A digital twin visualization system and method for engineering structure load response

Abstract

This invention discloses a digital twin visualization system and method for the load response of engineering structures. The system includes: a physical information input module for collecting measured physical data during physical tests of beam structures; a virtual modeling and simulation module for constructing a virtual model of the beam structure based on structural geometric data and structural material data, driving the virtual model to perform simulation calculations based on load force data, generating simulation data reflecting the mechanical response of the beam structure, and converting the simulation data into dynamic animation data; and a visualization and dynamic display module for receiving load force data and dynamic animation data, and synchronously displaying the deformation process and mechanical response distribution of the virtual model in a dynamic visualization format. This invention deeply integrates physical load tests with finite element simulation, achieving real-time interaction and synchronous visualization of physical data and virtual models, thus improving the analysis efficiency and intuitiveness of beam structure tests.

Description

Technical Field

[0001] This invention relates to the field of structural testing and computer simulation technology, and in particular to a digital twin visualization system and method for the load response of engineering structures. Background Technology

[0002] Structural load tests are generally divided into static load tests and dynamic load tests. Static load tests establish a finite element model of the beam and, combined with prior data such as load conditions, loading methods, and measuring point layout, calculate the mechanical responses of the beam structure, including internal forces, strain, and deflection. By comparing the theoretical calculations with the measured values, the actual load-bearing capacity of the beam is evaluated. Dynamic load tests are mainly used to evaluate the stiffness, deformation characteristics, dynamic performance, and operational status of the beam structure. In structural inspection and evaluation, the assessment of mechanical performance is directly related to the safety of structural use. However, some mechanical performance indicators are difficult to test, and the testing work generally faces challenges such as high cost, long cycle, limited data acquisition, and even difficulty in obtaining some test results or the existence of significant dispersion and randomness.

[0003] Finite element method (FEM) simulation technology can effectively supplement physical test results. Its core idea lies in describing complex mathematical problems in a finite discrete spatial scale. By dividing the problem into independent elements and approximating the overall solution with element numerical solutions, it achieves modeling and analysis of multi-feature structural problems. Traditional beam structure simulation usually adopts a parallel mode of finite element numerical analysis and physical load testing. However, finite element simulation requires the establishment of a parametric model and offline calculation, while load testing relies on sensors such as strain gauges and displacement gauges to collect measured data. The two are often carried out independently, making real-time interactive verification impossible, which reduces the confidence level of simulation results compared to experimental tests. Especially for beam structures with complex configurations and stresses, this fragmented mode can hardly meet the higher requirements of modern engineering for synchronous mapping and dynamic verification of virtual and real data.

[0004] Against this backdrop, digital twin technology can provide the ability to dynamically interact with virtual models and physical entities. By constructing a virtual model that is updated synchronously with the physical test process, it can realize the real-time mapping of load-response data, providing an effective solution to the problem of the disconnect between simulation calculations and measured data in traditional tests. Furthermore, it can use machine learning algorithms to dynamically correct the material parameters and boundary conditions of the finite element model, thereby continuously improving the accuracy of virtual simulation results. However, existing technologies lack the ability to systematically apply digital twin technology to the visualization and reconstruction of the stress test process of beam structures. Summary of the Invention

[0005] In view of the shortcomings of the prior art, the present invention provides a digital twin visualization system and method for the load response of engineering structures, which solves the problems in the prior art where finite element simulation and physical load test are isolated and asynchronous, making it impossible to achieve real-time interactive verification and difficult to meet the requirements of beam structure test for virtual-real synchronization and rapid feedback.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] In a first aspect, the present invention provides a digital twin visualization system for the load response of engineering structures, comprising:

[0008] The physical information input module is used to collect measured physical data during the physical testing of the beam structure;

[0009] The measured physical data includes structural geometric data, structural material data, structural strain data, and load force data;

[0010] The virtual modeling and simulation module, connected to the physical information input module, is used to construct a virtual model of the beam structure based on structural geometric data and structural material data. The virtual model is driven to perform simulation calculations through the load force data to generate simulation data reflecting the mechanical response of the beam structure, and the simulation data is converted into dynamic animation data.

[0011] The visualization and dynamic display module is connected to the physical information input module and the virtual modeling and simulation module, respectively. It is used to receive the load force data and dynamic animation data, and display them synchronously in a dynamic visualization form to show the deformation process and mechanical response distribution of the virtual model.

[0012] In an optional implementation, the physical information input module includes:

[0013] Several strain sensors are arranged on the surface of the beam structure to convert the surface mechanical strain of the beam structure into a differential voltage signal in order to obtain a structural strain simulation signal.

