.

[0024] The compensated output signal is compared with the reference measurement value representing the real physical quantity to obtain the comparison deviation. Using this comparison deviation as information, the boundary condition parameters of the multiphysics coupling model are inverted using an ensemble Kalman filter algorithm to obtain the optimal estimate of the boundary condition parameters.

[0025] The optimal estimation of boundary condition parameters is used to perform online closed-loop correction of the multiphysics coupling model, resulting in an updated full-field state dataset. Based on the updated full-field state dataset, the dynamic response characteristics of the instruments are determined.

[0026] In one exemplary embodiment, an application is provided for such Figure 1 The application environment shown is a simulation test method for instruments under multi-physics coupling. The application environment includes: server 101 and terminal 102.

[0027] Terminal 102 can communicate with server 101 via a network. The data storage system can store the data that server 101 needs to process. The data storage system can be integrated into server 101 or located in the cloud or on other network servers. Terminal 102 can be used to collect multiphysics parameters from the instrument's working environment and real-time observation data from the instrument, and upload it to server 101. Server 101 divides the multiphysics parameters into fast-changing and slow-changing subdomains, constructs a multiphysics coupling model, and asynchronously solves it to generate a full-field state dataset. It then completes the assimilation of hidden state variable data, real-time drift compensation, boundary condition parameter inversion, and online closed-loop correction of the multiphysics coupling model, ultimately determining the dynamic response characteristics of the instrument. Terminal 102 can be, but is not limited to, various data acquisition terminals, industrial monitoring terminals, tablet computers, portable testing equipment, and IoT monitoring equipment. Server 101 can be implemented using a standalone server or a server cluster composed of multiple servers.

[0028] In one exemplary embodiment, such as Figure 2 As shown, a simulation test method for instruments under multi-physics coupling is provided, taking server 101 as an example for illustration, including the following steps S1 to S4: Step S1: Obtain the multiphysics field of the instrument. Based on a time scale threshold, divide the multiphysics field into a fast-changing field subdomain and a slow-changing field subdomain. Construct a multiphysics coupling model by setting up a physical information neural network in the coupling region of the fast-changing field subdomain and the slow-changing field subdomain. Asynchronously solve the control equations of the multiphysics coupling model to generate a full-field state dataset.

[0029] Specifically, multiphysics refers to multiple physical fields that interact and influence each other in the working environment of instruments, including temperature fields, electromagnetic fields, stress fields, and flow fields. The time scale threshold refers to the time standard used to distinguish the rates of change of different physical fields. A rapidly changing field subdomain refers to a physical field subdomain with a relatively fast rate of change and sensitive time response, with a rate of change less than 0.01 seconds. A slowly changing field subdomain refers to a physical field subdomain with a relatively slow rate of change and a smooth time response, with a rate of change greater than 10 seconds. The coupling region refers to the region where the rapidly changing and slowly changing field subdomains overlap and interact. Asynchronous coupling solution refers to a solution method that uses different time steps to solve the rapidly changing and slowly changing field subdomains separately and achieves load transfer through the coupling region. The formula for asynchronous coupling solution can be:

[0030] In the formula, For the time step of the slowly varying field subdomain, To increase the time step of the field subdomain, The step size ratio is a positive integer multiple.

[0031] Optionally, server 101 can acquire multiphysics parameters from the instrument's operating environment via a terminal to obtain the instrument's multiphysics field. Server 101 can divide the multiphysics field into fast-changing and slow-changing subdomains based on a time-scale threshold, determining the coupling region between these subdomains. Server 101 can set up a physical information neural network in the coupling region, constructing a multiphysics coupling model by combining the physical characteristics of the fast-changing and slow-changing subdomains. Server 101 can solve the fast-changing subdomain using an explicit time integration method and the slow-changing subdomain using an implicit time integration method. Server 101 can use the physical information neural network to achieve load transfer in the coupling region, asynchronously couple the control equations of the multiphysics coupling model, and record the fast-changing time series and slow-changing state variables at each moment. Server 101 can integrate the fast-changing time series and slow-changing state variables to generate a full-field state dataset. Among them, the time scale threshold can refer to the critical time value used to determine whether a physical field belongs to a fast-changing field subdomain or a slow-changing field subdomain. The value of the time scale threshold can be set based on the working characteristics of the instrument, the inherent rate of change of the multi-physics field, and the simulation accuracy requirements. The time scale threshold can be used to quantitatively distinguish the rate of change of different physical fields.

[0032] Step S2: Based on the multiphysics coupling model and the full-field state dataset, the unmeasurable hidden state variables inside the instrument are assimilated by the reduced-order extended Kalman filter algorithm to obtain the optimal estimate of the hidden state variables.

[0033] Specifically, the order-reduced extended Kalman filter algorithm can refer to a method that first reduces the order of the entire field state data, and then combines it with the extended Kalman filter algorithm to achieve data assimilation. The recursive formula for the order-reduced extended Kalman filter algorithm can be: ; in, for The prior estimate of the state at time step [time]. Here is the state transition matrix. It is the transpose of the state transition matrix. for The posterior estimate of the state at time step [time]. To control the input matrix, for Constantly control the amount of input. for Prior estimation of the covariance matrix of the state at each time step. for The covariance matrix of the posterior estimate of the state at time step is... The process noise covariance matrix is... for Moment Kalman gain, For the observation matrix, Let be the transpose of the observation matrix. To observe the noise covariance matrix, for k The posterior estimate of the state at time step [time]. for Real-time observation data collected by instruments and meters. It is an identity matrix.

[0034] Impermanent state variables that are not measurable within an instrument can refer to state parameters that cannot be obtained through direct measurement but affect the dynamic response characteristics of the instrument. Data assimilation can refer to the process of fusing simulation data from a multiphysics coupled model with observation data from the instrument to correct model biases and improve the accuracy of state estimation.

