A digital twin modeling method for chemical pipeline networks
By performing three-dimensional reconstruction and fluid dynamic analysis of multi-view data of chemical pipeline network, dynamic operation model is constructed and collision detection is carried out, the problem of insufficient visual effect and immersion of virtual models in the existing technology is solved, and virtual models and reality are realized Improved consistency and immersion reduce hardware costs.
Patent Information
- Application Number
- CN202410797478.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-20
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2044-06-20
AI Technical Summary
The existing digital twin modeling method of chemical pipeline network only focuses on simulated operation status, ignores the visual effects and immersive experience of virtual models, and cannot effectively reflect the physical laws of the real world, resulting in decision-making lag and inaccurate.
By collecting chemical pipeline data from multiple perspectives, performing three-dimensional reconstruction and fitting, combining fluid dynamic analysis to build a dynamic operation model, performing collision detection and scene switching optimization, calculating virtual interaction accuracy and application quality index, and enhancing the consistency of virtual models Sex and immersion.
Achieve high consistency between virtual models and the real world, prevent physical errors, enhance immersive experience, and support model optimization through data evaluation, reducing hardware costs.
Smart Images

Figure CN118445960B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of digital modeling, and more specifically, to a digital twin modeling method for a chemical pipeline network. Background Art
[0002] In the existing chemical pipeline management, traditional physical models and manual monitoring methods cannot reflect the actual operating status of the pipeline in real time and accurately, resulting in delayed and inaccurate decision-making. Digital modeling and intelligent monitoring have gradually become hot topics.
[0003] The existing digital twin modeling method of chemical pipeline networks collects real-time data of chemical pipeline networks, including flow, pressure and temperature, through sensors and monitoring equipment, and then cleans, integrates and analyzes the collected data. Based on the processed data, a digital twin model is established using machine learning and artificial intelligence algorithms. This model can simulate the actual operating status of the chemical pipeline network, which improves operating efficiency and reduces operating risks to a certain extent.
[0004] However, the above system still has some problems: it only simulates the operating status of the chemical pipeline network and maintains the dynamic data consistency between the real world and the virtual world, but ignores the user's visual effects and immersive experience of the virtual digital twin model, which needs to be further strengthened. Summary of the invention
[0005] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a digital twin modeling method for a chemical pipeline network to solve the problems raised in the above-mentioned background technology.
[0006] To achieve the above object, the present invention provides the following technical solution: a digital twin modeling method for a chemical pipeline network, comprising the following steps:
[0007] S1: Collect pictures of each pipeline, valve, pump, and component equipment in the chemical pipeline network from multiple perspectives, physical parameters, connection methods of each component in the pipeline network, and historical operation data;
[0008] S2: annotate the collected images as original image data sets, perform 3D reconstruction and fitting on the original image data sets, and output the final 3D reconstruction model of the chemical pipeline network;
[0009] S3: Compare and check the three-dimensional reconstructed model of the chemical pipeline network with the standard model, and mark the three-dimensional reconstructed model that passes the test as the static environment model of the chemical pipeline network;
[0010] S4. Analyze the fluid dynamics in the chemical pipeline network, build a dynamic operation model of the chemical pipeline network based on the collected data other than the pictures, and output a verified dynamic operation model of the chemical pipeline network;
[0011] S5. The static environment model of the chemical pipeline network and the dynamic operation model of the chemical pipeline network are combined to obtain a virtual digital twin model, collision detection is performed on the three-dimensional model of the digital twin, and the virtual interaction accuracy coefficient is calculated based on the matching degree between the virtual collision detection result and the actual collision result;
[0012] S6. Solve the data loading problem of different virtual scenes through mouse click events. After determining the virtual scene, synchronously load the new scene and render it. After switching to the new scene, unload the use of the previous scene;
[0013] S7: Count the duration required for each scene switch within the preset model operation cycle, the number of digital twin 3D models contained in the switched scene, the number of scene switches, and the number of correct scene switches;
[0014] S8: Calculate the average scene switching efficiency coefficient based on the duration required for each scene switch within the preset model operation cycle and the number of digital twin three-dimensional models contained in the switched scene; calculate the scene switching accuracy based on the number of correct scene switches and the number of incorrect scene switches; calculate the twin model application quality index based on the virtual interaction accuracy coefficient, the average scene switching efficiency coefficient and the scene switching accuracy.
[0015] Preferably, the physical parameters of the pipeline in step S1 include diameter, length, wall thickness, thermal conductivity, and roughness; the physical parameters of the valve include nominal diameter, nominal pressure, applicable medium, test pressure, and pressure and temperature levels; the physical parameters of the pump include static head, device head, operating point, pump efficiency, shaft power, motor efficiency, transmission system efficiency, flow channel efficiency, and pump station efficiency; the physical parameters of the component equipment include working pressure, working temperature, material, and dimensional information; the historical operation data includes the inflow and outflow mass flow, flow velocity, pressure, temperature, and density of the flowing medium at each node of the chemical pipeline network.