[0014] Several force sensors are installed at the loading points or supports of the beam structure to convert the concentrated loads applied externally to the beam structure into voltage signals in order to obtain simulated load force signals.

[0015] The signal processing unit, connected to the strain sensor and the force sensor, is used to perform analog-to-digital conversion on the acquired structural strain simulation signal and load force value simulation signal to obtain structural strain data and load force value data.

[0016] The structural data acquisition unit is used to acquire the geometric and material data of the beam structure.

[0017] In one optional implementation, the virtual modeling and simulation module includes a simulation model database, a parametric modeling unit, a finite element analysis engine, a data matching unit, and an animation frame generation unit.

[0018] The simulation model database is connected to the parametric modeling unit and is used to store parametric model templates for beam structures.

[0019] The parametric modeling unit is connected to the physical information input module and the finite element analysis engine respectively, and is used to call the parametric model template of the beam structure according to the structural geometric data and structural material data to construct a virtual model corresponding to the physical test beam structure;

[0020] The finite element analysis engine is connected to the physical information input module and the data matching unit, respectively. It is used to receive the load force data, apply it as boundary load to the virtual model, and solve the mechanical equilibrium equation based on the linear elastic finite element method to obtain the simulation data of the beam structure. It is also used to receive the material parameter correction amount output by the data matching unit to dynamically update the virtual model and recalculate the simulation data of the beam structure.

[0021] The simulation data includes full-field deflection distribution data, full-field strain distribution data, and full-field nodal displacement data;

[0022] The data matching unit has a built-in multimodal machine learning model, which is connected to the physical information input module and the finite element analysis engine respectively. It is used to establish a mapping relationship between measured physical data and virtual model parameters based on the multimodal machine learning model, so as to realize the dynamic association and alignment between measured physical data and virtual model parameters. It is also used to calculate the material parameter correction amount in real time, use Kalman filtering to recursively update the correction amount, and send the updated material parameter correction amount to the finite element analysis engine in real time.

[0023] The animation frame generation unit is connected to the finite element analysis engine and is used to generate a continuous keyframe sequence through linear interpolation based on the simulation data of the beam structure to form dynamic animation data.

[0024] The dynamic animation data includes full-field deflection cloud map, full-field strain cloud map, and mechanical response animation frames reflecting the deformation process of the virtual model.

[0025] In one optional implementation, the multimodal machine learning model includes:

[0026] The graph neural network model is used to construct a mapping graph structure with the arrangement positions of the force sensor and strain sensor and the finite element nodes of the virtual model as graph nodes, and to interpolate the measured physical data into the full-field distribution data of the virtual model through graph convolution operation, so as to establish the spatial mapping relationship between the measured data and the finite element nodes in the virtual model.

[0027] A temporal convolutional network model is used to extract the temporal features of the measured physical data and the simulation data, establish the temporal alignment relationship between the two types of temporal data, eliminate the temporal offset between the measured physical data and the simulation data, and achieve temporal alignment.

[0028] A Bayesian neural network model is used to calculate the material parameter correction in real time, and the correction is recursively updated using Kalman filtering. The updated material parameter correction is then fed back to the finite element analysis engine in real time.

[0029] The material parameter corrections include corrections for elastic modulus and Poisson's ratio.

[0030] In one optional implementation, the structural data acquisition unit supports at least one of manual input, image recognition input, and computer-aided design drawing import.

[0031] In one optional implementation, the visualization dynamic display module includes a display terminal and a data visualization component;

[0032] The data visualization component, connected to the virtual modeling and simulation module, is used to stitch and render continuous frame cloud maps in dynamic animation data to generate full-field deflection dynamic cloud maps, full-field strain dynamic cloud maps, and mechanical response animations.

[0033] The display terminal is connected to the data visualization component and is used to synchronously display the full-field deflection dynamic cloud map, the full-field strain dynamic cloud map, the mechanical response animation, and the load force data.

[0034] Secondly, the present invention provides a digital twin visualization method for the load response of engineering structures, implemented using the digital twin visualization system for the load response of engineering structures as described above, including the following steps:

[0035] A. Obtain measured physical data during the physical testing of the beam structure;

[0036] The measured physical data includes structural geometric data, structural material data, structural strain data, and load force data;

[0037] B. Based on the structural geometry data and structural material data, call the pre-stored parametric model template to generate a virtual model corresponding to the physical test beam;

[0038] C. Apply the load force data as boundary loads to the virtual model, and solve the mechanical equilibrium equations based on the linear elastic finite element method to generate simulation data that reflects the mechanical response of the beam structure.

