Calculation and measurement fused high-fidelity digital twinning dynamic monitoring method

Abstract

The invention discloses a calculation and measurement fused high-fidelity digital twinning dynamic monitoring method for a cantilever beam structure. Firstly, a real-time updated database is constructed based on a physical entity system; secondly, constructing a three-dimensional digital model by using SolidWorks, and performing multi-physics field simulation analysis in combination with Ansys to obtain high-precision mechanism data of a structural key area; thirdly, performing reduced-order calculation on a simulation result by using MWorks, training and optimizing a reduced-order model through a feedforward neural network, and reducing an error between simulation data and reduced-order data; fusing the one-dimensional system data order reduction model and the three-dimensional field data order reduction model based on a Modelica language; and finally, integrating a mechanism-data hybrid model, constructing a dynamic response characteristic and spatial distribution characteristic integrated high-fidelity digital twinborn body, comparing twinborn body calculation data with sensor actual measurement data in real time, and accurately representing geometric morphology, mechanical properties and dynamic behaviors of a physical system. The method is suitable for high-precision monitoring, real-time state evaluation and performance optimization of the cantilever beam structure.

Description

Technical Field

[0001] The present invention relates to the technical field of digital twin modeling and structural monitoring, and particularly to a high-fidelity digital twin dynamic monitoring method for integrated calculation and measurement of a cantilever beam structure. Background Art

[0002] With the development of digital intelligent manufacturing software platforms such as Twin Builder, ThingWorx, and MWorks, the digital twin technology integrating multiple disciplines and multiple systems has gradually become the core technical means in the engineering field. This technology realizes the dynamic monitoring and simulation prediction of the entire life cycle of the physical system by mirroring the physical entity into the virtual space, acquiring and analyzing the operation data in real time, and provides important support for the optimization and management of complex systems.

[0003] However, there are still many challenges in the current digital twin monitoring methods in terms of accuracy, real-time performance, and multi-parameter integrated monitoring capabilities. Traditional engineering structure monitoring deploys limited sensors on physical objects to collect local data, which is difficult to comprehensively and accurately reflect the overall dynamic behavior of the system; while multi-physical field simulation is difficult to achieve real-time response due to the complexity of high-dimensional models and long calculation time. In addition, under complex working conditions, the completeness of data collection is insufficient, the dynamic update ability of the model is limited, and the simulation and reduced-order calculation efficiency is low, resulting in the existing digital twins being difficult to meet the requirements of high precision and high efficiency at the same time.

[0004] Especially in linear engineering structures such as cantilever beams, the monitoring and simulation accuracy of the structural dynamic behavior is crucial for system performance and safety. The existing technologies still face technical bottlenecks in the reduced-order calculation of high-dimensional simulation data, real-time dynamic response and visualization synchronization, and multi-parameter fusion modeling. The fusion of sensor data and virtual model calculation results is not efficient enough, the simulation and reduced-order calculation are time-consuming, and the dynamic visualization response of the model is delayed, which limits the application effect of digital twin technology in the monitoring and optimization of complex engineering structures.

[0005] The key to digital twin dynamic monitoring lies in realizing the accurate mapping between the virtual and physical twins, that is, through the deep fusion of data collected by multiple sensors and simulation calculation data, quickly and dynamically updating the virtual model to characterize the actual state of the physical system. This requires real-time data processing capabilities, efficient simulation and reduced-order calculation methods, and dynamic visualization capabilities to support the monitoring and optimization requirements under complex working conditions.