[0016] Preferably, the specific steps of generating the three-dimensional reconstruction model of the chemical pipeline network in step S2 are as follows:
[0017] S21. Design a feature extraction network structure, use a decoupled dynamic filter to decouple the depth dynamic filter into spatial and channel dynamic filters, perform convolution operations, and then construct a sparse cost map. The cost map is segmented in the depth direction. The cost volume is regularized using a sequence-sorted GUR network model to obtain a high-resolution sparse depth map. The sparse depth map and the corresponding high-resolution reference view are input as a matrix concatenated together, and different weights are obtained through training to achieve the effect from low resolution to high resolution. This process is recorded as a difference operation. The model automatically learns the weights and rules of interpolation to achieve reconstruction from sparse depth maps to dense depth maps. The dense depth map is refined using the Levenberg-Marquardt method, and the mean absolute error between the estimated depth map and the true depth map is used as the training loss function.
[0018] S22, obtaining a depth map of the three-dimensional reconstruction of the original image data set through the deep learning network training in step S21, and filtering the depth map using the photometric consistency and geometric consistency methods;
[0019] S23. The filtered depth map can be projected from the pixel coordinate system to the world coordinate system and converted into a coordinate point cloud, thereby realizing three-dimensional reconstruction. The reconstructed three-dimensional model is fitted to obtain a three-dimensional reconstructed model of the chemical pipeline network.
[0020] Preferably, the standard model in step S3 is a real three-dimensional coordinate point model of the chemical pipeline network obtained by using model making software to make a corresponding three-dimensional model based on the physical parameters of each pipeline, valve, pump, and component equipment in the chemical pipeline network and connecting them according to the connection method of each component in the collected pipeline network.
[0021] Preferably, the specific steps of comparing the three-dimensional reconstructed model of the chemical pipeline network with the standard model in step S3 are as follows:
[0022] S31, introducing an angle threshold for the chemical pipeline network standard model and the three-dimensional reconstruction model to extract edge contour points of the three-dimensional model, and using the contour points extracted from the standard model and the three-dimensional reconstruction model as their respective feature points;
[0023] S32. Count the number of feature points m extracted from the standard model te and the number of feature points m extracted from the 3D reconstruction model tc Calculate the number of feature points to extract the matching degree ζ t , the specific formula is: ;
[0024] S33, counting the number of feature points with consistent three-dimensional coordinates m ty And the number of remaining feature points m of the standard model es , the standard model feature points with inconsistent 3D coordinates and the 3D reconstruction model feature points are represented by (xei ,y ei ,z ei )、(x cj ,y cj ,z cj ) represents the calculation of the average value D of the minimum distance between the inconsistent standard model feature points and the 3D reconstruction model feature points ye , the specific formula is: The feature point position matching degree ζ is calculated based on the average value of the minimum distance between the inconsistent standard model feature points and the 3D reconstruction model feature points and the number of feature points with consistent 3D coordinates. w , the specific formula is: , where D yt The minimum distance allowed is preset;
[0025] S34, calculating the overall matching degree ζ of the feature points based on the matching degree of the number of feature points extracted and the matching degree of the feature point positions z , the specific formula is: , e is a natural constant;
[0026] S35. Compare the calculated overall matching degree of the feature points with the preset lower limit of the overall matching degree of the feature points. If the calculated value is greater than or equal to the preset lower limit, the three-dimensional reconstructed model is judged to have passed the inspection. Otherwise, the three-dimensional reconstructed model is judged to have failed the inspection. The three-dimensional reconstructed model that has passed the inspection is marked as a static environment model of the chemical pipeline network. When the composition of the chemical pipeline network changes, the static environment model of the chemical pipeline network will be automatically updated.
[0027] Preferably, the specific steps of model construction in step S4 are as follows:
[0028] S41, setting each node in the pipe network based on the collected physical parameters of each pipe, valve, pump, and component equipment in the chemical pipe network and the connection method of each component in the pipe network;
[0029] S42, performing mass conservation analysis, momentum conservation analysis and energy conservation analysis on each node in the pipe network and establishing corresponding mathematical expressions;
[0030] S43. Based on the characteristics of the dataset used and the model building requirements, the network architecture is selected and the network structure and number of layers are designed. After the model used is determined, the relevant packages are imported and important global parameters are modified. At the same time, the above mathematical expressions are used as constraints to build a dynamic operation model of the chemical pipeline network;
[0031] S44, divide the historical operation data into training set and test set to train and test the constructed model, and record the number of normal operation data in the test sample n yz , the number of abnormal running data n yf , the number of normal and abnormal operation data correctly identified nsz ;
[0032] S45. Calculate the running recognition accuracy λ based on the test results s , the specific formula is: ;
[0033] S46. Compare the calculated operation recognition accuracy with the preset standard operation recognition accuracy. If the calculated value is greater than or equal to the preset value, the model passes the test; otherwise, the model fails the test and needs to be optimized until it passes the test.