[0039] D. Based on a multimodal machine learning model, construct a spatial-temporal mapping relationship between measured physical data and virtual model parameters, calculate the material parameter correction amount to dynamically update the virtual model, and re-execute the simulation calculation to obtain updated simulation data;

[0040] E. Convert the updated simulation data into continuous dynamic animation data, and synchronously display the full-field deflection dynamic cloud map, full-field strain dynamic cloud map, mechanical response animation and load force data of the beam structure, thus restoring the entire process of the physical test of the beam structure.

[0041] In an optional implementation, in step D, the multimodal machine learning model includes:

[0042] The graph neural network model is used to construct a mapping graph structure with the arrangement positions of the force sensor and strain sensor and the finite element nodes of the virtual model as graph nodes, and to interpolate the measured physical data into the full-field distribution data of the virtual model through graph convolution operation, so as to establish the spatial mapping relationship between the measured data and the finite element nodes in the virtual model.

[0043] A temporal convolutional network model is used to extract the temporal features of the measured physical data and the simulation data, establish the temporal alignment relationship between the two types of temporal data, eliminate the temporal offset between the measured physical data and the simulation data, and achieve temporal alignment.

[0044] A Bayesian neural network model is used to calculate the material parameter correction in real time, and the correction is recursively updated using Kalman filtering. The updated material parameter correction is then fed back to the finite element analysis engine in real time.

[0045] The material parameter corrections include corrections for elastic modulus and Poisson's ratio.

[0046] The beneficial effects of the embodiments provided by the present invention include:

[0047] This invention deeply integrates the previously fragmented experimental measurements and numerical simulations by constructing a full-link collaborative architecture that combines physical information input, virtual modeling and simulation, and dynamic visualization. It utilizes a multimodal machine learning model in the data matching unit to establish a spatial mapping relationship between sensor measurement data and finite element nodes, achieving temporal alignment and dynamic correction of material constitutive parameters. This overcomes the technical bottleneck of asynchronous simulation calculations and physical experiments in existing technologies, which prevent real-time interactive verification. It truly achieves virtual-real synchronization and complementary advantages between the physical experimental process and virtual simulation analysis.

[0048] This invention transforms the full-field deflection and strain distribution data generated by the finite element analysis engine into continuous dynamic animation data through an animation frame generation unit, supporting multi-scale visualization observation from local micro-strain to macro-structural deformation. The visualization dynamic display module synchronously displays the full-field deflection dynamic cloud map, the full-field strain dynamic cloud map, the dynamic animation of structural deformation morphology, and the load force parameters that strictly correspond to each frame in a screen layout. This compresses the traditional long-cycle process of test-simulation-verification into an integrated real-time feedback process, effectively solving the problems of asynchronous data updates, low analysis efficiency, and delayed visualization in existing technologies.

[0049] This invention utilizes a data matching unit to dynamically update the material parameters of the virtual model during the experimental process, establishing a complementary verification mechanism between virtual simulation and physical experimentation. This reduces reliance on traditional repetitive graded loading tests, thereby lowering experimental costs and shortening the on-site testing cycle. Through real-time data fusion and automated analysis processes, this invention achieves refined synchronous observation from macroscopic structural deformation to microscopic cross-sectional strain, significantly improving the realism, on-site presence, and overall intelligence level of the perception of the evolution of the stress performance of the beam structure. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other embodiments can be obtained based on these drawings.

[0051] Figure 1 A schematic diagram of the digital twin visualization system for the load response of engineering structures in the embodiments of this specification is shown.

[0052] Figure 2 This specification shows a schematic diagram of the architecture of a digital twin visualization system for the load response of an engineering structure in an embodiment of this specification;

[0053] Figure 3A flowchart illustrating the digital twin visualization method for the load response of engineering structures in the embodiments of this specification is shown. Detailed Implementation

[0054] The features and exemplary embodiments of various aspects of the present invention will now be described in detail. In the following detailed description, numerous specific details are set forth in order to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that the invention may be practiced without requiring some of these specific details. The following description of embodiments is merely intended to provide a better understanding of the invention by illustrating examples of the invention.

[0055] In some of the processes described in the specification, claims, and accompanying drawings of this invention, multiple operations appear in a specific order. However, it should be clearly understood that these operations may not be performed in the order they appear herein, or they may be performed in parallel. The operation numbers, such as S1, S2, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be performed sequentially or in parallel.

[0056] Example 1

[0057] like Figures 1-2 As shown, this invention provides a digital twin visualization system for the load response of engineering structures. When the beam structure is in the linear elastic working stage, the system drives a virtual model to perform simulation calculations by real-time acquisition of measured data from physical experiments, and presents the simulation results in a dynamic visualization form. This allows experimental personnel to intuitively observe the deformation and mechanical response distribution of the beam structure during loading, including:

[0058] The physical information input module is used to collect measured physical data during the physical testing of the beam structure;

[0059] The measured physical data includes structural geometric data, structural material data, structural strain data, and load force data.