[0035] Optionally, server 101 establishes state transition equations and observation equations based on the control equations of the multiphysics coupling model. It performs intrinsic orthogonal decomposition on the state vectors in the full-field state dataset to obtain a reduced-order basis. The state transition equations and observation equations are projected onto the reduced-order space spanned by the reduced-order basis to obtain reduced-order state transition equations and reduced-order observation equations. Server 101 can use the projection of the state vectors onto the reduced-order basis as the initial modal coefficient vector. Based on the reduced-order state transition equations, reduced-order observation equations, and real-time observation data collected by the instruments, it recursively obtains the posterior estimate of the modal coefficient vector at the current moment using a Kalman filter algorithm. Server 101 can reconstruct the posterior estimate into a full-space state vector to obtain the optimal estimate of the unmeasurable hidden state variables within the instruments at the current moment. The intrinsic orthogonal decomposition includes eigenvalue solving and eigenvector selection.

[0036] Step S3: Decompose the optimal estimate into fast and slow components to obtain fast and slow components; determine the fast component drift and slow component drift based on the fast and slow components respectively; superimpose the fast and slow component drifts to obtain the real-time compensation values ​​for zero drift and sensitivity drift; and use the real-time compensation values ​​to compensate the original output signal acquired by the instrument to generate a compensated output signal.

[0037] Specifically, fast-slow decomposition refers to the process of separating components with different rates of change in the optimal estimate using filtering, and extracting the fast-changing and slow-changing characteristics separately. Fast-changing component drift refers to the amount of change in the instrument's output value due to the fast-changing component, while slow-changing component drift refers to the amount of change in the instrument's output value due to the slow-changing component. Zero-point drift refers to the amount of change in the instrument's output value due to the absence of an input signal, and sensitivity drift refers to the amount of change in the instrument's sensitivity due to changes in the operating environment, deviating from the sensitivity reference value.

[0038] Optionally, server 101 can extract the fast-changing component of the optimal estimate using a high-pass filter and the slow-changing component using a low-pass filter. Server 101 can establish a mapping relationship between the slow-changing component and its drift based on the thermophysical properties of the instrument's sensitive element, and can also establish a mapping relationship between the fast-changing component and its drift based on the dynamic response characteristics of the instrument's sensitive element. Based on the established mapping relationships, server 101 can obtain the current fast-changing component drift value and slow-changing component drift value by mapping the fast-changing component and slow-changing component respectively, and then superimpose the fast-changing component drift value and slow-changing component drift value to obtain the total zero-point drift and total sensitivity drift. Server 101 can acquire the zero-point reference value and sensitivity reference value of the instrument under standard conditions. It adds the total zero-point drift to the zero-point reference value to obtain the zero-point compensation value, and adds the total sensitivity drift to the sensitivity reference value to obtain the sensitivity compensation value. Server 101 can then compensate the original output signal acquired by the instrument based on the zero-point compensation value and the sensitivity compensation value, generating a compensated output signal (the compensated output signal is the original output signal plus the real-time compensation value; the deviation obtained by comparing it with the original output signal is the real-time compensation value, which is compared with the reference measurement value to reflect the difference between the model's predicted value and the actual value, used to correct the model). The cutoff frequencies of the high-pass filter and low-pass filter are set based on a time-scale threshold.

[0039] Step S4 compares the compensated output signal with the reference measurement value representing the real physical quantity to obtain the comparison deviation. Using this deviation as information, the boundary condition parameters of the multiphysics coupling model are inverted using an ensemble Kalman filter algorithm to obtain the optimal estimate of the boundary condition parameters. Based on the optimal estimate, the multiphysics coupling model is corrected online in a closed-loop manner to obtain an updated full-field state dataset. The dynamic response characteristics of the instrument are then determined based on this updated full-field state dataset.

[0040] Specifically, "new information" can refer to the comparison deviation used to correct the model parameters and is the basis for updating the ensemble Kalman filter algorithm; the ensemble Kalman filter algorithm can refer to the filtering algorithm that realizes parameter inversion and state update based on ensemble sampling; boundary condition parameters can refer to the key parameters that constrain the solution range and external effects of the multiphysics coupling model; and the benchmark measurement can refer to the high-precision reference output that represents the real physical quantity, which can be used to verify and correct the multiphysics coupling model.

[0041] Optionally, server 101 can obtain benchmark measurements representing real physical quantities from a standard metrology platform via a network. Server 101 can numerically compare the current compensated output signal with the benchmark measurements, calculate the comparison deviation between the compensated output signal and the benchmark measurements, and use this comparison deviation as the information for the ensemble Kalman filter algorithm. Server 101 can vectorize the boundary condition parameters of the multiphysics coupling model to generate an initial set. Server 101 can update each member in the initial set based on the information using the ensemble Kalman filter algorithm to obtain an updated boundary condition parameter set. Server 101 can calculate the average value of the updated boundary condition parameter set and use this average value as the optimal estimate of the boundary condition parameters at the current time. Server 101 can use the optimal estimate of the boundary condition parameters as new boundary conditions to input into the multiphysics coupling model, resolve the control equations of the multiphysics coupling model, and obtain an updated full-field state dataset. Server 101 can repeat steps S2 to S3 based on the updated full-field state dataset to update the optimal estimates of the hidden state variables and the compensated output signal until the deviation between the compensated output signal and the benchmark measurements meets a preset threshold. Server 101 can calculate the dynamic response characteristics of the instrument based on the final updated compensation output signal.

[0042] The above-mentioned simulation and testing method for instruments under multi-physics coupling effectively improves the accuracy and efficiency of multi-physics coupling simulation by using asynchronous coupling solution of fast and slow variable fields, physical information neural network modeling, hidden state data assimilation, real-time zero-point drift compensation, real-time sensitivity drift compensation, and online closed-loop correction of boundary conditions. It accurately estimates unmeasurable hidden state variables, reduces drift errors, enhances model adaptability and test reliability, and thus comprehensively solves the technical problems of low accuracy, poor efficiency, insufficient compensation, and difficulty in online correction in traditional simulation and testing.

[0043] In one exemplary embodiment, such as Figure 3 As shown, a schematic diagram of the full-field state dataset generation process is provided for application. Figure 1 Taking the server in the example, the following steps S11 to S14 are used for illustration: Step S11: Based on the working environment of the instrument, determine the multi-physics field of the instrument, decompose the multi-physics field based on the time scale threshold, divide the multi-physics field into fast-changing field subdomain and slow-changing field subdomain, and take the overlapping area of ​​the fast-changing field subdomain and the slow-changing field subdomain as the coupling region, set up the physical information neural network in the coupling region, and construct the multi-physics field coupling model.

