Large vertical ring counterforce device safety state visual system and implementation method

Abstract

The application discloses a large vertical annular counterforce device safety state visualization system and an implementation method, and relates to the field of large-scale test devices. The system comprises a physical layer, a perception layer, a twin model layer, a twin data layer and a visualization deduction layer. The perception layer collects multi-dimensional working condition data of the counterforce device in a running process in real time. The twin model layer integrates geometric, physical, behavior and rule models. The rule model is embedded with a mechanical response agent model based on a multi-layer perception network. Forward reasoning is performed by using finite element training data to realize online inversion of a global node stress and displacement prediction vector of the counterforce device. The visualization deduction layer maps the inversion data into a dynamic visualization cloud map in real time by three-dimensional grid reconstruction and barycentric coordinate interpolation technology, and synchronously displays the cloud map on a modularized interactive interface of a digital monitoring terminal. The application effectively solves the problems of incomplete and lagging stress state monitoring of large test devices, and realizes real-time quantitative evaluation and early warning of structural safety state.

Description

Technical Field

[0001] This invention relates to the field of tunnel engineering testing technology, and in particular to a safety status visualization system for a large vertical circumferential reaction device and a method for implementing the safety status visualization system for a large vertical circumferential reaction device. Background Technology

[0002] In the field of tunnel engineering, vertical tunnel real fire thermal coupling test systems have become an indispensable research platform for in-depth research into the evolution of the mechanical properties of lining structures under extreme high-temperature environments. In this test system, the reaction device, as the core load-bearing structure, must withstand an ultimate circumferential load of up to 65,000 kN under the coupled conditions of simulated real fire and high-pressure loads. Due to the complexity of the reaction device's structural system, the variety of loading conditions, and its service environment involving highly coupled multi-physics fields, the dynamic perception of its structural state and its safe operation and maintenance have become crucial aspects for ensuring the smooth conduct of the test.

[0003] However, under current technological conditions, monitoring the operational status of such large, heavy-duty reaction devices still faces the following technical bottlenecks:

[0004] 1. Monitoring methods suffer from localized perception and spatial blind spots. Traditional structural monitoring methods mainly rely on discretely distributed strain gauges or displacement gauges at key locations of the reaction device. Limited by the complex physical layout conditions on site and the limited space of the test system, this type of point-based monitoring can only obtain discrete data from local measuring points. It cannot truly reproduce the evolution characteristics of the global stress and displacement fields of the reaction device under extreme heavy loads and complex stress conditions, resulting in significant blind spots in the overall structural safety assessment.

[0005] 2. The high-dimensional nonlinearity of structural response leads to lag in state identification and inversion. Traditional mechanical analysis methods or simple linear sensor feedback are insufficient for real-time evaluation of structural performance during the experimental process. Furthermore, due to insufficient data processing dimensions and low information utilization, existing safety early warning methods are unable to achieve rapid decision-making and response in the event of sudden abnormal situations.

[0006] 3. Lack of intuitive physical mapping leads to significant challenges in operation and maintenance decision-making. Existing monitoring systems mostly use digital reports or discrete curves for information display, lacking a visual interactive mechanism that deeply integrates multi-source sensor data with physical structures. Because it is impossible to build a real-time synchronized digital twin model in the digital space, test personnel find it difficult to intuitively observe the evolution of the mechanical response characteristics of the reaction device. This results in poor intuitiveness in assessing safety risks during large-scale tests, hindering intelligent scientific decision-making. Summary of the Invention

[0007] This invention aims to at least partially solve one of the technical problems in related technologies. Therefore, the first objective of this invention is to propose a safety status visualization system for a large-scale vertical circumferential reaction device. By constructing a high-fidelity digital twin model layer and introducing a mechanical response proxy model, it achieves full-domain perception, real-time inversion, and three-dimensional dynamic visualization of the structural state under heavy load, providing reliable technical support for the digital transformation and intelligent safety monitoring of large-scale experimental systems.

[0008] The second objective of this invention is to provide an implementation method for a safety status visualization system for a large vertical circumferential reaction device.

[0009] To achieve the above objectives, the first aspect of the present invention proposes a safety status visualization system for a large vertical circumferential reaction device, comprising a physical layer, a perception layer, a twin model layer, a twin data layer, and a visualization and deduction layer.

[0010] Among them, the physical layer is the physical entity of the reaction device, which serves as the data acquisition object of the sensing layer;

[0011] The perception layer includes a sensor array deployed on the physical layer, which is used to collect multi-dimensional working condition data and local feature data of the reaction device in real time during operation, and send the collected signals to the twin data layer.

[0012] The twin data layer includes a data storage and interaction module, which is used to establish data interaction between the perception layer, the twin model layer and the visualization and inference layer, and to store the physical parameters and dynamic operation data of the reaction device;

[0013] The twin model layer is connected to the twin data layer to construct a digital twin of the reaction device, and to invert the stress and displacement indices of the reaction device based on the retrieved multi-dimensional working condition data.

[0014] The visualization inference layer is connected to the twin data layer, and the three-dimensional topological mesh model of the twin model layer is called based on the twin data layer to transform the global node prediction data obtained by inversion into a dynamic visualization cloud map through a spatial mapping algorithm.

[0015] In addition, the large-scale vertical circumferential reaction device safety status visualization system according to the above embodiments of the present invention may also have the following additional technical features:

[0016] According to one embodiment of the present invention, the twin model layer integrates a geometric model, a physical model, a behavioral model, and a rule model;

[0017] Among them, the geometric model describes the topology of the reaction device, the physical model gives the model dynamic properties, the behavioral model defines the action response logic, and the rule model embeds a mechanical response proxy model, which is used to invert the stress and displacement index of the reaction device based on multi-dimensional working condition data.

[0018] According to one embodiment of the present invention, constructing a twin model layer includes the following steps:

[0019] S1, Constructing the geometric model: Extract the geometric features of the reaction device using 3D modeling software, perform lightweight processing through a mesh reduction algorithm, and import it into the graphics rendering engine to perform geometric repair.