[0034] Preferably, the virtual digital twin model generation and collision detection process in step S5 is as follows:
[0035] S51, cutting the static environment model and the dynamic operation model of the chemical pipeline network according to the scene, integrating the static environment model and the dynamic operation model of each scene into the same scene, and finally obtaining a virtual digital twin model of the chemical pipeline network;
[0036] S52, use the bounding box of the digital twin 3D model to detect whether objects in the scene are in contact. The detection algorithm principle is as follows: any two digital twin 3D models A and B, respectively, are bounded by L A , W A , H A and L B , W B , H B The bounding box size represented by , the coordinates of the lower left corner vertex of the bounding box of the 3D models A and B are (X A ,Y A ,Z A) 、(X B ,Y B ,Z B ), then the coordinates of the upper right corner vertices of the bounding boxes of the three-dimensional models A and B of the digital twin are (X A +L A ,Y A +W A ,Z A +H A )、(X B +L B ,Y B +W B ,Z B +H B ),when When A and B are in contact, D X , D Y , D Z They represent the components of the center distance between objects A and B in the (X, Y, Z) direction of the coordinate axis. , , , the center coordinates of the two objects A and B are (A X ,A Y ,A Z ) and (B X ,B Y ,B Z );
[0037] S53, recording the collision detection results of any two objects in the digital twin model and the collision results of the corresponding two physical objects in the real world, counting the number of detection results with consistent detection results and the number of detection results with inconsistent detection results, and adjusting the twin model if the detection results are inconsistent;
[0038] S54, based on the number of consistent detection results m cy and inconsistent test results m cx Calculate the virtual interaction accuracy coefficient η z The specific formula is: .
[0039] Preferably, the specific data processing process in step S8 is as follows:
[0040] S81, based on the time t required for the i-th scene switching in the preset model operation cycle qi , the number of digital twin 3D models contained in the switched scene m qi And the number of scene switching times n within the preset model operation cycle q Calculate the average scene switching efficiency coefficient η xe The specific formula is: ;
[0041] S82: The number of correct scene switching times n within the preset model operation cycle qz and the number of scene switching n q Calculate the scene switching accuracy λ q The specific formula is: ;
[0042] S83. Calculate the twin model application quality index R based on the virtual interaction accuracy coefficient, the average scene switching efficiency coefficient, and the scene switching accuracy Y The specific formula is: , e is a natural constant.
[0043] Technical effects and advantages of the present invention:
[0044] The present invention collects pictures, physical parameters, connection methods of components in the chemical pipeline network from multiple perspectives of each pipeline, valve, pump, and component equipment, and historical operation data, annotates the collected pictures as original image data sets, performs three-dimensional reconstruction and fitting on the original image data sets, outputs the final three-dimensional reconstruction model of the chemical pipeline network, compares and verifies the three-dimensional reconstruction model of the chemical pipeline network with the standard model, and marks the three-dimensional reconstruction model that passes the test as the static environment model of the chemical pipeline network, which to a certain extent ensures the high consistency between the virtual digital twin model and the real objects in the real world; the present invention analyzes the fluid dynamics in the chemical pipeline network, constructs a dynamic operation model of the chemical pipeline network in combination with the remaining data collected except the pictures, outputs the verified dynamic operation model of the chemical pipeline network, and compares the static environment model of the chemical pipeline network with the dynamic environment model of the chemical pipeline network. The virtual digital twin model can be obtained by merging the two dynamically running models. Collision detection is performed on the three-dimensional model of the data twin, and the virtual interaction accuracy coefficient is calculated based on the degree of match between the virtual collision detection results and the actual collision results. The physical laws of the real world are more accurately reproduced to prevent abnormal physical behaviors such as objects passing through each other, which enhances the immersive experience to a certain extent. The data loading problem of different virtual scenes is solved through mouse click events. After determining the virtual scene, the new scene is loaded and rendered synchronously. After switching to the new scene, the use of the previous scene is uninstalled. While maintaining the dynamic update and switching of the new scene, it can also effectively reduce the hardware cost. A twin model application quality assessment method is provided, which shows the application effect of the twin model through intuitive data, and provides data support for the subsequent adjustment and optimization of the model. BRIEF DESCRIPTION OF THE DRAWINGS
[0045] Figure 1 It is a diagram of the method steps of the present invention.
[0046] Figure 2 It is a system structure block diagram of the present invention. DETAILED DESCRIPTION
[0047] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0048] like Figure 1 The present embodiment provides a digital twin modeling method for a chemical pipeline network, comprising the following steps:
[0049] S1: Collect pictures of each pipeline, valve, pump, and component equipment in the chemical pipeline network from multiple perspectives, physical parameters, connection methods of each component in the pipeline network, and historical operation data;
[0050] Furthermore, the physical parameters of the pipeline in step S1 include diameter, length, wall thickness, thermal conductivity, and roughness; the physical parameters of the valve include nominal diameter, nominal pressure, applicable medium, test pressure, and pressure and temperature levels; the physical parameters of the pump include static head, device head, operating point, pump efficiency, shaft power, motor efficiency, transmission system efficiency, flow channel efficiency, and pump station efficiency; the physical parameters of the component equipment include working pressure, working temperature, material, and dimensional information; the historical operation data includes the inflow and outflow mass flow, flow velocity, pressure, temperature, and density of the flowing medium at each node of the chemical pipeline network.