[0044] Specifically, physical information neural networks can refer to neural network models that integrate physical control equation constraints and have the ability to fit and map physical fields. Physical information neural networks can achieve accurate representation of multi-physics interaction behavior and load transfer within the coupled region through joint training of data-driven fitting and physical equation constraints.

[0045] Optionally, the server can collect the operating environment parameters of the instruments and meters, identify and determine the multiphysics field type and field quantity information corresponding to the instruments and meters under the current operating conditions based on the operating environment parameters; the server can judge the rate of change of each physical field based on the time scale threshold, classify physical fields with a rate of change higher than the time scale threshold into fast-changing field subdomains, and classify physical fields with a rate of change lower than the time scale threshold into slow-changing field subdomains; the server can determine the overlapping area of ​​the fast-changing field subdomains and slow-changing field subdomains through spatial coordinate matching, and mark the overlapping area as the coupling area; the server can embed a physical information neural network within the coupling area, and combine the control equations and boundary conditions of the fast-changing field subdomains and slow-changing field subdomains to complete the construction of the multiphysics coupling model.

[0046] Step S12: Solve the fast-changing field subdomain using the explicit time integration method to generate the fast-changing field time series, and solve the slow-changing field subdomain using the implicit time integration method to generate the slow-changing field state variables.

[0047] Specifically, the explicit time integration method refers to a numerical solution method that directly calculates the physical field state variables at the next time step based on the current physical field state variables using an explicit recursive formula. The calculation formula for the explicit time integration method can be: ; In the formula, For the first The physical field state quantities of a rapidly changing field subdomain. For the first The physical field state quantities of a rapidly changing field subdomain. To determine the solution time step for the rapidly varying field subdomain, For the first The time value at any given moment For the first Evolution operator of physical field state quantities at time step The evolution operator formula can be: ; Implicit time integration refers to a numerical solution method that establishes implicit equations based on the physical field state variables at the current and next time steps, and obtains the physical field state variables at the next time step through iterative solutions. The calculation formula for implicit time integration is as follows: ; In the formula, For the first The physical field state quantities of a time-varying slow field subdomain. For the first The physical field state variables to be solved in a time-varying slow field subdomain. Let be the solution time step for the slowly varying field subdomain. For the first The time value at any given moment For the first Evolution operator of physical field state quantities at time step The evolution operator formula can be: ; in, For the physical field diffusion term, The diffusion coefficient of the medium, For gradient operators, This is the physical field source term.

[0048] Optionally, the server can extract the control equations and initial state parameters of the rapidly changing field subdomain in the multiphysics coupling model, and use the explicit time integration method to recursively solve the rapidly changing field subdomain time-by-time, recording the rapidly changing field state variables at each time step, and organizing them in chronological order to generate a rapidly changing field time series; the server can also extract the control equations and initial state parameters of the slowly changing field subdomain in the multiphysics coupling model, and use the implicit time integration method to solve the slowly changing field subdomain time-by-time, eliminating numerical errors through iterative calculations to obtain the slowly changing field state variables at each time step.

[0049] Step S13: Fit the fast-changing field time series in the coupling region using a physical information neural network, construct the network loss function, and train the physical information neural network until convergence; call the trained physical information neural network to calculate the precise value of the slow-changing field subdomain at the current solution time point, and pass the precise value as an equivalent load to the control equation of the slow-changing field subdomain to generate the updated slow-changing field state variables.

[0050] Specifically, the equivalent load can refer to the external excitation that transforms the precise field quantities of the coupling region into the governing equations of the slowly varying field subdomain; the governing equations of the slowly varying field subdomain can refer to the second-order constant-coefficient partial differential equations that describe the variation law of the physical quantities in the slowly varying field subdomain.

[0051] Optionally, the server can input the fast-changing field time series as training samples into the physical information neural network within the coupling region. The server can construct a network loss function that includes data fitting loss and physical control equation constraint loss, and use an adaptive optimization algorithm to iteratively train the physical information neural network until the network loss function converges. The server can call the trained physical information neural network to infer and calculate the physical field quantities corresponding to the slow-changing field subdomain at the current solution time point to obtain accurate values ​​that satisfy the physical constraints. The server can convert these accurate values ​​into equivalent loads and substitute them into the control equations of the slow-changing field subdomain to resolve the slow-changing field subdomain, generating updated slow-changing field state quantities with higher accuracy.

[0052] Step S14: Perform eigenorthogonal decomposition on the slow-changing field state variables, extract time coefficients, and interpolate the time coefficients to obtain a continuously changing function. Based on the continuously changing function, calculate the slow-changing field boundary conditions corresponding to each time step of the fast-changing field subdomain and pass them to the control equations of the fast-changing field subdomain to generate an updated fast-changing field time series.

[0053] Specifically, intrinsic orthogonal decomposition can refer to a data analysis method used to extract the main characteristic information in the state variables of a slowly varying field. By performing eigenvalue decomposition on the state variable matrix of the slowly varying field, orthogonal characteristic modes and time coefficients are obtained. The governing equations of a rapidly varying field subdomain can refer to first-order constant-coefficient partial differential equations that describe the variation law of physical quantities in a rapidly varying field subdomain.

[0054] Optionally, the server can organize the slowly varying field state variables at different times into a state matrix, perform eigenorthogonal decomposition on the state matrix, solve for eigenvalues ​​and eigenvectors, and select eigenvectors whose cumulative contribution rate meets a preset threshold as spatial modes and extract time coefficient sequences. The server can use cubic spline interpolation to interpolate the discrete time coefficient sequences, eliminate jumps in discrete data, and obtain continuously varying functions. The server can traverse the continuously varying functions according to the solution time step of the rapidly varying field subdomain, calculate the time coefficients corresponding to each time step, and obtain the slowly varying field boundary conditions corresponding to that time step by combining spatial mode inversion. The server can pass the slowly varying field boundary conditions corresponding to each time step to the control equations of the rapidly varying field subdomain, call the explicit time integration method to re-solve the rapidly varying field subdomain, record the rapidly varying field state variables at each time step, and organize them to generate an updated rapidly varying field time series. The preset threshold can refer to the critical value of the cumulative contribution rate used to select effective spatial modes, which is preset to 95% or 99% according to the simulation accuracy requirements.