[0060] The virtual modeling and simulation module, connected to the physical information input module, is used to construct a virtual model of the beam structure based on structural geometric data and structural material data. The virtual model is driven to perform simulation calculations through the load force data to generate simulation data reflecting the mechanical response of the beam structure, and the simulation data is converted into dynamic animation data.

[0061] The visualization and dynamic display module is connected to the physical information input module and the virtual modeling and simulation module, respectively. It is used to receive the load force data and dynamic animation data, and display them synchronously in a dynamic visualization form to show the deformation process and mechanical response distribution of the virtual model.

[0062] For example, the physical information input module includes several strain sensors, several force sensors, a signal processing unit, and a structural data acquisition unit;

[0063] Specifically, several strain sensors are arranged on the surface of the beam structure to convert the surface mechanical strain of the beam structure into a differential voltage signal in order to obtain a structural strain simulation signal.

[0064] Optionally, the strain sensor may be a strain gauge based on the Wheatstone bridge principle. The Wheatstone bridge consists of four resistors and measures the change of physical quantity by utilizing the change of resistance. When the beam is subjected to force and deformation, the strain gauge attached to the surface of the test beam generates a resistance change proportional to the strain and outputs a differential voltage signal.

[0065] In terms of placement, based on the theoretical stress analysis of the beam structure, strain gauges are preferentially placed in the regions of maximum tensile strain and maximum compressive strain expected to occur under specific load conditions.

[0066] As an example arrangement, the region of maximum tensile strain may include the lower surface of the mid-span section of the simply supported beam, and the region of maximum compressive strain may include the upper surface of the support section of the simply supported beam. It should be noted that for statically indeterminate beams or complex beam structures, the location of the control sections should be determined based on the specific stress calculation results, and is not limited to the above example.

[0067] Several force sensors are installed at the loading points or supports of the beam structure to measure the external concentrated load applied to the loading points or the support reaction force generated at the supports, and convert the measured force values ​​into voltage signals to obtain simulated load force signals.

[0068] The signal processing unit, connected to the strain sensor and the force sensor, is used to perform analog-to-digital conversion on the acquired structural strain simulation signal and load force value simulation signal to obtain structural strain data and load force value data.

[0069] The load force data includes vertical force data;

[0070] Optionally, the signal processing unit may employ a 24-bit Δ−Σ analog-to-digital converter to convert analog signals into high-resolution digital signals through high-precision quantization, and transmit them to the virtual modeling and simulation module in real time via a data transmission interface.

[0071] The structural data acquisition unit is used to acquire geometric and material data of the beam structure.

[0072] The structural geometric data includes support parameters such as beam length, beam width, beam height, beam cross-sectional shape, beam dimensions, support spacing, and support diameter.

[0073] Structural material data includes mechanical parameters such as the elastic modulus, Poisson's ratio, and density of the materials used in the beam structure.

[0074] In some embodiments, the structural data acquisition unit supports at least one of manual input, image recognition input, and computer-aided design drawing import.

[0075] Regardless of the input method used, all input data must undergo unit standardization conversion and parameter logic verification to ensure data consistency and validity.

[0076] Specifically, when manually inputting structural data, the structural data acquisition unit can be equipped with an electronic digital caliper and a laser rangefinder to obtain the beam structure dimensions, support spacing, and support diameter through a combination of manual measurement and automatic input.

[0077] For image recognition input, the beam size ratio and support position can be automatically obtained by taking a picture. By entering the size of any component, the structural dimensions can be automatically generated according to the proportion shown in the figure.

[0078] For importing computer-aided design drawings, such as CAD drawings and Excel spreadsheets, geometric parameters can be automatically parsed.

[0079] For example, the virtual modeling and simulation module includes a simulation model database, parametric modeling units, a finite element analysis engine, a data matching unit, and an animation frame generation unit;

[0080] Specifically, the simulation model database is connected to the parametric modeling unit to store parametric model templates for beam structures;

[0081] Optionally, the parametric model template for beam structures includes parametric scripts for common geometric features and material properties of beam structures, supporting the rapid generation of virtual models using key parameters such as length and cross-sectional dimensions;

[0082] The simulation model database can also contain parametric APDL scripts for various standard beam components, and integrate the mechanical parameters and load conditions of commonly used materials.

[0083] The parametric modeling unit is connected to the signal processing unit, the structural data acquisition unit, and the finite element analysis engine, respectively. It is used to call the parametric model template of the beam structure based on the structural geometric data and structural material data to construct a virtual model corresponding to the physical test beam structure.