[0020] S2, Constructing the physical model: Configure rigid body components, collider components, and physical material properties for the geometric model in the graphics rendering engine to reproduce the mass distribution, collision boundaries, and friction characteristics of the reaction device;

[0021] S3, Constructing a Behavioral Model: Define the action response logic of the twin model and describe the spatial pose evolution of the reaction device during operation through matrix transformation;

[0022] S4, Constructing a rule model: An embedded mechanical response proxy model based on a multilayer perceptron network is used to perform a nonlinear mapping from the working condition input vector to the global node prediction vector.

[0023] According to an embodiment of the present invention, constructing a mechanical response proxy model for a twin model layer includes the following steps:

[0024] S41. A refined finite element model was established based on the geometric characteristics, material constitutive model and contact relationship of the reaction device, and mesh sensitivity analysis and experimental data comparison verification were carried out.

[0025] S42 uses the applied load and loading angle as input variables and employs a space-filling sampling algorithm to extract simulation working points within a preset factor level range.

[0026] S43 utilizes automated calculation scripts to drive the finite element model to perform batch calculations on simulation conditions, extracts stress and displacement response data of all nodes of the reaction device, and constructs a structural mechanical performance characteristic dataset.

[0027] S44 uses a structural mechanics performance feature dataset to train a multilayer perceptron network, and introduces random Gaussian noise at the input to perform robust training. The trained mechanical response surrogate model is then calibrated using physical experimental data to ensure that the global prediction error range is ≤5%.

[0028] S45 deploys the trained mechanical response proxy model in the rule model of the twin model layer, using the real-time collected load and loading angle as input vectors, and realizes the inversion output of the stress prediction vector and displacement prediction vector of the global nodes of the reaction device through the online inversion calculation logic of the proxy model.

[0029] According to one embodiment of the present invention, the online inversion calculation logic performed by the mechanical response proxy model of the twin model layer is as follows:

[0030] The input sequence of the current applied load and loading angle transmitted from the twin data layer is fed into the multilayer perceptron network. The feature transfer and state update formulas for each hidden layer node are as follows:

[0031] ;

[0032] In the formula, For the first The output tensor of the layer, This is the input sequence for the previous layer. This represents the mapping weight matrix that is fixed during the training phase. For physical bias terms, It is a nonlinear activation function; the output layer decouples the stress prediction vector and displacement prediction vector of each node in the entire domain.

[0033] According to one embodiment of the present invention, the transmission architecture of the twin data layer includes a data acquisition layer, a data driving layer, a data storage layer, and a data application layer; wherein,

[0034] The data acquisition layer obtains real-time data on the operation of the reaction device through the perception layer; the data-driven layer calls the proxy model through standardized software interfaces and data interaction protocols; the data storage layer uses a relational database as the underlying storage management center to uniformly manage the measured data and the predicted data of the proxy model; the data application layer drives the forward inference of the proxy model and the dynamic refresh of the visualization cloud map synchronously by building a cross-platform interface between the database, the 3D graphics rendering engine and the programming environment.

[0035] According to one embodiment of the present invention, the rendering logic of a dynamic visualization cloud map includes:

[0036] Extract the element topology information and node coordinates of the finite element model, divide the non-triangular elements into triangular patches, and reconstruct to generate a three-dimensional topological mesh model;

[0037] The received predicted values ​​of the reaction device nodes are normalized, and a linear gradient mapping relationship between the predicted values ​​and the color space is established.

[0038] Color rendering of pixels within an image is performed using a barycentric coordinate interpolation algorithm. The interpolation calculation model is defined as follows:

[0039] ;

[0040] In the formula, For any point within the sheet Color value calculation model, , and The color mapping value of the face vertex. , and For point The weights of the centroid coordinates within the patch, and satisfying... .

[0041] According to one embodiment of the present invention, the visualization simulation layer is configured with a visualization window integrated into a digital monitoring terminal, and the components integrated inside the visualization window include:

[0042] The functional configuration unit includes a data acquisition module, an operation status monitoring module, a fault early warning module, and an information management module; among them, the operation status monitoring module is used to drive the twin model layer to map and display the mechanical performance of the reaction device; the fault early warning module is used to trigger alarm information based on a preset safety threshold;

[0043] The dynamic rendering and display unit integrates dynamic curve drawing components and cloud map rendering engine. It is used to load modular interactive interfaces according to the test process and render and display the predicted data of the entire domain nodes of the reaction device and the sensor operating status in real time.

[0044] The human-computer interaction control unit is equipped with an event listening mechanism to capture physical operation commands triggered by interface interaction elements and convert them into system logic trigger signals to realize remote control mapping of the loading angle and distance adjustment of the reaction force device.

[0045] The status visualization feedback unit is used to dynamically update the status indication information of key nodes through a visualization window based on logical judgment of working condition signals and posture sensing data, so as to realize real-time progress prompts during the test phase.

[0046] According to one embodiment of the present invention, the end-to-end response time of a single-frame dynamic visualization cloud map is no more than 2 seconds.

[0047] To achieve the above objectives, a second aspect of the present invention provides an implementation method for a safety status visualization system for a large-scale vertical circumferential reaction device. The method employs the safety status visualization system for a large-scale vertical circumferential reaction device as described in the above embodiment, and includes the following steps:

[0048] Step 1: Deploy sensor groups in the key stress areas of the reaction device and establish a real-time data channel with the data-driven layer through an industrial gateway.

[0049] Step 2: Launch the visualization analysis platform, load the pre-trained surrogate model weights, and establish a connection with the relational database;

[0050] Step 3: During the experiment, the actuator operating condition signals are captured in real time, and the agent model is driven to perform forward inference calculations to obtain the mechanical response field data of all nodes of the device.

[0051] Step 4: Map the prediction results of the proxy model to the 3D topological mesh model in real time, generate a dynamic cloud map in the visualization window, and execute real-time warnings based on the preset safety threshold.