[0006] In summary, the existing technologies have limitations in aspects such as high-fidelity modeling of cantilever beams, real-time dynamic response, and multi-parameter integrated monitoring. There is an urgent need for a method that can simultaneously meet the requirements of high precision, real-time performance, and wide applicability. To this end, the present invention proposes a high-fidelity digital twin dynamic monitoring method integrating calculation and measurement, which combines technologies such as multi-sensor data acquisition, real-time data transmission and processing, multi-physical field modeling and simulation optimization, model reduction calculation and feedforward neural network training, and three-dimensional dynamic visualization to achieve high-precision dynamic characterization of the geometric shape and mechanical properties of the cantilever beam structure. Based on the technology of integrating calculation data from simulation and measured data from sensors, the present invention can accurately depict the dynamic behavior of the physical system, providing reliable technical support for the real-time monitoring, condition assessment, and performance optimization of the cantilever beam structure. Summary of the Invention

[0007] The purpose of the present invention is to overcome the deficiencies of the existing technologies and propose a high-fidelity digital twin dynamic monitoring method integrating calculation and measurement applicable to linear engineering structures such as cantilever beams. This method constructs a dynamic interaction mapping parallel system between the physical system and the digital twin to real-time characterize the dynamic behavior (such as displacement, strain) of the physical entity under different working conditions (such as load, boundary conditions). As Figure 1 shown, the main contents of the present invention include: physical system data acquisition and transmission, three-dimensional digital model construction and simulation, reduction calculation and neural network model training, three-dimensional data fusion modeling, and mechanism-data hybrid model integration and digital twin dynamic visualization display. The specific steps are as follows:

[0008] Step1: Based on measurement components such as acceleration sensors, strain gauges, displacement sensors, and force sensors pasted or fixed on the physical structure, as well as a transmitter for digital-to-analog signal conversion, a touch screen for displaying measurement data, a display screen for presenting the digital twin simulation interface, and a power supply module for power supply, jointly build a physical entity test system to provide hardware support for data acquisition and dynamic monitoring.

[0009] Step2: For the physical entity test system constructed in Step1, collect load force data, displacement data, and strain data through force sensors, displacement sensors, and resistance strain gauges respectively. Use a digital transmitter to convert the analog signal into a digital signal, then use the RS485 communication standard for data transmission, and upload it to the display screen based on hardware communication protocols such as Modbus. Use Ethernet and industrial wireless communication technologies to connect data acquisition devices such as sensors in the physical system and data processing tools such as simulation software to achieve real-time transmission and storage of multi-source data, providing a reliable data basis for subsequent modeling and analysis.

[0010] Step 3: For the physical entity system described in Step 2, based on physical parameters such as geometric structure dimensions, material properties, and boundary conditions, use SolidWorks to establish a three-dimensional digital model. You can also choose modeling software such as CAD, 3D Max, Creo, etc., to ensure that the model accurately represents the geometric characteristics and structural details of the physical system, providing a reliable basis for subsequent simulation analysis and digital twin modeling.

[0011] Step 4: For the three-dimensional model constructed in Step 3, use Ansys finite element analysis software to perform static and multi-physics field simulation analysis to obtain system data and field data respectively. You can also choose Abaqus, Comsol, Nastran, or other simulation software. To balance computational accuracy and efficiency, perform mesh encryption and refinement on the key areas of the structure to capture local gradient changes in physical parameters and achieve high-precision simulation and dynamic response characteristic analysis of the cantilever beam structure.

[0012] Step 5: For the high-dimensional simulation data generated in Step 4, adopt a reduced-order calculation method to reduce the model complexity and improve the simulation speed and computational efficiency. Preferably, use the MWorks physical modeling and simulation environment for reduced-order processing. This platform has high efficiency and accuracy in multi-physics field modeling and rapid reduced-order of complex systems, and at the same time supports seamless integration with other simulation software. You can also choose tools such as Ansys, Comsol, FreeFEM, etc. according to requirements, but there are differences in reduced-order speed and computational resource consumption.