[0084] Optionally, the parametric modeling unit is implemented using a parametric modeling tool, preferably ANSYSDesignModeler, which can automatically generate and update the virtual model of the beam based on a predefined geometric feature parameter system.

[0085] The finite element analysis engine is connected to the physical information input module and the data matching unit, respectively. It is used to receive the load force data, apply it as boundary load to the virtual model, and solve the mechanical equilibrium equations based on the linear elastic finite element method to obtain the simulation data of the beam structure. It is also used to receive the material parameter correction amount output by the data matching unit to dynamically update the virtual model and recalculate the simulation data of the beam structure.

[0086] The simulation data includes full-field deflection distribution data, full-field strain distribution data, and full-field nodal displacement data.

[0087] In some embodiments, the finite element analysis engine integrates the ANSYS Mechanical solver kernel. Under the assumption of small deformation in linear elasticity, it solves the equilibrium equations through the stiffness matrix to extract the vertical displacement components and interpolate and reconstruct the full-field deflection distribution data. Then, based on the element shape function and the strain-displacement matrix, it calculates the element strain from the nodal displacements and interpolates and stitches them together to obtain the full-field strain distribution data. The specific process is as follows:

[0088] The finite element analysis engine discretizes the virtual model and establishes structural equilibrium equations based on the assumption of small deformation in linear elasticity:

[0089]

[0090] In the formula, K is the overall structural stiffness matrix, u is the nodal displacement vector, and F is the external load vector;

[0091] After solving for the displacement vectors of each node in the virtual model, the total nodal displacement data of the virtual model can be obtained.

[0092] The vertical displacement components of the displacement vectors of each node in the virtual model are extracted and reconstructed by interpolation according to the spatial position of the nodes, so as to obtain the full-field deflection distribution data of the beam under the current load.

[0093] After obtaining the displacement vectors of each node in the virtual model, the element displacement field is established based on the element shape function:

[0094]

[0095] Where N is the shape function matrix; Let be the displacement vector of any point within the element; This represents the displacement vector of the element node.

[0096] The element strain vector is obtained by differentiating the element displacement field with respect to spatial coordinates:

[0097]

[0098] Where B is the strain-displacement matrix, This is the element strain vector.

[0099] The unit strain vectors at each node in the virtual model are interpolated and spliced ​​according to the spatial location of the nodes to form full-field strain distribution data.

[0100] The data matching unit has a built-in multimodal machine learning model, which is connected to the physical information input module and the finite element analysis engine respectively. It is used to establish a mapping relationship between measured physical data and virtual model parameters based on the multimodal machine learning model, so as to realize the dynamic association and alignment between measured physical data and virtual model parameters. It is also used to calculate the material parameter correction amount in real time, use Kalman filtering to recursively update the correction amount, and send the updated material parameter correction amount to the finite element analysis engine in real time.

[0101] In this embodiment, the data matching unit dynamically outputs material apparent parameters that are adapted to the current test stage based on the difference between measured data and simulation data within the current load range, and tracks the changes in the actual response of the structure by recalculating the finite element model.

[0102] An animation frame generation unit, connected to the finite element analysis engine, is used to generate a continuous sequence of keyframes through linear interpolation based on the simulation data of the beam structure, so as to form dynamic animation data.

[0103] Optionally, the animation frame generation unit receives and stores the simulation data output by the finite element analysis engine under different load states in a time sequence, and defines the independent calculation result under each load state as a key frame; when new load force data is input and the simulation calculation is completed, the animation frame generation unit appends the new result as the next key frame; subsequently, between all adjacent key frames, a dense and continuous sequence of intermediate frames is constructed based on a linear interpolation algorithm, thereby transforming the discrete static illustration results into a dynamic animation that shows the continuous deformation process of the beam from the initial state to the current load state.

[0104] The keyframe contains the complete geometric shape and mechanical data of the virtual model under that specific load.

[0105] The dynamic animation data includes full-field deflection contour maps, full-field strain contour maps, and mechanical response animation frames that reflect the deformation process of the virtual model.

[0106] Among them, the mechanical response animation frame is generated based on the full-field node displacement data of the virtual model. It is a continuous image of the geometric shape of the virtual model itself changing with the load, showing the macroscopic deformation of the structure such as bending and torsion, which is different from the cloud map that represents the magnitude of physical quantities through color mapping.