[0052] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0053] 1. Achieved full-domain perception and high-fidelity dynamic visualization of the structural state of a large-scale vertical circumferential loading reaction device. This invention overcomes the limitation of large and complex experimental devices relying solely on a limited number of discrete sensors for point-based monitoring. By constructing a twin model layer integrating geometric, physical, behavioral, and rule dimensions, and utilizing a spatial mapping algorithm, the multi-source heterogeneous signals from the perception layer are transformed in real time into full-domain stress and displacement cloud maps on a three-dimensional topological mesh model. This allows experimental personnel to intuitively and quantitatively grasp the overall mechanical evolution of the reaction device under a 65000kN ultimate load condition, eliminating the perception blind spots in traditional monitoring methods.

[0054] 2. Overcoming the computational efficiency and accuracy bottlenecks in mechanical state assessment under complex working conditions. Traditional finite element analysis methods, due to their long computation time, cannot meet the needs of real-time on-site monitoring, especially considering the high-dimensional nonlinear response characteristics of reaction devices. This invention introduces a multilayer perceptron network as a mechanical response proxy model, embedding the physical laws obtained from finite element analysis into the network weights. During online inversion, the system only needs to perform millisecond-level forward inference to output high-precision mechanical field data, with a single-frame cloud image end-to-end response time of no more than 2 seconds. This ensures real-time monitoring of the entire heavy-load test process while maintaining finite element-level computational accuracy.

[0055] 3. A robust multi-source heterogeneous data-driven and security early warning mechanism was constructed. The twin data layer and rule model proposed in this invention significantly improve the anti-interference ability and generalization performance of the surrogate model in complex experimental environments by introducing random Gaussian noise during the training phase. Combined with the modular functional interface in the visualization and inference layer, the system can compare sensor measured values ​​with model predicted values ​​in real time, and automatically trigger highlight flashing or alarm pop-ups according to preset safety thresholds, providing reliable technical support for the intelligent operation and maintenance and scientific decision-making of the device.

[0056] 4. This invention achieves digital management and interaction throughout the entire lifecycle of large-scale scientific instruments. By constructing a cross-platform data flow interface, it integrates multiple stages, including pre-experiment preparation, process monitoring, and historical data querying, into a unified digital monitoring terminal. Through smooth cloud map rendering achieved by a centroid coordinate interpolation algorithm and pose transformation synchronization based on a behavior model, a highly synchronized mapping between physical entities and digital space is established. This enables digital closed-loop management from experiment design and real-time monitoring to data archiving, significantly improving the operational efficiency of large-scale experimental systems.

[0057] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0058] Figure 1 A schematic diagram of the architecture of a large vertical circumferential reaction device safety status visualization system according to an embodiment of the present invention;

[0059] Figure 2 This is a schematic diagram illustrating the composition of the twin model layer according to an embodiment of the present invention;

[0060] Figure 3 This is a schematic diagram of the architecture of the twin data layer according to an embodiment of the present invention;

[0061] Figure 4 This is a schematic diagram of a finite element model according to an embodiment of the present invention;

[0062] Figure 5 This is a comparison chart of simulation and measured values ​​of representative measurement points in the finite element model according to an embodiment of the present invention;

[0063] Figure 6 This is a regression diagram of the simulated and measured values ​​of representative measurement points according to an embodiment of the present invention;

[0064] Figure 7 This is a schematic diagram illustrating the accuracy of the prediction task corresponding to stress and displacement according to an embodiment of the present invention.

[0065] Figure 8 This is a spatial distribution feature diagram of the reaction force device and adaptive loading device according to an embodiment of the present invention;

[0066] Figure 9 This is a heatmap of the correlation coefficient matrix of the sample operating conditions in the design space according to an embodiment of the present invention;

[0067] Figure 10 Flowchart for constructing a simulation dataset according to an embodiment of the present invention;

[0068] Figure 11 A flowchart for establishing batch simulation models according to embodiments of the present invention;

[0069] Figure 12 The flowchart for feature processing of the proxy model dataset provided by this invention;

[0070] Figure 13 This is a regression diagram showing the predicted and measured values ​​of a reaction device surrogate model according to an embodiment of the present invention.

[0071] Figure 14 This is a regression diagram of the predicted and measured values ​​of the reaction device surrogate model according to another embodiment of the present invention;

[0072] Figure 15 This is a schematic diagram illustrating the rendering and display principle of triangular facets in a visualized cloud image.

[0073] Figure 16 A visual mesh model of a reaction device according to an embodiment of the present invention;

[0074] Figure 17 This is a 3D visualization rendering of a reaction device mesh model according to an embodiment of the present invention. Detailed Implementation

[0075] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0076] The following description, with reference to the accompanying drawings, illustrates the implementation method of the large-scale vertical circumferential reaction device safety status visualization system and the method for implementing the large-scale vertical circumferential reaction device safety status visualization system proposed in the embodiments of the present invention.

[0077] like Figure 1As shown, the large-scale vertical circumferential reaction device safety status visualization system of this invention includes a physical layer 10, a sensing layer 20, a twin model layer 30, a twin data layer 40, and a visualization and deduction layer 50. The physical layer 10 is the physical entity of the reaction device, serving as the data acquisition object of the sensing layer 20. The sensing layer 20 includes a sensor array deployed on the physical layer 10, used to collect multi-dimensional operating condition data and local feature data of the reaction device in real time during operation, and to send the collected signals to the twin data layer 40. The twin data layer 40 includes a data storage and interaction module for establishing... The perception layer 20, the twin model layer 30, and the visualization and inference layer 50 interact with each other and store the physical parameters and dynamic operation data of the reaction device. The twin model layer 30 is connected to the twin data layer 40 to construct a digital twin of the reaction device and invert the stress and displacement indices of the reaction device based on the retrieved multi-dimensional working condition data. The visualization and inference layer 50 is connected to the twin data layer 40 and, based on the twin data layer 40, calls the three-dimensional topological mesh model of the twin model layer 30 to transform the inverted global node prediction data into a dynamic visualization cloud map through a spatial mapping algorithm.