[0013] Combining the static behavior and dynamic characteristics of the cantilever beam, the calculation formula for vector displacement is:

[0014]

[0015] where V is the vector displacement with direction and magnitude at a certain point on the cantilever beam, x is the distance from the fixed end of the beam, t is the time, F is the external load acting on the free end of the beam (such as the weight of the weight), E is the elastic modulus of the beam material, I is the cross-sectional moment of inertia of the beam, l is the length of the beam, p i (t) is the modal coefficient representing the change of the m modal functions representing displacement with time in the reduced-order method (such as singular value decomposition), and α i (x) is the i-th modal function. Among them, the modal function retains the stiffness matrix information. To balance the calculation speed and retain effective information, generally take the first 3-5 order modal functions.

[0016] By capturing complex transient behavior and response, the calculation formula for scalar strain is:

[0017]

[0018] Wherein, S is the strain at a certain point of the cantilever beam, y is the vertical distance from the neutral axis on the beam cross-section, q k (t) is the modal coefficient of the n modal functions representing the strain varying with time in the reduced-order method, μ k (x) is the k-th modal function. Among them, the modal function retains the information of the strain-displacement relationship matrix.

[0019] Step6: For the reduced-order calculation results in Step 5, in order to minimize the error between the original simulation data and the reduced-order fitting data, construct a multi-layer feedforward neural network f N (such as convolutional neural network CNN, recurrent neural network RNN, long short-term memory network LSTM, etc.) for training to improve the response prediction ability of the model for complex systems under different conditions. The loss function calculation formula of f N is:

[0020]

[0021] Wherein, L is the loss function, f N is the mapping function of the neural network from input to output, N is the set of network model parameters (including the number of hidden layers C, neuron size Y, batch size Batch size, number of iterations Epoch, learning rate Rate, weight parameter w, bias parameter b, convergence threshold θ, etc.), T is the total number of reduced-order data training samples, and S sim are the original vector and scalar simulation data at position x j and time t j respectively, and S mor are the reduced-order vector and scalar data at the same position and time respectively, and f N (S mor ) are the predicted values of the neural network model for the reduced-order vector and scalar data respectively.

[0022] When L(N) reaches the convergence condition, that is, satisfies the threshold θ, the final system reduced-order data output by the model is:

[0023]

[0024] Wherein, and S * (x, t) are the accurately fitted displacement and strain reduced-order data respectively, argmin N is the optimized set of neural network parameters.

[0025] Step 7: For the system order reduction data output in Step 6, use Modelica language to establish a one-dimensional model and achieve integrated modeling with the three-dimensional field data model. Combining the spatial distribution characteristics of the three-dimensional coordinate system and the order reduction data, adopt 3-5 order modal functions to capture the dynamic characteristics in different direction dimensions, and obtain field order reduction data that conforms to the three-dimensional structure response and spatio-temporal evolution law of the physical system:

[0026]

[0027] Where x, y, and z are the distribution of the physical field in the three spatial directions over time t, and are the dynamic distribution representations of the order reduction data of vectors and scalars in the three-dimensional visualization model respectively.

[0028] Step 8: For the three-dimensional structure simulation model in Step 4, it provides high-precision mechanism data; the simulation data order reduction model in Step 5 retains key modes through dimensionality reduction, reducing the computational complexity; the neural network training model in Step 6 further fits the order reduction data to improve the response prediction ability under different working conditions and external conditions; the three-dimensional visualization model in Step 7 realizes the spatial reconstruction of vector and scalar physical quantities and is used for real-time dynamic display in combination with modal functions; based on interface standards and modeling specifications such as FMI and OPC UA, encapsulate and integrate the above multi-source heterogeneous models to construct a high-fidelity digital twin of the physical system.

[0029] Step 9: For the digital twin model constructed in Step 8, run it under different loads, boundary conditions or working conditions, and compare and analyze by collecting the sensor measurement data of the cantilever beam structure in real time with the calculation output of the twin. Calculate the relative error between the measurement data and the calculation data. Based on the difference analysis results, dynamically adjust the modal functions or neural network parameters in the twin model to continuously optimize the prediction accuracy and dynamic response ability of the model, and achieve high-precision dynamic characterization of the geometric shape and mechanical properties of the cantilever beam structure. Finally, the error between the twin data and the real data for different working condition models does not exceed 3%, and the response time is not greater than 0.3 s, forming a stable and highly confident calculation-measurement fusion high-fidelity digital twin model, providing reliable technical support for the real-time monitoring and condition assessment of the cantilever beam structure.