[0051] S2: annotate the collected images as original image data sets, perform 3D reconstruction and fitting on the original image data sets, and output the final 3D reconstruction model of the chemical pipeline network;
[0052] Furthermore, the specific steps of generating the three-dimensional reconstruction model of the chemical pipeline network in step S2 are as follows:
[0053] S21. Design a feature extraction network structure, use a decoupled dynamic filter to decouple the depth dynamic filter into spatial and channel dynamic filters, perform convolution operations, and then construct a sparse cost map. The cost map is segmented in the depth direction. The cost volume is regularized using a sequence-sorted GUR network model to obtain a high-resolution sparse depth map. The sparse depth map and the corresponding high-resolution reference view are input as a matrix concatenated together, and different weights are obtained through training to achieve the effect from low resolution to high resolution. This process is recorded as a difference operation. The model automatically learns the weights and rules of interpolation to achieve reconstruction from sparse depth maps to dense depth maps. The dense depth map is refined using the Levenberg-Marquardt method, and the mean absolute error between the estimated depth map and the true depth map is used as the training loss function.
[0054] S22, obtaining a depth map of the three-dimensional reconstruction of the original image data set through the deep learning network training in step S21, and filtering the depth map using the photometric consistency and geometric consistency methods;
[0055] S23. The filtered depth map can be projected from the pixel coordinate system to the world coordinate system and converted into a coordinate point cloud, thereby realizing three-dimensional reconstruction. The reconstructed three-dimensional model is fitted to obtain a three-dimensional reconstructed model of the chemical pipeline network.
[0056] It should be specifically noted in this embodiment that all steps in the three-dimensional reconstruction process of the chemical pipeline network belong to the existing technology and will not be explained in detail here.
[0057] S3: Compare and check the three-dimensional reconstructed model of the chemical pipeline network with the standard model, and mark the three-dimensional reconstructed model that passes the test as the static environment model of the chemical pipeline network;
[0058] Furthermore, the standard model in step S3 is a real three-dimensional coordinate point model of the chemical pipeline network obtained by using model making software to make a corresponding three-dimensional model based on the physical parameters of each pipeline, valve, pump, and component equipment in the chemical pipeline network and connecting them according to the connection method of each component in the collected pipeline network.
[0059] Furthermore, the specific steps of comparing the three-dimensional reconstructed model of the chemical pipe network with the standard model in step S3 are as follows:
[0060] S31, introducing an angle threshold for the chemical pipeline network standard model and the three-dimensional reconstruction model to extract edge contour points of the three-dimensional model, and using the contour points extracted from the standard model and the three-dimensional reconstruction model as their respective feature points;
[0061] Specifically, in this embodiment, a feature point extraction method is provided, including the following steps:
[0062] S311, the target point p b Instead of projecting the points within the search radius onto the least squares plane it constructs, let the number of points within the search radius be n c , the nearest point in the neighborhood of the target point is recorded as p ti , with p b p ti As the reference vector, take any other point p in the neighborhood tj , let vector p b p tj ;
[0063] S312, calculate p b p ti With p b p tj The angle α between the vectors, α∈[0,π] and the cross product vector , and use this as the reference vector, when When α i Remain unchanged when hour, ;
[0064] S313, vector angle sequence a={α1,α2,......,α nc} are arranged in ascending order and two extreme angles are added to obtain a new vector angle sequence a={0,α1,α2,......,α nc ,2π}, then the angle θ between each adjacent line segmenti The calculation formula is: ;
[0065] S314, according to θ i Whether the target point is a boundary contour point is determined by whether it exceeds the set threshold, and the determined contour point is marked as a feature point.