[0078] Specifically, the working principle of the large-scale vertical circumferential reaction device safety status visualization system of this invention embodiment is based on the construction of a real-time mapping channel between physical entities and virtual models using a five-layer digital twin architecture: the reaction device in the physical layer 10 serves as the monitoring object, and the perception layer 20 collects multi-dimensional working condition data (such as load, loading angle, etc.) in real time through the deployed sensor group, and sends the data to the twin data layer 40 for unified storage and management; the twin model layer 30 retrieves dynamic operation data from the twin data layer 40, and uses the embedded mechanical response proxy model to invert the stress and displacement indices of the entire domain nodes of the reaction device; the visualization inference layer 50 indirectly calls the three-dimensional topological mesh model of the twin model layer 30 through the twin data layer 40, and uses the spatial mapping algorithm to convert the inverted global node prediction data into dynamic color cloud maps in real time, thereby completing a complete mapping closed loop from physical signal acquisition, data analysis to visualization presentation.

[0079] Specifically, during the trial operation, the sensor array of the sensing layer 20 continuously acquires multi-dimensional operating condition data and local feature data of key parts of the reaction device using high-frequency sampling, and pushes the standardized signals to the twin data layer 40 in real time. The twin data layer 40 is a key element driving the visualization monitoring and prediction of the structural state of the reaction device, and mainly consists of system physical parameters, information collected by the sensing layer 20, system dynamic operation data, and environmental parameters, such as... Figure 3As shown, the twin data layer 40 classifies, stores, and indexes the data, while providing a real-time data interface for the twin model layer 30. Based on the current loading conditions, the twin model layer 30 calls the pre-trained mechanical response proxy model to quickly perform forward inference calculations, outputting the stress prediction values ​​and displacement prediction values ​​of all nodes of the entire reaction device within milliseconds. The visualization simulation layer 50 extracts the three-dimensional topological mesh model through the twin data layer 40, and uniformly renders the node prediction values ​​onto the model surface through spatial mapping algorithms such as centroid coordinate interpolation, forming dynamically updated stress cloud maps and displacement cloud maps, which are then displayed in real time on the monitoring terminal.

[0080] According to one embodiment of the present invention, such as Figure 2 As shown, the twin model layer 30 integrates a geometric model, a physical model, a behavioral model, and a rule model. Among them, the geometric model describes the topology of the reaction device, the physical model assigns dynamic attributes to the model, the behavioral model defines the action response logic, and the rule model embeds a mechanical response proxy model, which is used to invert the stress and displacement indices of the reaction device based on multi-dimensional working condition data.

[0081] Specifically, the geometric model describes the geometric parameters, internal structure, and spatial relationships of the experimental system; the physical model describes the physical properties of the reaction device, providing a physical basis for its simulation and analysis; the behavioral model simulates the actions and behaviors of the physical entity based on the geometric model; and the rule model extracts state evolution patterns by mining historical operating data and uses algorithms to establish mapping relationships between data, enabling the prediction of key indicators such as stress and displacement of the reaction device. These models achieve a digital mapping of the reaction device from different dimensions, providing model support for condition monitoring.

[0082] According to an embodiment of the present invention, constructing the twin model layer 30 includes the following steps:

[0083] S1, Constructing the geometric model: Extract the geometric features of the reaction device using 3D modeling software, perform lightweight processing through a mesh reduction algorithm, and import it into the graphics rendering engine to perform geometric repair.

[0084] S2, Build the physical model: Configure rigid body components, collision body components and physical material properties for the geometric model in the graphics rendering engine to reproduce the mass distribution, collision boundaries and friction characteristics of the reaction device.

[0085] S3, Constructing a Behavioral Model: Define the action response logic of the twin model and describe the spatial pose evolution of the reaction device during operation through matrix transformation.

[0086] Specifically, in this embodiment, the pose transformation is described by the following basic transformation matrix:

[0087] (1) Translation transformation matrix:

[0088] ;

[0089] In the formula, Represents the translation matrix; express Directional displacement; express Directional displacement; express Directional displacement.

[0090] (2) Rotation matrix:

[0091] ;

[0092] ;

[0093] ;

[0094] In the formula: Indicates circling Axis rotation matrix; Indicates circling Axis rotation matrix; Indicates circling Axis rotation matrix; Indicates circling The rotation angle of the shaft; Indicates circling The rotation angle of the shaft; Indicates circling The rotation angle of the axis.

[0095] (3) Scaling matrix:

[0096] ;

[0097] In the formula: Represents the scaling matrix; express Axis scaling factor; express Axis scaling factor; express Axis scaling factor.

[0098] For complex actions that require linkage, they are decomposed into combinations of the above basic transformations, and the coordinated linkage of each basic transformation is achieved through matrix operations.

[0099] S4, Constructing a rule model: An embedded mechanical response proxy model based on a multilayer perceptron network is used to perform a nonlinear mapping from the working condition input vector to the global node prediction vector.

[0100] According to an embodiment of the present invention, constructing a mechanical response proxy model for twin model layer 30 includes the following steps:

[0101] S41. A refined finite element model was established based on the geometric characteristics, material constitutive model and contact relationship of the reaction device, and mesh sensitivity analysis and experimental data comparison verification were carried out.

[0102] Specifically, a refined finite element model is established based on the geometric characteristics, material constitutive model, and contact relationship of the reaction device, such as... Figure 4 As shown in the figure. This embodiment uses finite element analysis to simulate the reaction device based on the working condition data collected during the experiment, and compares and analyzes the simulation results with the measured data to ensure the accuracy of the numerical model. The simulation results are as follows: Figure 5 As shown, the measured data are as follows: Figure 6 and Figure 7 As shown.

[0103] S42 uses the applied load and loading angle as input variables and employs a space-filling sampling algorithm to extract simulation working points within a preset factor level range.