[0030] The present invention has the following beneficial effects:

[0031] The present invention proposes a high-fidelity digital twin dynamic monitoring method for measurement and calculation integration applicable to linear engineering structures such as cantilever beams, which solves the deficiencies in real-time performance and dynamic monitoring capabilities of digital twin modeling for geometric shapes and structural mechanical properties in the prior art, and breaks through the bottleneck of low prediction accuracy caused by limited sensor measured data and low-fidelity simulation data in traditional methods. In contrast, the advantages of the present invention are: simplicity (reducing the modeling complexity through measurement and calculation integration and reducing the dependence on large-scale sensor arrangements), speed (optimizing the model reduction calculation technology, significantly reducing the simulation calculation time, and improving the real-time response ability), accuracy (improving the accuracy of dynamic behavior characterization under different working conditions), and good generalization (applicable to digital twin modeling and monitoring of typical linear engineering structures such as cantilever beams). Therefore, the present invention provides an efficient solution for the real-time monitoring and condition assessment of cantilever beam structures, and has broad application prospects and promotion value. Description of the Drawings

[0032] Figure 1 is the basic process of the present invention

[0033] Figure 2 is the schematic diagram of the physical system

[0034] Figure 3 is the three-dimensional model of the cantilever beam structure

[0035] Figure 4 is the one-dimensional model of the Modelica text

[0036] Figure 5 is the comparison result of the measured data and the simulation data

[0037] Figure 6 is the simulation result of the displacement and strain combined field data Detailed Embodiment

[0038] In order to more clearly describe the technical features of the present invention, the implementation process of the present invention will be described in detail below with reference to the accompanying drawings. It should be emphasized that the cantilever beam example is only one implementation of the present invention, aiming to show the application method of the present invention and should not be regarded as a limitation of the present invention. Other implementation manners that conform to the technical idea and innovation points of the present invention are equally applicable. Therefore, the present invention has broad applicability and flexibility and can be appropriately adjusted and optimized according to specific application requirements.

[0039] (1) Considering the mechanical characteristics of the cantilever beam structure and the influence of additional materials on its deformation and movement, select appropriate force sensors, displacement sensors, and resistance strain gauges, and make a reasonable layout according to the experimental requirements. The specific installation method of the physical system is as Figure 2As shown in the figure, the entire test bench consists of the experimental system connection control box on the left and the cantilever beam test bench on the right. The left side of the 304 stainless steel cantilever beam with a size of 340mm×15mm×5mm is fixed to the bench through a clamping device. The first resistance strain gauge is pasted on the upper surface of the beam, close to the right side of the clamping device (40mm from the root of the beam), and another one is pasted every 80mm. The positions of the four strain gauges are recorded as x 1 、x 2 、x 3 and x 4 The S-type tension and pressure sensor is connected to the free end of the right side of the cantilever beam through a screw rod, and a weight plate is suspended below the sensor. Preferably, the displacement sensor is a point laser displacement sensor with a model of SGI150, a range of 150mm, and an accuracy of 0.001mm, which can measure the deformation displacement of the free end of the cantilever beam with high precision.

[0040] (2) The pressure sensor signal and the resistance strain gauge signal are converted into digital signals through digital transmitters, and then the RS485 communication standard is used for data transmission, and the data is uploaded to the Weiluntong display screen based on the Modbus communication protocol. The LRS-150W single-group output power supply is used to convert 220V AC power into 24V DC power to power the display screen, sensors and other equipment. For easy operation, the above equipment is integrated into the experimental system connection control box on the left and controlled by the buttons on the shell.