[0055] Step S15: Repeat steps S12 to S14 until the solution converges. Integrate the fast-changing field time series and slow-changing field state variables at all times to generate a full-field state dataset.

[0056] Optionally, the server can set a convergence threshold and repeatedly execute the asynchronous coupled solution process from S12 to S14. After each iteration, the server compares the numerical deviations of the fast-changing field time series and slow-changing field state variables between the current iteration and the previous iteration to determine if the deviation is less than the preset convergence threshold. If the deviation does not meet the convergence condition, the iterative calculation continues; if the deviation meets the convergence condition, the iteration stops, and the solution is considered converged. The server can extract the fast-changing field time series and slow-changing field state variables at all times after iterative convergence, and perform unified integration, standardization, and associated storage of the fast-changing field time series and slow-changing field state variables at all times according to the chronological order and spatial coordinate information to generate a full-field state dataset. The convergence threshold is adaptively set according to the instrument type, simulation accuracy requirements, and computational efficiency requirements.

[0057] In one exemplary embodiment, the formula for calculating the network loss function includes: ; in, The output of the physical information neural network, These are the training parameters for the physical information neural network. For time, The number of solution steps for a rapidly varying field subdomain within a single slowly varying field subdomain. Let this be the current solution time point for the slowly varying field subdomain. To determine the solution time step for the rapidly varying field subdomain, For fast-changing field time series, To balance the coefficients of the data fitting term and the physical constraint term, For the physical governing equations satisfied by the rapidly varying field subdomain, In for .

[0058] Specifically, The calculation formula can be: ); In the formula, Indicates The physical information of the network parameters is mapped to the neural network. Time is the independent variable. The vector of parameters to be trained consists of network weights and biases.

[0059] The calculation formula can be: ; In the formula, For fast-changing field diffusion terms, The diffusion coefficient of the medium, For gradient operators, It consists of the fast-changing field source term, external excitation, and multi-physics coupling term.

[0060] In one exemplary embodiment, a method for generating optimal estimates of hidden state variables is provided, which is then applied to... Figure 1 Taking the server in the example, the explanation includes the following steps S21 to S24. Wherein:

[0061] Step S21: Based on the control equations of the multiphysics coupling model, establish the state transition equations and observation equations.

[0062] Specifically, the state transition equation can refer to the mathematical equation obtained by discretizing the control equation based on the multiphysics coupling model, which describes the recursive evolution relationship of the hidden state variables of the system over time; the observation equation can refer to the mathematical equation that describes the mapping relationship between the hidden state variables of the system and the observation data collected in real time by instruments.

[0063] Optionally, the server can extract the governing equations of the multiphysics coupling model, and use the finite difference method to discretize the governing equations in time and space, transforming the continuous physical field evolution relationship into a discrete-time recursive form, and establishing state transition equations. The server can also establish a mapping relationship between hidden state variables and the real-time observation data acquired by the instruments based on the actual measurement location, the type of measured physical quantity, and the measurement principle, thereby generating observation equations.

[0064] Step S22: Perform eigenorthogonal decomposition on the state vectors in the full-field state dataset to obtain a reduced-order basis. Project the state transition equation and observation equation onto the reduced-order space spanned by the reduced-order basis to obtain the reduced-order state transition equation and the reduced-order observation equation.

[0065] Specifically, a reduced basis can refer to a basis matrix composed of orthogonal eigenvectors whose cumulative contribution rate satisfies a preset threshold obtained from eigenorthogonal decomposition; a reduced space can refer to a low-dimensional linear space spanned by the column vectors of the reduced basis.

[0066] Optionally, the server can extract state vectors from the full-field state dataset to construct a state vector set matrix. The server can then perform eigenorthogonal decomposition on this matrix and filter by cumulative contribution rate to obtain a reduced-order basis. The server can project the state transition equation and observation equation onto the reduced-order space spanned by the reduced-order basis, respectively, and perform matrix operations to reduce the dimensionality of the equations, obtaining the reduced-order state transition equation and reduced-order observation equation. The calculation formulas for the reduced-order state transition equation and reduced-order observation equation can be:

[0067] ; In the formula, This is the reduced-order state vector at the current moment; for The reduced-order state vector at time t, This is a reduced-order basis obtained from eigenorthogonal decomposition; This is the state transition matrix; For observation vectors; The observation matrix; This is process noise; To observe noise.

[0068] Step S23: The projection of the state vector onto the reduced basis is used as the initial modal coefficient vector. Based on the reduced state transition equation, the reduced observation equation, and the observation data collected in real time by the instruments, the posterior estimate of the modal coefficient vector at the current time is obtained by recursion through the Kalman filter algorithm.

[0069] Optionally, the server can project the initial state vector onto a reduced-order basis to obtain the initial mode coefficient vector. The server can then use a Kalman filter algorithm to predict and iteratively update the mode coefficient vector based on the reduced-order state transition equation and the reduced-order observation equation. The server can also combine real-time observation data from instruments to complete filtering corrections and output the posterior estimate of the mode coefficient vector at the current time.

[0070] Step S24 reconstructs the posterior estimate into a full-space state vector to obtain the optimal estimate of the hidden state variable at the current time.

[0071] Specifically, reconstruction can refer to the computational process of mapping low-dimensional modal coefficient vectors back to the original high-dimensional space using a reduced-order basis.

[0072] Optionally, the server can perform matrix multiplication on the posterior estimate and the reduced basis to reconstruct the full-space state vector. The server can then use the reconstructed full-space state vector as the optimal estimate of the latent state variables at the current time step.