[0104] Specifically, an optimized Latin hypercube sampling algorithm is used to extract simulation working points. When extracting these points, not only are the upper and lower limits of single variables constrained, but multivariate coupling constraints must also be met. For example, the load difference between two adjacent loading devices must be less than a preset safety threshold, or when a certain loading angle reaches its extreme value, its corresponding upper load limit is nonlinearly reduced to eliminate invalid sample points that would cause the finite element calculation to fail to converge or do not conform to the logic of actual engineering.

[0105] In this embodiment, taking into account both the fitting accuracy of the surrogate model and the time cost of finite element calculation, 500 effective sample working points were selected to ensure high-dimensional coverage of the design space.

[0106] like Figure 8 As shown, the loading system of the reaction device is symmetrically arranged with the vertical centerline as the reference. This embodiment selects nine design variables, including the load of each group of loading devices (range 0-5000kN) and the loading angle (range -30° to 30°). This embodiment optimizes the sample distribution based on the maximum-minimum distance (Maxmin) criterion, the mathematical expression of which is as follows:

[0107] ;

[0108] In the formula, The number of sample points. For any two sample points , Distance between:

[0109] .

[0110] To verify the uniformity of the distribution of the sample working conditions in the design space and the independence between the features, a correlation analysis was performed on the sample points. Figure 9 The heatmap of the correlation coefficient matrix shows that, except for the autocorrelation term on the diagonal, the correlation coefficients between the input variables are all at a low level, with the largest absolute value being 0.093. This indicates that the linear correlation between the feature variables is weak, which can meet the requirements of the sample distribution quality for subsequent surrogate model training.

[0111] S43 utilizes automated calculation scripts to drive the finite element model to perform batch calculations on simulation conditions, extracting stress and displacement response data of all nodes of the reaction device, and constructing a structural mechanical performance characteristic dataset.

[0112] Specifically, adopting such Figure 10 The process shown constructs a numerical simulation dataset and uses automated scripts (such as secondary development scripts based on high-level programming languages) to drive the finite element kernel to perform batch calculations. The process is as follows: Figure 11 As shown, the coordinates, equivalent displacements, and Mises equivalent stresses of all nodes in the model are extracted, and the above input parameters are aligned with the corresponding mechanical response results and integrated into a structural mechanical performance feature dataset.

[0113] S44 uses a structural mechanics performance feature dataset to train a multilayer perceptron network, and introduces random Gaussian noise at the input to perform robust training. The trained mechanical response surrogate model is then calibrated using physical experimental data to ensure that the global prediction error range is ≤5%.

[0114] Specifically, to effectively avoid overfitting or underfitting during the training of the surrogate model and to improve the model's generalization ability under complex working conditions, the structural mechanical performance feature dataset constructed by S43 is preprocessed, including:

[0115] (1) Data set partitioning: The entire feature dataset is randomly divided into training set and test set in a ratio of 8:2. The test set is isolated and does not participate in model training and hyperparameter adjustment. It is only used to evaluate the generalization ability of the model.

[0116] (2) Feature Standardization: To eliminate the impact of differences in the dimensions of different features on the gradient descent and weight optimization of the neural network, the input variables are standardized using Z-Score before training. It should be noted that the statistical parameters of this standardization process are calculated only based on the training set to avoid data leakage. The expression is as follows:

[0117] ;

[0118] In the formula, Representing the One data point, and These represent the corresponding standard deviation and mean, respectively. This is the standardized data.

[0119] (3) Noise Injection Data Augmentation: To quantitatively evaluate the stability of the surrogate model under complex working conditions, noise is introduced during the training of the neural network model for data augmentation. In this embodiment, Gaussian white noise is superimposed on the input features to simulate the random disturbances introduced by the measurement errors of the sensor in the perception layer 20 in the actual engineering environment, thus playing a regularization role. The random variable corresponding to the Gaussian noise follows a normal distribution, and its probability density function is defined as follows:

[0120] ;

[0121] In the formula, It represents the mean of the noise and reflects the central location of the noise distribution; The standard deviation represents the amplitude of noise fluctuations.

[0122] Since all features have been normalized (mean is 0, variance is 1) before being input into the proxy model, in order not to change the original data distribution characteristics, this embodiment directly superimposes zero-mean Gaussian noise into the standardized feature space, as shown below:

[0123] ;

[0124] In the formula, This is the noise level coefficient, used to control the disturbance amplitude.

[0125] (4) Cross-validation strategy: Based on the processed training set described above, a five-fold cross-validation method is used to train and validate the model. Specifically, the training set is divided into five mutually exclusive subsets of equal size. One subset is selected as the validation set, and the remaining four subsets are used as the training set. The model parameters are optimized by minimizing the prediction error function, and the model's performance on different data subsets is comprehensively evaluated. The feature processing flow is as follows: Figure 12 As shown.

[0126] S45 deploys the trained mechanical response proxy model in the regular model of the twin model layer 30, using the real-time collected load and loading angle as input vectors, and realizes the inversion output of the stress prediction vector and displacement prediction vector of the global nodes of the reaction device through the online inversion calculation logic of the proxy model.

[0127] According to one embodiment of the present invention, the online inversion calculation logic executed by the mechanical response proxy model of the twin model layer 30 is as follows: the input sequence of the current applied load and loading angle transmitted by the twin data layer 40 is fed into the multilayer perceptron network, and the feature transfer and state update formulas of each hidden layer node are as follows:

[0128] ;

[0129] In the formula, For the first The output tensor of the layer, This is the input sequence for the previous layer. This represents the mapping weight matrix that is fixed during the training phase. For physical bias terms, The activation function is nonlinear; the output layer decouples the stress prediction vector and displacement prediction vector for all nodes in the global domain. In this embodiment, the regression scatter distribution of the surrogate model in the stress and displacement prediction task is as follows: Figure 13 , Figure 14 As shown.