[0041] (3) Preferably, a three-dimensional digital model is established using SolidWorks for the cantilever beam and its connected related structures, such as Figure 3 In this example, the selected BMB350 strain gauge is small in size and light in weight, and its effect on the original structure of the cantilever beam is negligible, so it is not included in the construction of the 3D model.

[0042] (4) Static simulation of the three-dimensional model was performed using Ansys Workbench, with the load starting from 0 and increasing gradually from 1 N to 45 N. The simulation results under 45 load conditions were exported as .rst format files and saved as system data in Excel format using the Export function. In addition, EnSight software was used to convert the result files into field data in .case format for post-processing.

[0043] (5) Import the system data into the model reduction toolbox of MWORKS-Sysplorer, set the load force as the input variable, and the displacement and strain as the output variables. Combined with the vibration equation and boundary conditions of the cantilever beam, the calculation formula of the displacement can be converted to:

[0044]

[0045] Wherein, V represents the displacement at position x on the cantilever beam at time t, x is the distance from the fixed end of the beam (0 - 340 mm), F(t) is the time-varying external load acting on the free end of the beam (0 - 45 N), E is the elastic modulus of the beam material (2×10 11 N / m 2 ), I is the moment of inertia of the beam cross-section (1.5625×10 -10 m 4 ), and l is the length of the beam (340 mm).

[0046] Select the first 3 modal components, then m = 3 in the above formula,

[0047] where ρ is the material density (8000 kg / m 3 ), and A is the cross-sectional area (75 mm 2 ).

[0048] By capturing the dynamic condition response, the calculation formula of strain can be transformed into:

[0049]

[0050] Wherein, S represents the strain at position x on the cantilever beam at time t, y is the vertical distance from the neutral axis on the beam cross-section (0.0025 m), and the meanings and values of the parameters E, I, and F(t) are the same as those in formula (1).

[0051] Select the first 3 modal components, then n = 3 in the above formula,

[0052] where the values of ρ and A are the same as those in formula (1).

[0053] Meanwhile, read the displacement vector data (including the x, y, and z directions) of each node and the strain and stress scalar data of each element in the.case file. Select the first 3 modal functions to perform dimensionality reduction processing on the field data, and randomly divide it into a training set and a validation set according to a ratio of 7:3.

[0054] (6) Preferably, a Convolutional Neural Network (CNN) model is constructed to train the system data and field data respectively. Initially, the number of hidden layers C is set to 3, the number of neurons in the hidden layers Y are 16, 32, and 64 respectively, the activation function is selected as Sigmoid, the loss function is Mean Squared Error (MSE), the optimizer uses AdamW with an adaptive learning rate, the batch size is 32, the number of epochs is 500, the learning rate is 0.01, the initial values of the weight parameter w and the bias parameter b are 0.5 and 0.05 respectively, and the convergence threshold θ is 0.005. The sum of the MSEs based on displacement and strain, L(N), is used as the loss function for neural network training:

[0055]

[0056] where the mapping function The value of the displacement simulation data is 0.419F (mm), and the value of the strain simulation data S sim is 24F, with the unit of microstrain με, and F = the weight of the weight at the free end of the beam (kg) × 9.8 m / s 2 ; N represents the set of network model parameters {C, Y, Batch size, Epoch, Rate, w, b, θ}; T is the system reduced-order training data sample of displacement and strain, a total of 225 × 0.7 = 158 groups; the predicted values of displacement and strain by the CNN model and f N (S mor ), compared with the original simulation data, the error ranges are ±0.5% and ±0.9% respectively.

[0057] When L(N) reaches the convergence condition, that is, the error is minimized, it is usually considered that the training is completed when the real-time loss value (Loss) is less than the threshold of 0.005. The final reduced-order data output by the model is:

[0058]

[0059] where the reduced-order displacement data after precise fitting has a value range of [0, 18.94 mm], and the reduced-order strain data S * (x, t) has a value range of [0, 1089.72 με], and the optimized set of neural network parameters argmin N = {C = 3, Y = [4, 8, 16], Batch size = 16, Epoch = 1000, Rate = 0.001, w = 0.55, b = 0.01, θ = 0.005}.