[0073] In one exemplary embodiment, such as Figure 4 As shown, a schematic diagram of the compensation output signal generation process is provided for application. Figure 1 Taking the server in the example, the explanation includes the following steps S31 to S34. Wherein:

[0074] Step S31: Based on the optimal estimate, extract the fast-changing component of the optimal estimate through a high-pass filter and extract the slow-changing component of the optimal estimate through a low-pass filter.

[0075] Specifically, a high-pass filter can refer to a filtering system that allows high-frequency signals to pass through and suppresses low-frequency signals, used to separate signal components in the optimal estimate that fluctuate rapidly over time; a low-pass filter can refer to a filtering system that allows low-frequency signals to pass through and suppresses high-frequency signals, used to separate signal components in the optimal estimate that change gradually over time.

[0076] Optionally, the server can invoke high-pass and low-pass filters, inputting the optimal estimates of the hidden state variables at the current time into the high-pass and low-pass filters. The server can use the high-pass filter to filter out low-frequency interference components in the optimal estimate and extract the fast-changing components. The server can use the low-pass filter to filter out high-frequency interference components in the optimal estimate and extract the slow-changing components.

[0077] Step S32: Based on the material thermophysical properties of the sensitive element of the instrument, establish the mapping relationship between the slow-changing component and the slow-changing drift to obtain the slow-changing mapping relationship; based on the dynamic response characteristics of the sensitive element of the instrument, establish the mapping relationship between the fast-changing component and the fast-changing drift to obtain the fast-changing mapping relationship.

[0078] Specifically, slow drift refers to the deviation of the instrument's output signal that changes slowly over time; fast drift refers to the deviation of the instrument's output signal that fluctuates rapidly over time.

[0079] Optionally, the server can acquire the material thermophysical properties and dynamic response characteristics of the sensitive elements of the instrument, collect experimental data on slow and fast drift under different operating conditions, and use the least squares method to fit the initial mapping relationships between the slow components and slow drift, and between the fast components and fast drift, respectively, based on the experimental data. The server can perform error verification on the initial mapping relationships, adjust the parameters of the mapping relationships to reduce errors, and obtain the slow and fast mapping relationships.

[0080] Step S33: Based on the fast-changing mapping relationship, obtain the fast-changing component drift value at the current moment according to the fast-changing component mapping; based on the slow-changing mapping relationship, obtain the slow-changing component drift value at the current moment according to the slow-changing component mapping; superimpose the fast-changing component drift value and the slow-changing component drift value to obtain the total zero-point drift and the total sensitivity drift.

[0081] Optionally, the server can input the fast-changing component at the current moment into the fast-changing mapping relationship to calculate the fast-changing component drift value at the current moment. The server can also input the slow-changing component at the current moment into the slow-changing mapping relationship to calculate the slow-changing component drift value at the current moment. The server can then perform a superposition operation on the fast-changing component drift value and the slow-changing component drift value to obtain the total zero-point drift and total sensitivity drift of the instrument at the current moment, respectively.

[0082] Step S34: Obtain the zero-point reference value and sensitivity reference value of the instrument under standard conditions. Add the total zero-point drift and the zero-point reference value to obtain the zero-point compensation value. Add the total sensitivity drift and the sensitivity reference value to obtain the sensitivity compensation value. Compensate the original output signal based on the zero-point compensation value and the sensitivity compensation value to generate a compensated output signal.

[0083] Specifically, the zero-point reference value can refer to the output reference signal value of an instrument when the input is zero under standard conditions; the sensitivity reference value can refer to the change in output signal corresponding to a unit input physical quantity of an instrument under standard conditions.

[0084] Optionally, the server can retrieve the zero-point reference value and sensitivity reference value of the instrument under standard conditions. The server can add the total zero-point drift and the zero-point reference value to obtain the zero-point compensation value. The server can also add the total sensitivity drift and the sensitivity reference value to obtain the sensitivity compensation value. The server can then correct the zero-point deviation of the original output signal based on the zero-point compensation value and correct the sensitivity deviation of the original output signal based on the sensitivity compensation value, thereby generating a compensated output signal.

[0085] In one exemplary embodiment, a method for generating dynamic response characteristic indicators is provided, which is then applied to... Figure 1 Taking the server in the example, the explanation includes the following steps S41 to S45. Wherein:

[0086] Step S41 compares the current compensated output signal with the reference measurement value, calculates the comparison deviation between the compensated output signal and the observed data, and uses the comparison deviation as information.

[0087] Optionally, the server can obtain the current compensated output signal and the reference measurement value, perform a difference operation on the compensated output signal and the reference measurement value, calculate the comparison deviation between the compensated output signal and the reference measurement value, and the server can use the calculated comparison deviation as information.

[0088] Step S42: Vectorize the boundary condition parameters to generate an initial set; based on the information, update each member in the initial set using the ensemble Kalman filter algorithm to obtain the updated boundary condition parameter set; calculate the average value of the updated boundary condition parameter set and use the average value as the optimal estimate of the boundary condition parameters at the current time.

[0089] Specifically, boundary condition parameters can refer to the inherent constraint parameters that instruments and meters must follow during operation, including the operating temperature threshold, operating voltage range, operating air pressure range, upper limit of input signal range, lower limit of input signal range, and output signal threshold of the sensitive elements of the instruments and meters.

[0090] Optionally, the server can vectorize the boundary condition parameters to generate multiple boundary condition parameter vectors, forming an initial set. Based on the information, the server can update each parameter vector member in the initial set one by one using an ensemble Kalman filter algorithm to obtain an updated boundary condition parameter set. The server can calculate the average value of all parameter vector members in the updated boundary condition parameter set and use this average value as the optimal estimate of the boundary condition parameters at the current time.

[0091] Step S43: Input the optimal estimate as a new boundary condition into the multiphysics coupling model, resolve the control equations of the multiphysics coupling model, and obtain the updated full-field state dataset.