[0130] According to one embodiment of the present invention, such as Figure 3 As shown, the transmission architecture of the twin data layer 40 includes a data acquisition layer, a data driving layer, a data storage layer, and a data application layer. Among them, the data acquisition layer acquires real-time data of the reaction device operation through the perception layer 20; the data driving layer calls the proxy model through standardized software interfaces and data interaction protocols; the data storage layer uses a relational database as the underlying storage management center to uniformly manage the measured data and the proxy model prediction data; the data application layer synchronously drives the forward inference of the proxy model and the dynamic refresh of the visualization cloud map by constructing a cross-platform interface between the database, the 3D graphics rendering engine, and the programming environment.

[0131] Specifically, the transmission architecture of the twin data layer 40 achieves efficient data flow and collaborative driving through four layers: the data acquisition layer, as the data entry point, is responsible for acquiring real-time operating condition data and local feature data of the reaction device from the perception layer 20; the data driving layer, through standardized software interfaces and data interaction protocols, calls and manages the mechanical response proxy model in the twin model layer 30, ensuring that the model can perform forward inference as needed; the data storage layer uses a relational database as the underlying storage hub to uniformly store, index, and manage sensor measured data and proxy model prediction results, providing reliable data persistence support for the system; and the data application layer constructs a cross-platform interface between the database, the 3D graphics rendering engine, and the programming environment, realizing deep integration between data, models, and visualization modules. This four-layer architecture together constitutes a complete data link from data acquisition to model inference and then to visualization, laying the foundation for the system's real-time performance, consistency, and scalability.

[0132] In the actual operation, the data acquisition layer first receives real-time operating data (such as load, loading angle, etc.) from the perception layer 20 at high frequency and performs preliminary formatting. Then, the data-driven layer, based on the current operating conditions, calls a pre-trained proxy model through a standardized interface, inputting the real-time data as an input vector and triggering forward inference calculations. Simultaneously, the data storage layer synchronously writes the collected measured data and the model's output prediction data into a relational database for archiving management. Finally, the data application layer extracts the prediction results from the database through a cross-platform interface, drives the 3D graphics rendering engine to dynamically refresh the 3D topological mesh model's cloud map, and presents the refreshed visualization results in real-time on the monitoring terminal. The entire process achieves closed-loop synchronization of "acquisition—storage—inference—rendering," ensuring that the end-to-end response time for a single frame of dynamic visualization cloud map does not exceed 2 seconds.

[0133] According to one embodiment of the present invention, the rendering logic of a dynamic visualization cloud map includes:

[0134] Extract the element topology information and node coordinates of the finite element model, divide the non-triangular elements into triangular patches, and reconstruct to generate a three-dimensional topological mesh model;

[0135] The received predicted values ​​of the reaction device nodes are normalized, and a linear gradient mapping relationship between the predicted values ​​and the color space is established.

[0136] Color rendering of pixels within an image is performed using a barycentric coordinate interpolation algorithm. The interpolation calculation model is defined as follows:

[0137] ;

[0138] In the formula, For any point within the sheet Color value calculation model, , and The color mapping value of the face vertex. , and For point The weights of the centroid coordinates within the patch, and satisfying... .

[0139] Specifically, such as Figures 15 to 17 As shown, the rendering logic of the dynamic visualization cloud map is as follows:

[0140] (1) Reconstructing the 3D topological mesh model. In this embodiment, the finite element model of the reaction device is mainly composed of S4R elements. The spatial coordinates and element topology information of all nodes are extracted using a script, and the quadrilateral elements are divided into triangular patches to generate a mesh object with the same topology as the original model. Figure 15 Taking the quadrilateral facet shown as an example, to divide it into two triangles and ensure it is visible from the outside view, the vertex index sequences of the triangles are [0,3,1] and [1,3,2], respectively. The reconstructed mesh model looks like this. Figure 16 As shown.

[0141] (2) Mesh rendering and interpolation. The node prediction values ​​output by the proxy model are mapped to the mesh vertices. Normalization is performed first:

[0142] ;

[0143] In the formula, The node prediction values ​​output by the surrogate model; This represents the minimum value of all predicted data under a single working condition. This represents the maximum value of all predicted data under a single working condition. These are the standard values ​​after normalization.

[0144] A smooth color transition within a triangular face is achieved using barycentric coordinate interpolation. Let the vertices of the triangular face be A, B, and C, and any point within the face... physical quantity The calculation model is as follows:

[0145] ;

[0146] In the formula, , and The color mapping value of the face vertex. , and For point The weights of the centroid coordinates within the patch, and satisfying... .

[0147] According to an embodiment of the present invention, the visualization simulation layer 50 is configured with a visualization window integrated into a digital monitoring terminal. The components integrated inside the visualization window include: a functional configuration unit, including a data acquisition module, an operation status monitoring module, a fault early warning module, and an information management module; wherein, the operation status monitoring module is used to drive the twin model layer 30 to map and display the mechanical performance of the reaction device; the fault early warning module is used to trigger alarm information based on a preset safety threshold; a dynamic rendering display unit, integrating a dynamic curve drawing component and a cloud map rendering engine, is used to load a modular interactive interface according to the test process, and to render and display the predicted data of all nodes of the reaction device and the sensor operation status in real time; a human-machine interaction control unit, configured with an event listening mechanism, is used to capture the physical operation commands triggered by the interface interaction elements and convert them into system logic trigger signals to realize remote control mapping of the loading angle and distance adjustment of the reaction device; and a status visualization feedback unit, used to dynamically update the status indication information of key nodes through the visualization window according to the logical judgment of the working condition signal and the pose sensing data, to realize real-time progress prompts during the test phase.

[0148] Specifically, the UI interface developed for the visualization simulation layer 50 mainly includes the main interface, preliminary preparation interface, process monitoring interface, status monitoring interface, and historical data query interface, among which:

[0149] The core function of the system's main interface is to comprehensively display the overall operation of the experimental system, providing researchers with a global monitoring perspective. Operators can access the main interface by clicking the digital experimental platform icon. The interface centrally displays the 3D model of the reaction device, the overall operational status of various sensors, and on-site video footage. A navigation bar at the bottom of the interface allows for quick navigation to detailed monitoring pages via icon clicks, enabling phased management of the experimental process.