[0060] The training effect of the model is further quantified by calculating the relative error between the original data and the reduced-order data under each load condition. If the relative error is less than 5%, it indicates that the model training is relatively accurate and the parameter settings are reasonable. After the system data and field data are trained by the optimized neural network model, if the above convergence conditions and error thresholds are met, the reduced-order data containing the physical field distribution under the load will be exported respectively for subsequent co-simulation use.

[0061] (7) Establish a comparison model from the physical measurement data to the simulation reduced-order data based on the Modelica language, as Figure 4 shown. Access the MySQL database through Python code, convert the mass measured by the force sensor into the load force, which is used as the input of the reduced-order model. Use C language code to call the DLL function to access the database, and the system model can output the simulation displacement value at the free end of the cantilever beam and the strain data at the positions of the 4 strain gauges [x 1 , x 2 , x 3 and x 4 in step (1). The three-dimensional field model can generate visual field data:

[0062]

[0063] where x, y, and z are the distribution fields of the reduced-order data in the directions of length [0, 300 mm], width [0, 15 mm], and height [0, 5 mm] respectively, and the displacement takes values in the three directions based on the range of the external load changing with time as {x ∈ [0, 18.85 mm], y ∈ [0, 2.10 mm], z ∈ [0, 18.85 mm]}, and the strain takes values in the three directions changing with the external load as {x ∈ [0, 1080 με], y ∈ [0, 324 με], z ∈ [0, 2.77 με]}.

[0064] (8) Integrate the above models to construct a digital twin of the cantilever beam. During the process of gradually increasing the weight of the weights (0 - 4600 g) on the weight tray, the co-simulation of the twin model is carried out synchronously. The measured displacement and microstrain (με) are compared with the reduced-order simulation values of the system model at an update interval of 0.2 s, as Figure 5 shown. The results show that the simulation data of the twin model has high consistency with the measured data, the average error is less than 2%, and the maximum error is less than 3%, meeting the requirements of high-precision engineering monitoring, and verifying the accuracy and reliability of the digital twin.

[0065] (9) In the Figure 2 display screen, the distribution field data of displacement, strain, and stress are dynamically displayed, and at the same time, the combined fields of displacement + strain and displacement + stress are supported for real-time visualization, as Figure 6As shown. The results indicate that the constructed digital twin system can achieve the follow-up monitoring of the cantilever beam structure with high fidelity, and accurately simulate and real-time characterize its geometric morphology and structural mechanical properties, providing effective technical support for state assessment and performance optimization under different working conditions.