[0066] S32. Count the number of feature points m extracted from the standard model te and the number of feature points m extracted from the 3D reconstruction model tc Calculate the number of feature points to extract the matching degree ζ t , the specific formula is: ;
[0067] S33, counting the number of feature points with consistent three-dimensional coordinates m ty And the number of remaining feature points m of the standard model es , the standard model feature points with inconsistent 3D coordinates and the 3D reconstruction model feature points are represented by (x ei ,y ei ,z ei )、(x cj ,y cj ,z cj ) represents the calculation of the average value D of the minimum distance between the inconsistent standard model feature points and the 3D reconstruction model feature points ye , the specific formula is: The feature point position matching degree ζ is calculated based on the average value of the minimum distance between the inconsistent standard model feature points and the 3D reconstruction model feature points and the number of feature points with consistent 3D coordinates. w , the specific formula is: , where D yt The minimum distance allowed is preset;
[0068] S34, calculating the overall matching degree ζ of the feature points based on the matching degree of the number of feature points extracted and the matching degree of the feature point positions z , the specific formula is: , e is a natural constant;
[0069] S35. Compare the calculated overall matching degree of the feature points with the preset lower limit of the overall matching degree of the feature points. If the calculated value is greater than or equal to the preset lower limit, the three-dimensional reconstructed model is judged to have passed the inspection. Otherwise, the three-dimensional reconstructed model is judged to have failed the inspection. The three-dimensional reconstructed model that has passed the inspection is marked as a static environment model of the chemical pipeline network. When the composition of the chemical pipeline network changes, the static environment model of the chemical pipeline network will be automatically updated.
[0070] S4. Analyze the fluid dynamics in the chemical pipeline network, build a dynamic operation model of the chemical pipeline network based on the collected data other than the pictures, and output a verified dynamic operation model of the chemical pipeline network;
[0071] Furthermore, the specific steps of model construction in step S4 are as follows:
[0072] S41, setting each node in the pipe network based on the collected physical parameters of each pipe, valve, pump, and component equipment in the chemical pipe network and the connection method of each component in the pipe network;
[0073] S42, performing mass conservation analysis, momentum conservation analysis and energy conservation analysis on each node in the pipe network and establishing corresponding mathematical expressions;
[0074] It should be specifically noted in this embodiment that fluid dynamics analysis and establishment of mathematical expressions are both existing technologies and will not be described in detail here.
[0075] S43. Based on the characteristics of the dataset used and the model building requirements, the network architecture is selected and the network structure and number of layers are designed. After the model used is determined, the relevant packages are imported and important global parameters are modified. At the same time, the above mathematical expressions are used as constraints to build a dynamic operation model of the chemical pipeline network;
[0076] S44, divide the historical operation data into training set and test set to train and test the constructed model, and record the number of normal operation data in the test sample n yz , the number of abnormal running data n yf , the number of normal and abnormal operation data correctly identified n sz ;
[0077] S45. Calculate the running recognition accuracy λ based on the test results s , the specific formula is: ;
[0078] S46. Compare the calculated operation recognition accuracy with the preset standard operation recognition accuracy. If the calculated value is greater than or equal to the preset value, the model passes the test; otherwise, the model fails the test and needs to be optimized until it passes the test.
[0079] It should be specifically noted in this embodiment that the preset values used are selected based on actual needs and are not limited to specific values.
[0080] S5. The static environment model of the chemical pipeline network and the dynamic operation model of the chemical pipeline network are combined to obtain a virtual digital twin model, collision detection is performed on the three-dimensional model of the digital twin, and the virtual interaction accuracy coefficient is calculated based on the matching degree between the virtual collision detection result and the actual collision result;
[0081] Furthermore, the virtual digital twin model generation and collision detection process in step S5 is as follows:
[0082] S51, cutting the static environment model and the dynamic operation model of the chemical pipeline network according to the scene, integrating the static environment model and the dynamic operation model of each scene into the same scene, and finally obtaining a virtual digital twin model of the chemical pipeline network;
[0083] It should be specifically explained in this embodiment that there are two types of models in the virtual digital twin model, static environment models and dynamic motion models. The static environment models are intended to provide visual effects to enhance the immersion of the digital twin system. They are mainly used to create details and background elements in the virtual environment, while the dynamic motion models are intended to maintain the dynamic consistency between the physical space and the virtual digital twin model.
[0084] S52, use the bounding box of the digital twin 3D model to detect whether objects in the scene are in contact. The detection algorithm principle is as follows: any two digital twin 3D models A and B, respectively, are bounded by L A , W A , H A and L B , W B , H B The bounding box size represented by , the coordinates of the lower left corner vertex of the bounding box of the 3D models A and B are (X A ,Y A ,Z A) 、(X B ,Y B ,Z B ), then the coordinates of the upper right corner vertices of the bounding boxes of the three-dimensional models A and B of the digital twin are (X A +L A ,Y A +W A ,Z A +H A )、(X B +L B ,Y B +W B ,Z B +H B ),when When A and B are in contact, D X , D Y , D Z They represent the components of the center distance between objects A and B in the (X, Y, Z) direction of the coordinate axis. , , , the center coordinates of the two objects A and B are (A X ,A Y ,AZ ) and (B X ,B Y ,B Z );
[0085] S53, recording the collision detection results of any two objects in the digital twin model and the collision results of the corresponding two physical objects in the real world, counting the number of detection results with consistent detection results and the number of detection results with inconsistent detection results, and adjusting the twin model if the detection results are inconsistent;
[0086] S54, based on the number of consistent detection results m cy and inconsistent test results m cx Calculate the virtual interaction accuracy coefficient η z The specific formula is: .