[0092] Optionally, the server can input the optimal estimate as a new boundary condition into the multiphysics coupling model; the server can call a numerical solution algorithm to re-solve the governing equations of the multiphysics coupling model. The server can extract the full-field state data output by the multiphysics coupling model and generate an updated full-field state dataset. The numerical solution algorithm uses the finite difference method for discrete solution, and the core recursive formula can be:

[0093] ; In the formula, For the first Time grid nodes The physical field variables, No. Time grid nodes The physical field variables, For time step, For gradient operators, For the discrete operators of the multiphysics coupling control equations, The calculation formula can be: ; In the formula, For the discrete terms of physical field diffusion, The diffusion coefficient of the medium, This is a multiphysics coupling source term.

[0094] Step S44: Based on the updated full-field state dataset, repeat steps S2 to S3 to update the optimal estimate and the compensation output signal until the deviation between the compensation output signal and the benchmark measurement meets the preset threshold.

[0095] Specifically, the preset threshold can refer to the critical deviation value used to determine whether the compensation accuracy meets the standard. The preset threshold can be combined with the instrument's factory accuracy level, actual working scenario requirements, and allowable measurement error range. The maximum allowable measurement error under the instrument's standard conditions is used as the benchmark, and 50%-80% of the benchmark value is taken as the preset threshold. At the same time, the threshold must not be greater than the minimum response threshold of the instrument's sensitive element.

[0096] Optionally, the server can repeatedly execute the optimal estimation process of S2 based on the updated full-field state dataset to obtain the updated optimal estimates of the hidden state variables. The server can then repeatedly execute the drift compensation process of S3 to obtain the updated compensated output signal. The server can calculate the deviation between the updated compensated output signal and the benchmark measurement value, determine whether the deviation meets a preset threshold, and if not, continue the loop update; if it meets the threshold, stop the loop and output the final compensated output signal.

[0097] Step S45: Calculate the dynamic response characteristics of the instrument based on the updated compensated output signal.

[0098] Optionally, the server can extract the time-domain characteristic parameters of the compensated output signal. Based on the time-domain characteristic parameters, the server can calculate the four dynamic response characteristic indicators of the instrument: rise time, settling time, peak time, and overshoot.

[0099] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0100] Based on the same inventive concept, this application also provides a simulation testing system for instruments under multi-physics coupling to implement the above-mentioned simulation testing method for instruments under multi-physics coupling. The system includes: The data acquisition module is used to acquire various physical fields that interact and influence each other in the working environment of the instrument under test.

[0101] The multiphysics coupling simulation module is used to divide multiple physical fields into fast-changing and slow-changing subdomains based on a time-scale threshold. A pre-trained physical information neural network is used to fit the coupling region between the fast-changing and slow-changing subdomains to obtain a multiphysics coupling model. The governing equations of the multiphysics coupling model are then solved asynchronously to obtain a full-field state dataset.

[0102] The hidden state assimilation module is used to assimilate the hidden state variables of the instrument under test based on the multi-physics coupling model and the full-field state dataset, and obtain the optimal estimate of the hidden state variables.

[0103] The dynamic compensation module is used to decompose the optimal estimate of the hidden state variable into fast and slow components. Based on the fast and slow components, drift compensation is performed on the raw output signal acquired in real time by the instrument under test to obtain the compensated output signal.

[0104] The boundary condition correction module compares the compensated output signal with the reference measurement value representing the real physical quantity to obtain the comparison deviation. Using this deviation as information, the boundary condition parameters of the multiphysics coupling model are inverted using an ensemble Kalman filter algorithm to obtain the optimal estimate of the boundary condition parameters.

[0105] The evaluation module is used to perform online closed-loop correction of the multiphysics coupling model based on the optimal estimation of boundary condition parameters, obtain an updated full-field state dataset, and determine the dynamic response characteristics of the instruments based on the updated full-field state dataset.

[0106] The solution provided by this system is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the simulation test system for instruments under multi-physics coupling provided below can be found in the limitations of the simulation test method for instruments under multi-physics coupling described above, and will not be repeated here.

[0107] In one exemplary embodiment, such as Figure 5 As shown, a simulation testing system 500 for instruments under multi-physics coupling is provided to implement the methods in the above embodiments. The system includes: The multiphysics coupling simulation module 501 is used to acquire the multiphysics field of the instrument. Based on the time scale threshold, the multiphysics field is divided into a fast-changing field subdomain and a slow-changing field subdomain. By setting up a physical information neural network in the coupling region of the fast-changing field subdomain and the slow-changing field subdomain, a multiphysics coupling model is constructed. The control equations of the multiphysics coupling model are solved asynchronously and coupled to generate a full-field state dataset. The hidden state assimilation module 502 is used to assimilate the unmeasurable hidden state variables inside the instrument based on the multi-physics coupling model and the full-field state dataset, and obtain the optimal estimate of the hidden state variables by using the extended Kalman filter algorithm based on reduced order. The dynamic compensation module 503 is used to decompose the optimal estimate into fast and slow components to obtain fast and slow components; determine the fast component drift and slow component drift according to the fast and slow components respectively; superimpose the fast and slow component drifts to obtain the real-time compensation values ​​for zero drift and sensitivity drift; and use the real-time compensation values ​​to compensate the original output signal collected by the instrument to generate a compensated output signal. The boundary condition correction and evaluation module 504 is used to compare the compensated output signal with the reference measurement value representing the real physical quantity to obtain the comparison deviation; using the comparison deviation as information, the boundary condition parameters of the multiphysics coupling model are inverted through the ensemble Kalman filter algorithm to obtain the optimal estimate of the boundary condition parameters; based on the optimal estimate, the multiphysics coupling model is corrected online in closed loop to obtain the updated full-field state dataset; and based on the updated full-field state dataset, the dynamic response characteristic index of the instrument is determined.