[0150] The preliminary preparation interface primarily integrates sensor status monitoring and fault diagnosis functions, providing technical support for the smooth conduct of preliminary preparations for the experiment. The interface dynamically presents the sensor number, name, and operating status through a real-time updated "alarm data table," enabling visualized monitoring of sensor operation. When the system detects an anomaly in a sensor, it automatically triggers an alarm mechanism. Operators can quickly locate the physical installation location of the abnormal sensor based on its number in the alarm data table and remotely verify the on-site situation by accessing the corresponding location's video surveillance footage, thus preparing for subsequent experiments. The layout of the interface in the later stages of the experiment remains consistent with the preliminary preparation interface, ensuring the continuity of the operational process.

[0151] The process monitoring interface integrates real-time monitoring data from all sensors during the experiment, primarily including parameters such as displacement and strain of the reaction device, internal temperature of the tube segments, tube segment displacement, and thermal imaging information. Sensor data can be visualized through dynamic line graphs and statistical tables. Users can quickly retrieve data curves for key areas by switching sensor numbers via drop-down menus within the interface. Based on the cross-platform data fusion architecture described above, the real-time operating information of the adaptive loading system is synchronously transmitted to the Python data analysis module, driving a pre-trained surrogate model to rapidly predict the stress and displacement of the reaction device. The prediction results are displayed in the center of the interface as a cloud map, providing intuitive decision support for the safe operation of the experimental system.

[0152] The core function of the status monitoring interface is to display the real-time operational progress of the test system. Through visual prompts of key node statuses, it helps operators quickly grasp the current stage of the test. For example, when the transport trolley reaches the designated turning position or the adaptive loading device adjusts to the preset angle, the corresponding status button on the interface will flash to indicate that the step has been completed. This process continues until the entire test procedure is finished.

[0153] After clicking the "Animation Demonstration" button on the main interface, the system will automatically play a 3D animation of the experimental system's operation, fully presenting the overall operation process of the experimental device. This animation is mainly used to help researchers quickly familiarize themselves with the operating principles and procedures of the experimental system, providing a cognitive foundation for subsequent practical operations.

[0154] The historical data interface systematically presents the sensor's number, type, and numerical records throughout the complete test cycle in a tabular format. Operators can quickly access the corresponding sensor's data details page by clicking on the sensor number in the table, and review the data change trend of that sensor throughout the entire test.

[0155] According to one embodiment of the present invention, the end-to-end response time of a single-frame dynamic visualization cloud map is no more than 2 seconds.

[0156] Corresponding to the above embodiments, the present invention also proposes an implementation method for a safety status visualization system for a large vertical circumferential reaction device.

[0157] The implementation method of the safety status visualization system for a large vertical circumferential reaction device according to an embodiment of the present invention, using the safety status visualization system for a large vertical circumferential reaction device as described above, includes the following steps:

[0158] Step 1: Deploy sensor groups in the key stress areas of the reaction device and establish a real-time data channel with the data-driven layer through an industrial gateway.

[0159] Step 2: Launch the visualization analysis platform, load the pre-trained surrogate model weights, and establish a connection with the relational database;

[0160] Step 3: During the experiment, the actuator operating condition signals are captured in real time, and the agent model is driven to perform forward inference calculations to obtain the mechanical response field data of all nodes of the device.

[0161] Step four involves mapping the surrogate model's prediction results to the 3D topological mesh model in real time, simultaneously generating a dynamic cloud map in the visualization window, and executing real-time warnings based on preset safety thresholds (such as material yield strength or displacement limit values). When the predicted value exceeds the threshold, a risk warning is issued in the visualization interface through color flashing and alarm pop-ups.

[0162] It should be noted that for details not disclosed in the safety status visualization system for the large vertical circumferential reaction device in this embodiment of the invention, please refer to the details disclosed in the implementation method of the safety status visualization system for the large vertical circumferential reaction device in this embodiment of the invention, which will not be repeated here.

[0163] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0164] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0165] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0166] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A safety status visualization system for a large-scale vertical circumferential reaction device, characterized in that, It includes the physical layer, the perception layer, the twin model layer, the twin data layer, and the visualization and inference layer; The physical layer is the physical entity of the reaction device, which serves as the data acquisition object of the sensing layer. The sensing layer includes a sensor group deployed on the physical layer, used to collect multi-dimensional working condition data and local feature data of the reaction device in real time during operation, and send the collected signals to the twin data layer; The twin data layer includes a data storage and interaction module, which is used to establish data interaction between the perception layer, the twin model layer and the visualization and deduction layer, and to store the physical parameters and dynamic operation data of the reaction device; The twin model layer is connected to the twin data layer to construct a digital twin of the reaction device and to invert the stress and displacement indices of the reaction device based on the retrieved multi-dimensional working condition data. The visualization inference layer is connected to the twin data layer, and the three-dimensional topological mesh model of the twin model layer is called based on the twin data layer to transform the global node prediction data obtained by inversion into a dynamic visualization cloud map through a spatial mapping algorithm.

2. The safety status visualization system for a large vertical circumferential reaction device according to claim 1, characterized in that, The twin model layer integrates a geometric model, a physical model, a behavioral model, and a rule model; The geometric model describes the topology of the reaction device, the physical model assigns dynamic properties to the model, the behavioral model defines the action response logic, and the rule model embeds a mechanical response proxy model, which is used to invert the stress and displacement indices of the reaction device based on the multi-dimensional working condition data.