[0087] S6. Solve the data loading problem of different virtual scenes through mouse click events. After determining the virtual scene, synchronously load the new scene and render it. After switching to the new scene, unload the use of the previous scene;
[0088] S7: Count the duration required for each scene switch within the preset model operation cycle, the number of digital twin 3D models contained in the switched scene, the number of scene switches, and the number of correct scene switches;
[0089] S8: Calculate the average scene switching efficiency coefficient based on the duration required for each scene switch within the preset model operation cycle and the number of digital twin three-dimensional models contained in the switched scene; calculate the scene switching accuracy based on the number of correct scene switches and the number of incorrect scene switches; calculate the twin model application quality index based on the virtual interaction accuracy coefficient, the average scene switching efficiency coefficient and the scene switching accuracy.
[0090] Further, the specific data processing process in step S8 is as follows:
[0091] S81, based on the time t required for the i-th scene switching in the preset model operation cycle qi , the number of digital twin 3D models contained in the switched scene m qi And the number of scene switching times n within the preset model operation cycle q Calculate the average scene switching efficiency coefficient η xe The specific formula is: ;
[0092] S82: The number of correct scene switching times n within the preset model operation cycle qz and the number of scene switching n q Calculate the scene switching accuracy λ q The specific formula is: ;
[0093] S83. Calculate the twin model application quality index R based on the virtual interaction accuracy coefficient, the average scene switching efficiency coefficient, and the scene switching accuracy Y The specific formula is: , e is a natural constant.
[0094] like Figure 2 The embodiment shown provides a digital twin modeling system for a chemical pipeline network, including a chemical pipeline network data acquisition module, a chemical pipeline network three-dimensional reconstruction module, a three-dimensional reconstruction model verification module, a chemical pipeline network dynamic operation model construction module, a model fusion module, a digital twin model collision detection module, a scene switching rule setting module, a scene switching recording module, a model application quality assessment module, and a database. The chemical pipeline network data acquisition module is connected to the chemical pipeline network three-dimensional reconstruction module and the chemical pipeline network dynamic operation model construction module, the chemical pipeline network three-dimensional reconstruction module is connected to the three-dimensional reconstruction model verification module, the three-dimensional reconstruction model verification module and the chemical pipeline network dynamic operation model construction module are connected to the model fusion module, the model fusion module, the digital twin model collision detection module, the scene switching rule setting module, the scene switching recording module, and the model application quality assessment module are connected in sequence, and all modules in the system are connected to the database.
[0095] Chemical pipeline network data acquisition module: collects pictures of each pipeline, valve, pump, and component equipment in the chemical pipeline network from multiple perspectives, physical parameters, connection methods of each component in the pipeline network, and historical operation data;
[0096] Chemical pipe network 3D reconstruction module: annotate the collected images as original image data sets, perform 3D reconstruction and fitting on the original image data sets, and output the final 3D reconstruction model of the chemical pipe network;
[0097] 3D reconstruction model verification module: compare and verify the 3D reconstruction model of the chemical pipeline network with the standard model, and mark the 3D reconstruction model that passes the verification as the static environment model of the chemical pipeline network;
[0098] Chemical pipeline network dynamic operation model construction module: analyze the fluid dynamics in the chemical pipeline network, combine the collected data except the pictures to build a chemical pipeline network dynamic operation model, and output the verified chemical pipeline network dynamic operation model;
[0099] Model fusion module: The static environment model of the chemical pipeline network and the dynamic operation model of the chemical pipeline network are combined to obtain a virtual digital twin model;
[0100] Digital twin model collision detection module: performs collision detection on the three-dimensional model of the digital twin, and calculates the virtual interaction accuracy coefficient based on the matching degree between the virtual collision detection results and the actual collision results;
[0101] Scene switching rule setting module: solves the data loading problem of different virtual scenes through mouse click events. After determining the virtual scene, it synchronously loads the new scene and renders it. After switching to the new scene, the use of the previous scene is unloaded;
[0102] Scene switching record module: counts the duration of each scene switching within the preset model operation cycle, the number of digital twin 3D models contained in the switched scene, the number of scene switching times, and the number of correct scene switching times;
[0103] Model application quality assessment module: Calculate the average scene switching efficiency coefficient based on the duration of each scene switch within the preset model operation cycle and the number of digital twin 3D models contained in the switched scene; calculate the scene switching accuracy based on the number of correct scene switches and the number of incorrect scene switches; calculate the twin model application quality index based on the virtual interaction accuracy coefficient, the average scene switching efficiency coefficient and the scene switching accuracy;
[0104] Database: used to store data information of all modules in the system.