[0108] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0109] The embodiments described above are merely examples of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A simulation testing method for instruments under multi-physics coupling, characterized in that, include: To acquire the various physical fields that interact and influence each other in the working environment of the instrument under test; Based on a time scale threshold, multiple physical fields are divided into fast-changing field subdomains and slow-changing field subdomains. A pre-trained physical information neural network is used to fit the coupling region between the fast-changing field subdomains and the slow-changing field subdomains to obtain a multi-physics coupling model. The control equations of the multi-physics coupling model are solved asynchronously to obtain the full-field state dataset. Based on the multiphysics coupling model and the full-field state dataset, the hidden state variables of the instrument under test are assimilated by the reduced-order extended Kalman filter algorithm to obtain the optimal estimate of the hidden state variables. The optimal estimate of the hidden state variable is decomposed into fast and slow components. Drift compensation is performed on the raw output signal acquired in real time by the instrument under test based on the fast and slow components to obtain the compensated output signal. The compensation output signal is compared with the reference measurement value of the reference output that characterizes the real physical quantity to obtain the comparison deviation; Using the comparison deviation as information, the boundary condition parameters of the multiphysics coupling model are inverted through the ensemble Kalman filter algorithm to obtain the optimal estimate of the boundary condition parameters; The optimal estimation of boundary condition parameters is used to perform online closed-loop correction of the multiphysics coupling model, resulting in an updated full-field state dataset. Based on the updated full-field state dataset, the dynamic response characteristics of the instruments are determined.

2. The simulation testing method for instruments under multi-physics coupling as described in claim 1, characterized in that, Training the physical information neural network specifically includes: The system acquires various historical physical fields that interact and influence each other in the working environment of the instrument, and divides these historical physical fields into historical fast-changing field subdomains and historical slow-changing field subdomains based on time scale thresholds. The historical rapidly changing field subdomain is solved recursively time-by-time using the explicit time integration method to record the historical rapidly changing field state variables at each time point, thus obtaining the historical rapidly changing field time series. The historical slowly changing field subdomain is solved using the implicit time integration method to obtain the historical slowly changing field state variables at each time point. A physical information neural network is used to fit the fast-changing field time series in the coupling region. During the fitting process, the network loss function is constructed based on the following formula: ； )； ； in, The output of the physical information neural network, These are the training parameters for the physical information neural network. For time, The number of solution steps for a rapidly varying field subdomain within a single slowly varying field subdomain. This represents the current solution time for the slowly varying field subdomain. To determine the solution time step for the rapidly varying field subdomain, For fast-changing field time series, To balance the coefficients of the data fitting term and the physical constraint term, For the physical governing equations satisfied by the rapidly varying field subdomain, Indicated by The physical information of the network parameters is mapped to the neural network. For fast-changing field diffusion terms, The diffusion coefficient of the medium, For gradient operators, These are terms representing the rapidly changing field source term, external excitation, and multiphysics coupling interaction term. The physical information neural network is trained with the goal of minimizing the network loss function, resulting in the trained physical information neural network.

3. The simulation testing method for instruments under multi-physics coupling as described in claim 1, characterized in that, The process involves fitting the coupling region between the rapidly varying and slowly varying field subdomains using a pre-trained physical information neural network to obtain a multi-physics coupling model. The governing equations of this model are then solved asynchronously to obtain the full-field state dataset. The fast-changing field subdomain is solved recursively time-by-time using the explicit time integration method to record the fast-changing field state variables at each time step, thus obtaining the fast-changing field time series. The slow-changing field subdomain is solved using the implicit time integration method to obtain the slow-changing field state variables at each time step. The precise value of the slowly varying field subdomain at each time step is determined by a pre-trained physical information neural network, and the precise value is passed as an equivalent load to the control equation of the slowly varying field subdomain to obtain the updated slowly varying field state quantity. At each time step, the state variables of the slowly varying field are decomposed into eigenorthogonal components to extract time coefficients and interpolate them to obtain a continuously varying function. Based on the continuously varying function, the boundary conditions of the slowly varying field corresponding to each time step of the rapidly varying field subdomain are determined and passed to the control equations of the rapidly varying field subdomain to obtain the updated rapidly varying field time series. The updated fast-changing field time series and the updated slow-changing field state variables at all times are integrated to obtain the full-field state dataset; The governing equations for the slowly varying field subdomains include second-order partial differential equations with constant coefficients that describe the variation of physical quantities in the slowly varying field subdomains. The governing equations for the rapidly varying field subdomains include first-order partial differential equations with constant coefficients that describe the variation of physical quantities in the rapidly varying field subdomains.

4. The simulation testing method for instruments under multi-physics coupling as described in claim 3, characterized in that, Based on the following formula, the rapidly changing field subdomain is recursively solved time-by-time using the explicit time integration method to record the rapidly changing field state variables at each time step, thus obtaining the rapidly changing field time series: ； ； Based on the following formula, the state variables of the slowly varying field at each time step are obtained by solving the slowly varying field subdomain using the implicit time integration method: ； ； Among them, in the two formulas The first The physical field state quantities of the rapidly varying field subdomain and the slowly varying field subdomain, in the two formulas The first The physical field state quantities of the rapidly changing field subdomain and the slowly changing field subdomain. To determine the solution time step for the rapidly varying field subdomain, For the first The time value at any given moment For the first Evolution operator of physical field state quantities at time step Let be the solution time step for the slowly varying field subdomain. For the first The time value at any given moment For the first Evolution operator of physical field state quantities at time step For the physical field diffusion term The diffusion coefficient of the medium, For gradient operators, This is the physical field source term.

5. The simulation testing method for instruments under multi-physics coupling as described in claim 1, characterized in that, The method, based on a multiphysics coupling model and a full-field state dataset, uses a reduced-order extended Kalman filter algorithm to assimilate the latent state variables of the instrument under test, obtaining the optimal estimates of the latent state variables. Specifically, this includes: Based on the control equations of the multiphysics coupling model, state transition equations are constructed to describe the recursive evolution of hidden state variables over time, and observation equations are constructed to describe the mapping relationship between hidden state variables and the raw output signals acquired in real time by instruments. The state vectors in the full-field state dataset are subjected to intrinsic orthogonal decomposition, and the basis matrix composed of orthogonal eigenvectors whose cumulative contribution rate satisfies a preset threshold during the intrinsic orthogonal decomposition is used as the reduced-order basis. Based on the following equation, the state transition equation and observation equation are projected onto a low-dimensional linear space composed of column vectors of the reduced-order basis to obtain the reduced-order state transition equation and reduced-order observation equation: ； in, This is the reduced-order state vector at the current moment. for The reduced-order state vector at time t, This is a reduced-order basis obtained from eigenorthogonal decomposition. Here is the state transition matrix. For the observation vector, For the observation matrix, For process noise, To observe noise; The projection of the state vector onto the reduced basis is used as the initial modal coefficient vector. Based on the reduced state transition equation, the reduced observation equation, and the original output signal collected in real time by the instrument, the posterior estimate of the modal coefficient vector at the current time is obtained by Kalman filtering recursively. By using a reduced basis, the posterior estimate of the modal coefficient vector at the current moment is mapped to the original high-dimensional space of the full-field state dataset, thus obtaining the optimal estimate of the latent state variable at the current moment.