Claims

1. A high-fidelity digital twin dynamic monitoring method integrating calculation and measurement, characterized in that: Step 1: Build a physical entity test system based on accelerometers, strain gauges, displacement sensors and force sensors attached or fixed to the physical structure, as well as transmitters for digital-to-analog signal conversion, touch screens for displaying measurement data, displays for displaying the digital twin simulation interface and power modules for power supply; Step 2: For the physical entity test system constructed in step 1, load force data, displacement data and strain data are collected through force sensors, displacement sensors and resistance strain gauges respectively; analog signals are converted into digital signals using digital transmitters, and then data is transmitted using the RS485 communication standard, and then uploaded to the display screen based on the hardware communication protocol; Ethernet and industrial wireless communication technology are used to interconnect the data acquisition equipment and data processing tools in the physical system to achieve real-time transmission and storage of multi-source data; Step 3: Model the physical entity system described in step 2 using modeling software based on geometric structure dimensions, material properties and boundary conditions; Step 4: For the three-dimensional model constructed in step 3, use Ansys finite element analysis software to perform statics and multi-physics field simulation analysis to obtain system data and field data respectively; Step 5: For the simulation data generated in step 4, the order reduction calculation method is used to reduce the model complexity and improve the simulation speed and calculation efficiency; Combining the static behavior and dynamic characteristics of the cantilever beam, the calculation formula for the vector displacement is: Where V is the vector displacement with direction and magnitude at a point on the cantilever beam, x is the distance from the fixed end of the beam, t is the time, F is the external load acting on the free end of the beam, E is the elastic modulus of the beam material, I is the section moment of inertia of the beam, l is the length of the beam, and p is i (t) is the time-varying modal coefficient of the m modal functions representing the displacement in the reduced-order method, α i (x) is the i-th mode function; the mode function retains the stiffness matrix information; in order to balance the calculation speed and retain effective information, the first 3-5 order mode functions are taken; By capturing the complex transient behavior and response, the scalar strain is calculated as: Where S is the strain at a certain point on the cantilever beam, y is the vertical distance from the neutral axis on the beam section, and q k (t) is the modal coefficient of the n modal functions representing the strain in the reduced-order method, μ k (x) is the kth mode function; where the mode function retains the strain-displacement relationship matrix information; Step 6: Based on the reduced-order calculation results in step 5, in order to minimize the error between the original simulation data and the reduced-order fitting data, a multi-layer feedforward neural network f is constructed. N Conduct training; N The loss function calculation formula is: Where L is the loss function, f N is the mapping function of the neural network from input to output, N is the set of network model parameters, including the number of hidden layers C, neuron size Y, batch size Batch size, number of iterations Epoch, learning rate Rate, weight parameter w, bias parameter b, convergence threshold θ; T is the total number of reduced-order data training samples, and S sim The positions x j and time t j The original vector and scalar simulation data at and S mor are the reduced-order vector and scalar data at the same position and time, respectively. and f N (S mor ) are the predicted values ​​of the neural network model for the reduced-order vector and scalar data, respectively; When L(N) reaches the convergence condition, that is, satisfies the threshold θ, the model outputs the final system reduction data as follows: In the formula, and S * (x, t) are the displacement and strain reduction data respectively, argmin N is the set of optimized neural network parameters; Step 7: For the system reduced-order data output in step 6, use the Modelica language to establish a one-dimensional model and realize the fusion modeling with the three-dimensional field data model; combine the spatial distribution characteristics of the three-dimensional coordinate system with the reduced-order data, use 3-5 order modal functions to capture the dynamic characteristics in different directions and dimensions, and obtain field reduced-order data that conforms to the three-dimensional structural response and spatiotemporal evolution laws of the physical system: In the formula, x, y, and z are the distribution of the physical field in the three directions of space with time t. and They are respectively the dynamic distribution representation of reduced-order data of vectors and scalars in three-dimensional visualization models; Step 8: For the three-dimensional structural simulation model in step 4, it provides high-precision mechanism data; the simulation data reduction model in step 5 retains key modes by reducing the dimension and reduces the computational complexity; the neural network training model in step 6 further fits the reduced-order data to improve the response prediction capability under different working conditions and external conditions; the three-dimensional visualization model in step 7 realizes the spatial reconstruction of vector and scalar physical quantities, combined with modal functions for real-time dynamic display; based on interface standards and modeling specifications, the above multi-source heterogeneous models are packaged and integrated to build a digital twin of the physical system; Step 9: The digital twin model constructed in step 8 is operated under different loads, boundary conditions or working conditions, and the sensor measurement data of the cantilever beam structure is collected in real time to characterize the geometric shape and mechanical properties of the cantilever beam structure.