[0107] In some embodiments, the multimodal machine learning model includes:

[0108] A graph neural network model is used to construct a measured-simulation spatial mapping graph structure using the arrangement positions of the force sensor and strain sensor and the finite element nodes of the virtual model as graph nodes, and the spatial adjacency relationship between the graph nodes and the mesh topology connection relationship between the finite element nodes of the model as edges. The measured physical data is interpolated into the full-field distribution data of the virtual model through graph convolution operation to establish the spatial mapping relationship between the measured data and the finite element nodes in the virtual model.

[0109] The temporal convolutional network model is used to extract the temporal features of the full-field distribution data and the simulation data, establish the temporal alignment relationship between the two types of temporal data, and output the aligned measured response data and simulation response data to eliminate the time offset between the measured physical data and the simulation data and achieve time alignment.

[0110] The time offset may originate from factors such as transmission delay of the data acquisition system, buffer queue delay, or sensor response lag. Temporal convolutional networks learn to output the time offset and align the data.

[0111] The Bayesian neural network model receives the aligned measured response data and simulated response data output by the temporal convolutional network model, and outputs the material parameter correction amount based on the deviation between the two. At the same time, the Bayesian neural network model also receives the current load, boundary condition information, and the material parameter estimate value of the previous time step to calculate the posterior probability distribution of the material parameter correction amount. The posterior probability distribution is used as the observation value, and the material parameter correction amount is recursively updated using Kalman filtering. The updated material parameter correction amount is sent to the finite element analysis engine in real time to update the material constitutive parameters of the virtual model.

[0112] The material parameter corrections include corrections for elastic modulus and Poisson's ratio.

[0113] In this embodiment, the specific process of the multimodal machine learning model working together is as follows:

[0114] Collect measured physical data and obtain the sensor placement locations and model node coordinates in the virtual model;

[0115] The graph neural network model uses sensor placement locations and finite element nodes as graph nodes, assigns node features to each graph node, and outputs the predicted response values ​​on each finite element node through graph convolution operations, thus expanding the discrete measurement point data into the full-field distribution data of the virtual model.

[0116]

[0117] In the formula, Let be the node features of the i-th graph node; Represents the spatial coordinates of graph nodes; These represent strain characteristics, displacement characteristics, and force characteristics, respectively.

[0118] The temporal convolutional network model receives the measured full-field response time series mapped by the graph neural network and the simulated response time series output by the finite element analysis engine. It extracts temporal features and outputs time offsets, performs alignment correction on the two types of data, and outputs aligned measured response data and simulated response data.

[0119] The Bayesian neural network model calculates the material parameter correction based on the deviation between the aligned measured response data and the simulated response data, and calculates the posterior probability distribution of the material parameter correction based on the measured response data, the simulated response data, the current load and boundary condition information, and the model parameter estimates from the previous time step.

[0120] The Bayesian neural network uses the output posterior probability distribution as an observation, and uses Kalman filtering for recursive updates to obtain the updated material parameter corrections, which are then sent to the finite element analysis engine to update the virtual model. This improves the dynamic mapping accuracy between measured physical data and the virtual model, and enhances the ability of the full-field deflection, full-field strain, and dynamic visualization results to reproduce the response of the real structure.

[0121] The finite element analysis engine recalculates the full-field deflection and strain distribution data based on the updated constitutive parameters and outputs them to the animation frame generator and display module, thereby achieving dynamic and high-precision mapping from physical measurement data to the virtual model.

[0122] For example, the visualization dynamic display module includes a display terminal and a data visualization component;

[0123] The data visualization component, connected to the virtual modeling and simulation module, is used to stitch and render continuous frame cloud maps in dynamic animation data to generate full-field deflection dynamic cloud maps, full-field strain dynamic cloud maps, and mechanical response animations.

[0124] Specifically, the splicing rendering process refers to the data visualization component continuously playing a sequence of single-frame static cloud images at a rate that matches the refresh rate of the display terminal, thereby forming a smooth and continuous dynamic visualization effect.

[0125] The display terminal is connected to the data visualization component and is used to synchronously display the full-field deflection dynamic cloud map, the full-field strain dynamic cloud map, the mechanical response animation, and the load force data.

[0126] Specifically, the display terminal uses an LCD screen as the hardware carrier and adopts a dual dynamic graph + force value panel layout on the same screen. The force value panel is used to display the vertical force value data generated based on the load force value data in real time in the form of numerical values ​​or curves. When displayed synchronously, each frame of dynamic cloud map corresponds strictly to the vertical force value data at the corresponding moment, intuitively revealing the mapping relationship between load-deformation-strain.

[0127] It should be noted that the term "real-time" in this invention should be understood as online synchronous processing during physical experiments, with end-to-end latency controllable to the level of several seconds. The system can control the scale of the finite element model and perform offline training and online inference on the machine learning model, so that the overall process can meet the requirements of near real-time visualization.