[0105] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A digital twin modeling method for a chemical pipeline network, characterized in that: The following steps are involved: S1: Collect pictures of each pipeline, valve, pump, and component equipment in the chemical pipeline network from multiple perspectives, physical parameters, connection methods of each component in the pipeline network, and historical operation data; S2: annotate the collected images as original image data sets, perform 3D reconstruction and fitting on the original image data sets, and output the final 3D reconstruction model of the chemical pipeline network; S3: Compare and check the three-dimensional reconstructed model of the chemical pipeline network with the standard model, and mark the three-dimensional reconstructed model that passes the test as the static environment model of the chemical pipeline network; S4. Analyze the fluid dynamics in the chemical pipeline network, build a dynamic operation model of the chemical pipeline network based on the collected data other than the pictures, and output a verified dynamic operation model of the chemical pipeline network; S5. The static environment model of the chemical pipeline network and the dynamic operation model of the chemical pipeline network are combined to obtain a virtual digital twin model, collision detection is performed on the three-dimensional model of the digital twin, and the virtual interaction accuracy coefficient is calculated based on the matching degree between the virtual collision detection result and the actual collision result; The virtual digital twin model generation and collision detection process in step S5 is as follows: S51, cutting the static environment model and the dynamic operation model of the chemical pipeline network according to the scene, integrating the static environment model and the dynamic operation model of each scene into the same scene, and finally obtaining a virtual digital twin model of the chemical pipeline network; S52, use the bounding box of the digital twin 3D model to detect whether objects in the scene are in contact. The detection algorithm principle is as follows: any two digital twin 3D models A and B, respectively, are bounded by L A , W A , H A and L B , W B , H B The bounding box size represented by , the lower left corner vertex coordinates of the bounding boxes of 3D models A and B are (X A ,Y A ,Z A )、(X B ,Y B ,Z B ), then the coordinates of the upper right corner vertices of the bounding boxes of the three-dimensional models A and B of the digital twin are (X A +L A ,Y A +W A ,Z A +H A )、(X B +L B ,Y B +W B ,Z B +H B ),when When A and B are in contact, D X , D Y , D Z They represent the components of the center distance between objects A and B in the (X, Y, Z) direction of the coordinate axis. , , , the center coordinates of the two objects A and B are (A X ,A Y ,A Z ) and (B X ,B Y ,B Z ); S53, recording the collision detection results of any two objects in the digital twin model and the collision results of the corresponding two physical objects in the real world, counting the number of detection results with consistent detection results and the number of detection results with inconsistent detection results, and adjusting the twin model if the detection results are inconsistent; S54, based on the number of consistent detection results m cy and inconsistent test results m cx Calculate the virtual interaction accuracy coefficient η z The specific formula is: ; S6. Solve the data loading problem of different virtual scenes through mouse click events. After determining the virtual scene, synchronously load the new scene and render it. After switching to the new scene, unload the use of the previous scene; S7: Count the duration required for each scene switch within the preset model operation cycle, the number of digital twin 3D models contained in the switched scene, the number of scene switches, and the number of correct scene switches; S8: Calculate the average scene switching efficiency coefficient based on the duration of each scene switching within the preset model operation cycle and the number of digital twin 3D models contained in the switched scene; calculate the scene switching accuracy based on the number of correct scene switching and the number of incorrect scene switching; calculate the twin model application quality index based on the virtual interaction accuracy coefficient, the average scene switching efficiency coefficient and the scene switching accuracy; The specific data processing process in step S8 is as follows: S81, based on the time t required for the i-th scene switching in the preset model operation cycle qi , the number of digital twin 3D models contained in the switched scene m qi And the number of scene switching times n within the preset model operation cycle q Calculate the average scene switching efficiency coefficient η xe The specific formula is: ; S82: The number of correct scene switching times n within the preset model operation cycle qz and the number of scene switching n q Calculate the scene switching accuracy λ q The specific formula is: ; S83. Calculate the twin model application quality index R based on the virtual interaction accuracy coefficient, the average scene switching efficiency coefficient, and the scene switching accuracy Y The specific formula is: , e is a natural constant.
2. A digital twin modeling method for a chemical pipeline network according to claim 1, characterized in that: The physical parameters of the pipeline in step S1 include diameter, length, wall thickness, thermal conductivity, and roughness; the physical parameters of the valve include nominal diameter, nominal pressure, applicable medium, test pressure, and pressure and temperature levels; the physical parameters of the pump include static head, device head, operating point, pump efficiency, shaft power, motor efficiency, transmission system efficiency, flow channel efficiency, and pump station efficiency; the physical parameters of the component equipment include working pressure, working temperature, material, and dimensional information; the historical operation data includes the inflow and outflow mass flow, flow velocity, pressure, temperature, and density of the flowing medium at each node of the chemical pipeline network.