2. The method according to claim 1, characterized in that: (1) Comprehensively considering the mechanical properties of the cantilever beam structure and the influence of the additional materials on its deformation and movement, force sensors, displacement sensors and resistance strain gauges were selected to build a physical entity test system; (2) The pressure sensor signal and the resistance strain gauge signal are converted into digital signals through digital transmitters, and then the RS485 communication standard is used for data transmission, and they are uploaded to the Weiluntong display screen based on the Modbus communication protocol; the LRS-150W single-group output power supply is used to convert 220V AC power into 24V DC power; (3) Use SolidWorks to build a three-dimensional digital model; (4) Static simulation of the three-dimensional model was performed using Ansys Workbench, with the load starting from 0 and increasing to 45 N at intervals of 1 N. The simulation results under 45 load conditions were exported as .rst format files and saved as system data in Excel format using the Export function. In addition, EnSight software was used to convert the result files into field data in .case format for post-processing. (5) Import the system data into the model reduction toolbox of MWORKS-Sysplorer, set the load force as the input variable, and the displacement and strain as the output variables; combined with the vibration equation and boundary conditions of the cantilever beam, the calculation formula of the displacement is converted to: Where V represents the displacement of the cantilever beam at position x at time t, x is the distance from the fixed end of the beam, F(t) is the time-varying external load acting on the free end of the beam, E is the elastic modulus of the beam material, I is the section moment of inertia of the beam, and l is the length of the beam; Select the first three modal components, then m=3 in the above formula, Where ρ is the material density and A is the cross-sectional area; By capturing the dynamic working condition response, the strain calculation formula is transformed into: Where S represents the strain at x on the cantilever beam at time t, y is the vertical distance from the neutral axis on the beam section, and the meanings and values ​​of the parameters E, I, and F(t) are the same as those in formula (1); Select the first three modal components, then n=3 in the above formula, At the same time, read the displacement vector data of each node in the .case file, including the three directions of x, y, and z, as well as the strain and stress scalar data of each unit; select the first three modal functions to reduce the dimension of the field data, and randomly divide it into training set and validation set; (6) Construct a convolutional neural network (CNN) model to train the system data and field data respectively; the number of hidden layers C is initially set to 3, the number of hidden layer neurons Y is 16, 32, and 64 respectively, the activation function is Sigmoid, the loss function is the mean square error, the optimizer uses the adaptive learning rate AdamW, the batch size is 32, the number of iterations is 500, the learning rate is 0.01, the initialization values ​​of the weight parameter w and the bias parameter b are 0.5 and 0.05 respectively, and the convergence threshold θ is 0.005; the sum of the MSE based on displacement and strain L(N) is used as the loss function for neural network training: In the formula, the mapping function Displacement simulation data Strain simulation data S sim , the unit is micro strain με, F = weight of the free end of the beam (kg) × 9.8m / s 2 ; N represents the network model parameter set {C, Y, Batchsize, Epoch, Rate, w, b, θ}; T is the system reduced-order training data sample of displacement and strain; The predicted value of displacement and strain by CNN model and f N (S mor ); When L(N) reaches the convergence condition, that is, the error is minimized, the training is considered to be completed when the real-time loss value (Loss) is less than the threshold value 0.005; the final reduced-order data output by the model is: The training effect of the model is further quantified by calculating the relative error between the original data and the reduced-order data under each load condition. If the relative error is less than 5%, it means that the model training is relatively accurate and the parameter settings are reasonable. After the system data and field data are trained by the optimized neural network model, if the above convergence conditions and error thresholds are met, the reduced-order data containing the physical field distribution under the load will be exported separately for subsequent joint simulation. (7) A comparison model from physical measurement data to simulation reduced-order data is established based on the Modelica language; the MySQL database is accessed through Python code, the mass measured by the force sensor is converted into load force as the input of the reduced-order model, and the DLL function is called using C language code to access the database. The system model outputs the simulated displacement value of the free end of the cantilever beam and the strain data at the four strain gauge positions [x1, x2, x3 and x4] in step (1); the three-dimensional field model generates visual field data: Where x, y, and z are the reduced-order data distribution fields in the length, width, and height directions, respectively. is the time distribution of displacement in three directions based on external loads, is the distribution of strain in three directions as the external load changes.

Images