6. The simulation testing method for instruments under multi-physics coupling as described in claim 5, characterized in that, The reduced-order extended Kalman filter algorithm is expressed by the following formula: ； in, for The prior estimate of the state at time step [time]. Here is the state transition matrix. It is the transpose of the state transition matrix. for The posterior estimate of the state at time step [time]. To control the input matrix, for Constantly control the amount of input. for Prior estimation of the covariance matrix of the state at each time step. for The covariance matrix of the posterior estimate of the state at time step is... The process noise covariance matrix is... for Moment-time Kalman gain, For the observation matrix, Let be the transpose of the observation matrix. To observe the noise covariance matrix, for k The posterior estimate of the state at time step [time]. for Real-time observation data collected by instruments and meters. It is an identity matrix.

7. The simulation testing method for instruments under multi-physics coupling as described in claim 1, characterized in that, The optimal estimate of the hidden state variable is decomposed into fast and slow components to obtain fast and slow components. Drift compensation is performed on the raw output signal acquired in real time by the instrument under test based on the fast-changing and slow-changing components to obtain the compensated output signal, which specifically includes: Based on the optimal estimate of the hidden state variables, the fast-changing component of the optimal estimate is extracted by a high-pass filter, and the slow-changing component of the optimal estimate is extracted by a low-pass filter. Based on the material thermophysical properties of the sensitive element of the instrument under test, a mapping relationship between the slowly varying component and the slowly varying drift is constructed to obtain the slowly varying mapping relationship; based on the dynamic response characteristics of the sensitive element of the instrument under test, a mapping relationship between the rapidly varying component and the rapidly varying drift is constructed to obtain the rapidly varying mapping relationship. Based on the fast-changing mapping relationship, the fast-changing component drift value at the current moment is obtained by mapping the fast-changing components; based on the slow-changing mapping relationship, the slow-changing component drift value at the current moment is obtained by mapping the slow-changing components. The fast-changing component drift value and the slow-changing component drift value are superimposed to obtain the total zero-point drift and the total sensitivity drift. The zero-point reference value and sensitivity reference value of the instrument under test are obtained under standard conditions. The total zero-point drift and the zero-point reference value are added together to obtain the zero-point compensation value. The total sensitivity drift and the sensitivity reference value are added together to obtain the sensitivity compensation value. Based on the zero-point compensation value and the sensitivity compensation value, the original output signal of the instrument under test is compensated in real time to obtain the compensated output signal.

8. The simulation testing method for instruments under multi-physics coupling as described in claim 1, characterized in that, The process of using the comparison deviation as information and inverting the boundary condition parameters of the multiphysics coupling model using an ensemble Kalman filter algorithm to obtain the optimal estimate of the boundary condition parameters specifically includes: The current compensated output signal is compared with the original output signal to determine the comparison deviation between the compensated output signal and the original output signal, and the comparison deviation is used as the information. Vectorize the boundary condition parameters of the multiphysics coupling model to obtain the initial set; Based on the new information, each member in the initial set is updated using the ensemble Kalman filter algorithm to obtain the updated boundary condition parameter set, and the average value of the updated boundary condition parameter set is used as the optimal estimate of the boundary condition parameters at the current time.

9. The simulation testing method for instruments under multi-physics coupling as described in claim 1, characterized in that, The optimal estimation based on boundary condition parameters is used to perform online closed-loop correction of the multiphysics coupling model, resulting in an updated full-field state dataset. Based on this updated full-field state dataset, the dynamic response characteristics of the instruments are determined, specifically including: The optimal estimate based on the boundary condition parameters is used as the new boundary condition input to the multiphysics coupling model to resolve the control equations of the multiphysics coupling model and obtain an updated full-field state dataset. Based on the updated full-field state dataset, the optimal estimate and the compensation output signal are updated until the deviation between the compensation output signal and the original output signal meets the preset threshold, and the updated compensation output signal is obtained. The dynamic response characteristics of the instrument under test are determined based on the updated compensated output signal.

10. A simulation testing system for instruments under multi-physics coupling, characterized in that, include: The data acquisition module is used to acquire various physical fields that interact and influence each other in the working environment of the instrument under test. The multiphysics coupling simulation module is used to divide multiple physical fields into fast-changing field subdomains and slow-changing field subdomains based on time scale thresholds. The coupling region between the fast-changing field subdomains and the slow-changing field subdomains is fitted by a pre-trained physical information neural network to obtain a multiphysics coupling model. The control equations of the multiphysics coupling model are solved asynchronously to obtain the full-field state dataset. The hidden state assimilation module is used to assimilate the hidden state variables of the instrument under test based on the multi-physics coupling model and the full-field state dataset, and obtain the optimal estimate of the hidden state variables. The dynamic compensation module is used to decompose the optimal estimate of the hidden state variable into fast and slow components. Drift compensation is performed on the raw output signal acquired in real time by the instrument under test based on the fast and slow components to obtain the compensated output signal. The boundary condition correction module is used to compare the compensated output signal with the reference measurement value of the reference output that characterizes the real physical quantity to obtain the comparison deviation; Using the comparison deviation as information, the boundary condition parameters of the multiphysics coupling model are inverted through the ensemble Kalman filter algorithm to obtain the optimal estimate of the boundary condition parameters; The evaluation module is used to perform online closed-loop correction of the multiphysics coupling model based on the optimal estimation of boundary condition parameters, obtain an updated full-field state dataset, and determine the dynamic response characteristics of the instruments based on the updated full-field state dataset.

Images