[0128] Example 2

[0129] like Figure 3 As shown, this invention provides a digital twin visualization method for the load response of engineering structures, implemented using the digital twin visualization system for the load response of engineering structures as described in Example 1, including the following steps:

[0130] S1. Obtain measured physical data during the physical test of the beam structure;

[0131] The measured physical data includes structural geometric data, structural material data, structural strain data, and load force data;

[0132] S2. Based on the structural geometry data and structural material data, call the pre-stored parametric model template to generate a virtual model corresponding to the physical test beam.

[0133] S3. Apply the load force data as boundary loads to the virtual model, and solve the mechanical equilibrium equations based on the linear elastic finite element method to generate simulation data that reflects the mechanical response of the beam structure.

[0134] S4. Based on a multimodal machine learning model, construct a spatial-temporal mapping relationship between measured physical data and virtual model parameters, calculate the material parameter correction amount to dynamically update the virtual model, and re-execute the simulation calculation to obtain updated simulation data.

[0135] In some embodiments, in step S4, the multimodal machine learning model includes:

[0136] A graph neural network model is used to construct a graph structure with the arrangement positions of the force sensor and strain sensor as nodes and the finite element mesh topology of the virtual model as edges. The measured physical data is interpolated into the full-field distribution data of the virtual model through graph convolution operation to establish a spatial mapping relationship between the measured data and the finite element nodes in the virtual model.

[0137] A temporal convolutional network model is used to extract the temporal features of the measured physical data and the simulation data, establish the temporal alignment relationship between the two types of temporal data, eliminate the temporal offset between the measured physical data and the simulation data, and achieve temporal alignment.

[0138] A Bayesian neural network model is used to calculate the material parameter correction in real time, and the correction is recursively updated using Kalman filtering. The updated material parameter correction is then fed back to the finite element analysis engine in real time.

[0139] The material parameter corrections include corrections for elastic modulus and Poisson's ratio.

[0140] S5. Convert the updated simulation data into continuous dynamic animation data, and synchronously display the full-field deflection dynamic cloud map, full-field strain dynamic cloud map, mechanical response animation and load force data of the beam structure, thus restoring the entire process of the physical test of the beam structure.

[0141] The above description is merely a preferred embodiment of the present invention. It will be apparent to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects. The scope of the present invention is defined by the appended claims rather than the foregoing description, and thus all variations falling within the meaning and scope of the equivalents of the claims are intended to be included within the present invention.

Claims

1. A digital twin visualization system for the load response of engineering structures, characterized in that, include: The physical information input module is used to collect measured physical data during the physical testing of the beam structure; The measured physical data includes structural geometric data, structural material data, structural strain data, and load force data; The virtual modeling and simulation module, connected to the physical information input module, is used to construct a virtual model of the beam structure based on structural geometric data and structural material data. The virtual model is driven to perform simulation calculations through the load force data to generate simulation data reflecting the mechanical response of the beam structure, and the simulation data is converted into dynamic animation data. The visualization and dynamic display module is connected to the physical information input module and the virtual modeling and simulation module, respectively. It is used to receive the load force data and dynamic animation data, and display them synchronously in a dynamic visualization form to show the deformation process and mechanical response distribution of the virtual model.

2. The visualization system according to claim 1, characterized in that, The physical information input module includes: Several strain sensors are arranged on the surface of the beam structure to convert the surface mechanical strain of the beam structure into a differential voltage signal in order to obtain a structural strain simulation signal. Several force sensors are installed at the loading points or supports of the beam structure to convert the concentrated loads applied externally to the beam structure into voltage signals in order to obtain simulated load force signals. A signal processing unit, connected to the strain sensor and the force sensor, is used to perform analog-to-digital conversion on the acquired structural strain simulation signal and load force value simulation signal to obtain structural strain data and load force value data. The structural data acquisition unit is used to acquire the geometric and material data of the beam structure.