3. The digital twin modeling method for a chemical pipeline network according to claim 1, characterized in that: The specific steps of generating the three-dimensional reconstruction model of the chemical pipeline network in step S2 are as follows: S21. Design a feature extraction network structure, use a decoupled dynamic filter to decouple the depth dynamic filter into spatial and channel dynamic filters, perform convolution operations, and then construct a sparse cost map. The cost map is segmented in the depth direction. The cost volume is regularized using a sequence-sorted GUR network model to obtain a high-resolution sparse depth map. The sparse depth map and the corresponding high-resolution reference view are input as a matrix concatenated together, and different weights are obtained through training to achieve the effect from low resolution to high resolution. This process is recorded as a difference operation. The model automatically learns the weights and rules of interpolation to achieve reconstruction from sparse depth maps to dense depth maps. The dense depth map is refined using the Levenberg-Marquardt method, and the mean absolute error between the estimated depth map and the true depth map is used as the training loss function. S22, obtaining a depth map of the three-dimensional reconstruction of the original image data set through the deep learning network training in step S21, and filtering the depth map using photometric consistency and geometric consistency methods; S23. The filtered depth map can be projected from the pixel coordinate system to the world coordinate system and converted into a coordinate point cloud, thereby realizing three-dimensional reconstruction. The reconstructed three-dimensional model is fitted to obtain a three-dimensional reconstructed model of the chemical pipeline network.
4. The digital twin modeling method for a chemical pipeline network according to claim 1, characterized in that: The standard model in step S3 is a real three-dimensional coordinate point model of the chemical pipeline network obtained by using model making software to make a corresponding three-dimensional model based on the physical parameters of each pipeline, valve, pump, and component equipment in the chemical pipeline network and connecting them according to the connection method of each component in the collected pipeline network.
5. The digital twin modeling method for a chemical pipeline network according to claim 1, characterized in that: The specific steps of comparing the three-dimensional reconstructed model of the chemical pipe network with the standard model in step S3 are as follows: S31, introducing an angle threshold for the chemical pipeline network standard model and the three-dimensional reconstruction model to extract edge contour points of the three-dimensional model, and using the contour points extracted from the standard model and the three-dimensional reconstruction model as their respective feature points; S32. Count the number of feature points m extracted from the standard model te and the number of feature points m extracted from the 3D reconstruction model tc Calculate the number of feature points to extract the matching degree ζ t , the specific formula is: ; S33, counting the number of feature points with consistent three-dimensional coordinates m ty And the number of remaining feature points m of the standard model es , the standard model feature points with inconsistent 3D coordinates and the 3D reconstruction model feature points are represented by (x ei ,y ei ,z ei )、(x cj ,y cj ,z cj ) represents the calculation of the average value D of the minimum distance between the inconsistent standard model feature points and the 3D reconstruction model feature points ye , the specific formula is: The feature point position matching degree ζ is calculated based on the average value of the minimum distance between the inconsistent standard model feature points and the 3D reconstruction model feature points and the number of feature points with consistent 3D coordinates. w , the specific formula is: , where D yt The minimum distance allowed is preset; S34, calculating the overall matching degree ζ of the feature points based on the matching degree of the number of feature points extracted and the matching degree of the feature point positions z , the specific formula is: , e is a natural constant; S35. Compare the calculated overall matching degree of the feature points with the preset lower limit of the overall matching degree of the feature points. If the calculated value is greater than or equal to the preset lower limit, the three-dimensional reconstructed model is judged to have passed the inspection. Otherwise, the three-dimensional reconstructed model is judged to have failed the inspection. The three-dimensional reconstructed model that has passed the inspection is marked as a static environment model of the chemical pipeline network. When the composition of the chemical pipeline network changes, the static environment model of the chemical pipeline network will be automatically updated.
6. The digital twin modeling method for a chemical pipeline network according to claim 1, characterized in that: The specific steps of model construction in step S4 are as follows: S41, setting each node in the pipe network based on the collected physical parameters of each pipe, valve, pump, and component equipment in the chemical pipe network and the connection method of each component in the pipe network; S42, performing mass conservation analysis, momentum conservation analysis and energy conservation analysis on each node in the pipe network and establishing corresponding mathematical expressions; S43. Based on the characteristics of the dataset used and the model building requirements, the network architecture is selected and the network structure and number of layers are designed. After the model used is determined, the relevant packages are imported and important global parameters are modified. At the same time, the above mathematical expressions are used as constraints to build a dynamic operation model of the chemical pipeline network; S44, divide the historical operation data into training set and test set to train and test the constructed model, and record the number of normal operation data in the test sample n yz , the number of abnormal running data n yf , the number of normal and abnormal operation data correctly identified n sz ; S45. Calculate the running recognition accuracy λ based on the test results s , the specific formula is: ; S46. Compare the calculated operation recognition accuracy with the preset standard operation recognition accuracy. If the calculated value is greater than or equal to the preset value, the model passes the test; otherwise, the model fails the test and needs to be optimized until it passes the test.
Citation Information
Patent Citations
Pipe network digital twinborn scene generation method
CN115546394A
Digital twin modeling method and simulation test system thereof
CN116822100A
Pipeline detection system based on digital twinning, system construction method and detection method
CN117807877A
Chemical pipeline visualization method based on BIM and UE5
CN117874893A
Oil-gas pipe network digital twin modeling method and system based on parallel system
CN117973194A