3. The safety status visualization system for a large vertical circumferential reaction device according to claim 2, characterized in that, Constructing the twin model layer includes the following steps: S1, Construct the geometric model: Extract the geometric features of the reaction device using 3D modeling software, perform lightweight processing using a mesh reduction algorithm, and import it into a graphics rendering engine to perform geometric repair. S2, Construct the physical model: Configure rigid body components, collision body components and physical material properties for the geometric model in the graphics rendering engine to reproduce the mass distribution, collision boundary and friction characteristics of the reaction device; S3, Construct the behavior model: Define the action response logic of the twin model, and describe the spatial pose evolution of the reaction device during operation through matrix transformation; S4, Construct the rule model: embed a mechanical response proxy model based on a multilayer perceptron network to perform a nonlinear mapping from the working condition input vector to the global node prediction vector.

4. The safety status visualization system for a large vertical circumferential reaction device according to claim 3, characterized in that, Constructing the mechanical response proxy model of the twin model layer includes the following steps: S41. Based on the geometric characteristics, material constitutive model and contact relationship of the reaction device, a refined finite element model is established, and mesh sensitivity analysis and experimental data comparison verification are performed. S42 uses the applied load and loading angle as input variables and employs a space-filling sampling algorithm to extract simulation working points within a preset factor level range. S43 utilizes automated calculation scripts to drive the finite element model to perform batch calculations on simulation conditions, extracts stress and displacement response data of all nodes of the reaction device, and constructs a structural mechanical performance characteristic dataset. S44, a multilayer perceptron network is trained using the aforementioned structural mechanical performance feature dataset, and random Gaussian noise is introduced at the input to perform robust training. The trained mechanical response surrogate model is then calibrated using physical experimental data to ensure that the global prediction error range is ≤5%. S45 deploys the trained mechanical response proxy model in the rule model of the twin model layer, using the real-time collected load and loading angle as input vectors, and realizes the inversion output of the stress prediction vector and displacement prediction vector of the global nodes of the reaction device through the online inversion calculation logic of the proxy model.

5. The safety status visualization system for a large vertical circumferential reaction device according to claim 4, characterized in that, The online inversion calculation logic executed by the mechanical response proxy model of the twin model layer is as follows: The input sequence of the current applied load and loading angle transmitted from the twin data layer is fed into the multilayer perceptron network. The feature transfer and state update formulas for each hidden layer node are as follows: ； In the formula, For the first The output tensor of the layer, This is the input sequence for the previous layer. This represents the mapping weight matrix that is fixed during the training phase. For physical bias terms, It is a nonlinear activation function; the output layer decouples the stress prediction vector and displacement prediction vector of each node in the entire domain.

6. The safety status visualization system for a large vertical circumferential reaction device according to claim 5, characterized in that, The transmission architecture of the twin data layer includes a data acquisition layer, a data-driven layer, a data storage layer, and a data application layer; among which... The data acquisition layer acquires real-time data on the operation of the reaction device through the perception layer; the data driving layer calls the proxy model through standardized software interfaces and data interaction protocols; the data storage layer uses a relational database as the underlying storage management center to uniformly manage the measured data and the proxy model's predicted data; the data application layer constructs a cross-platform interface between the database, the 3D graphics rendering engine, and the programming environment to synchronously drive the proxy model's forward inference and the dynamic refresh of the visualization cloud map.

7. The safety status visualization system for a large vertical circumferential reaction device according to claim 6, characterized in that, The rendering logic of the dynamic visualization cloud map includes: Extract the element topology information and node coordinates of the finite element model, divide the non-triangular elements into triangular patches, and reconstruct to generate a three-dimensional topological mesh model; The received predicted values ​​of the reaction device nodes are normalized, and a linear gradient mapping relationship between the predicted values ​​and the color space is established. Color rendering of pixels within an image is performed using a barycentric coordinate interpolation algorithm. The interpolation calculation model is defined as follows: ； In the formula, For any point within the sheet Color value calculation model, , and The color mapping value of the face vertex. , and For point The weights of the centroid coordinates within the patch, and satisfying... .

8. The safety status visualization system for a large vertical circumferential reaction device according to claim 1, characterized in that, The visualization simulation layer is configured with a visualization window integrated into the digital monitoring terminal, and the components integrated within the visualization window include: The functional configuration unit includes a data acquisition module, an operation status monitoring module, a fault early warning module, and an information management module; wherein, the operation status monitoring module is used to drive the twin model layer to map and display the mechanical performance of the reaction device; the fault early warning module is used to trigger alarm information based on a preset safety threshold; The dynamic rendering and display unit integrates dynamic curve drawing components and cloud map rendering engine. It is used to load modular interactive interfaces according to the test process and render and display the predicted data of the entire domain nodes of the reaction device and the sensor operating status in real time. The human-computer interaction control unit is equipped with an event listening mechanism to capture physical operation commands triggered by interface interaction elements and convert them into system logic trigger signals to realize remote control mapping of the loading angle and distance adjustment of the reaction force device. The status visualization feedback unit is used to dynamically update the status indication information of key nodes through a visualization window based on logical judgment of working condition signals and posture sensing data, so as to realize real-time progress prompts during the test phase.

9. The safety status visualization system for a large vertical circumferential reaction device according to claim 1, characterized in that, The end-to-end response time for a single-frame dynamic visualization cloud map is no more than 2 seconds.

10. A method for implementing a safety status visualization system for a large-scale vertical circumferential reaction device, characterized in that, The safety status visualization system for a large vertical circumferential reaction device as described in any one of claims 1-9 includes the following steps: Step 1: Deploy sensor groups in the key stress areas of the reaction device and establish a real-time data channel with the data-driven layer through an industrial gateway. Step 2: Launch the visualization analysis platform, load the pre-trained surrogate model weights, and establish a connection with the relational database; Step 3: During the experiment, the actuator operating condition signal is captured in real time, and the proxy model is driven to perform forward inference calculation to obtain the mechanical response field data of the entire node of the device. Step 4: Map the prediction results of the proxy model to the 3D topological mesh model in real time, generate a dynamic cloud map in the visualization window, and execute real-time warnings based on the preset safety threshold.

Images