3. The visualization system according to claim 1, characterized in that, The virtual modeling and simulation module includes a simulation model database, parametric modeling units, a finite element analysis engine, a data matching unit, and an animation frame generation unit. The simulation model database is connected to the parametric modeling unit and is used to store parametric model templates for beam structures. The parametric modeling unit is connected to the physical information input module and the finite element analysis engine respectively, and is used to call the parametric model template of the beam structure according to the structural geometric data and structural material data to construct a virtual model corresponding to the physical test beam structure; The finite element analysis engine is connected to the physical information input module and the data matching unit, respectively, to receive the load force data, apply it as boundary load to the virtual model, and solve the mechanical equilibrium equation based on the linear elastic finite element method to obtain the simulation data of the beam structure. It is also used to receive the material parameter correction amount output by the data matching unit to dynamically update the virtual model and recalculate the simulation data of the beam structure; The simulation data includes full-field deflection distribution data, full-field strain distribution data, and full-field nodal displacement data; The data matching unit has a built-in multimodal machine learning model, which is connected to the physical information input module and the finite element analysis engine respectively. It is used to establish a mapping relationship between measured physical data and virtual model parameters based on the multimodal machine learning model, so as to realize the dynamic association and alignment between measured physical data and virtual model parameters. It is also used to calculate the material parameter correction amount in real time, and to recursively update the correction amount using Kalman filtering, and send the updated material parameter correction amount to the finite element analysis engine in real time. The animation frame generation unit is connected to the finite element analysis engine and is used to generate a continuous keyframe sequence through linear interpolation based on the simulation data of the beam structure to form dynamic animation data. The dynamic animation data includes full-field deflection cloud map, full-field strain cloud map, and mechanical response animation frames reflecting the deformation process of the virtual model.

4. The visualization system according to claim 3, characterized in that, The multimodal machine learning model includes: The graph neural network model is used to construct a mapping graph structure with the arrangement positions of the force sensor and strain sensor and the finite element nodes of the virtual model as graph nodes, and to interpolate the measured physical data into the full-field distribution data of the virtual model through graph convolution operation, so as to establish the spatial mapping relationship between the measured data and the finite element nodes in the virtual model. A temporal convolutional network model is used to extract the temporal features of the full-field distributed data and the simulation data, establish the temporal alignment relationship between the two types of temporal data, eliminate the temporal offset between the full-field distributed data and the simulation data, and achieve time alignment. A Bayesian neural network model is used to calculate the material parameter correction in real time, and the correction is recursively updated using Kalman filtering. The updated material parameter correction is then fed back to the finite element analysis engine in real time. The material parameter corrections include corrections for elastic modulus and Poisson's ratio.

5. The visualization system according to claim 2, characterized in that, The structural data acquisition unit supports at least one of the following methods: manual input, image recognition input, and import of computer-aided design drawings.

6. The visualization system according to claim 1, characterized in that, The visualization dynamic display module includes a display terminal and a data visualization component; The data visualization component, connected to the virtual modeling and simulation module, is used to stitch and render continuous frame cloud maps in dynamic animation data to generate full-field deflection dynamic cloud maps, full-field strain dynamic cloud maps, and mechanical response animations. The display terminal is connected to the data visualization component and is used to synchronously display the full-field deflection dynamic cloud map, the full-field strain dynamic cloud map, the mechanical response animation, and the load force data.

7. A digital twin visualization method for the load response of engineering structures, characterized in that, The digital twin visualization system for the load response of engineering structures as described in any one of claims 1 to 6 is used, comprising the following steps: A. Obtain measured physical data during the physical testing of the beam structure; The measured physical data includes structural geometric data, structural material data, structural strain data, and load force data; B. Based on the structural geometry data and structural material data, call the pre-stored parametric model template to generate a virtual model corresponding to the physical test beam; C. Apply the load force data as boundary loads to the virtual model, and solve the mechanical equilibrium equations based on the linear elastic finite element method to generate simulation data that reflects the mechanical response of the beam structure. D. Based on a multimodal machine learning model, construct a spatial-temporal mapping relationship between measured physical data and virtual model parameters, calculate the material parameter correction amount to dynamically update the virtual model, and re-execute the simulation calculation to obtain updated simulation data; E. Convert the updated simulation data into continuous dynamic animation data, and synchronously display the full-field deflection dynamic cloud map, full-field strain dynamic cloud map, mechanical response animation and load force data of the beam structure, thus restoring the entire process of the physical test of the beam structure.

8. The visualization method according to claim 7, characterized in that, In step D, the multimodal machine learning model includes: The graph neural network model is used to construct a mapping graph structure with the arrangement positions of the force sensor and strain sensor and the finite element nodes of the virtual model as graph nodes, and to interpolate the measured physical data into the full-field distribution data of the virtual model through graph convolution operation, so as to establish the spatial mapping relationship between the measured data and the finite element nodes in the virtual model. A temporal convolutional network model is used to extract the temporal features of the measured physical data and the simulation data, establish the temporal alignment relationship between the two types of temporal data, eliminate the temporal offset between the measured physical data and the simulation data, and achieve temporal alignment. A Bayesian neural network model is used to calculate the material parameter correction in real time, and the correction is recursively updated using Kalman filtering. The updated material parameter correction is then fed back to the finite element analysis engine in real time. The material parameter corrections include corrections for elastic modulus and Poisson's ratio.

Images