Cable-strut tension structure force finding method, device, equipment, medium and product

Abstract

The invention discloses a force finding method, device and equipment for a cable-strut tension structure, a medium and a product, and relates to the field of computer aided design. The method comprises the steps that geometric topology information of a cable-strut structure is analyzed from a CAD file, graph network data are constructed according to the geometric topology information, and the graph network data comprise an incidence matrix; wherein nodes in the graph network data comprise stay bar nodes and boundary nodes, and edges comprise stay bar units and cable units; inputting the incidence matrix into a pre-trained force distribution prediction model, and obtaining unit force prediction values corresponding to the stay bar unit and the cable unit; wherein a loss function of the training force distribution prediction model is constructed based on physical constraints, and comprises force distribution constraint loss and force balance residual error loss. According to the scheme, the topology and physical dual-drive concept is adopted, the force finding precision of a high-dimensional complex topological structure is greatly improved, meanwhile, universality and general applicability are enhanced, and the automation degree is improved.

Description

Technical Field

[0001] This application relates to the field of computer-aided design technology, and in particular to a method, device, equipment, medium and product for finding force in a cable tension structure. Background Technology

[0002] Prestressed cable-stayed structures (such as cable domes and tensioned monolithic structures) are the core form of large-span spatial structures, and their stiffness relies entirely on the self-balancing prestress between the cables (tension) and the rods (compression). Force-finding analysis is the core step in the design of this type of structure, aiming to determine a reasonable prestress distribution so that the structure meets the self-balancing condition under zero external load.

[0003] Traditional force analysis methods, including nonlinear finite element methods, dynamic relaxation methods, singular value decomposition methods, and heuristic algorithms, require manual grouping to consider symmetry conditions, pre-set initial stress values, and involve complex iterative calculations and convergence conditions. Data-driven neural networks, such as BPNN (Back-Propagation Neural Network) and RBFNN (Radial Basis Function Neural Network), rely on large amounts of finite element simulation data for training. Obtaining high-fidelity samples in practical engineering is costly, and their generalization ability to untrained novel structural topologies is weak. Novel neural network architectures like Physical-Informed Neural Networks (PINNs), while embedding equilibrium equations into loss functions to reduce data dependence, fail to explicitly express the topological relationships of the encoded structure, as well as the characteristics of elements and nodes (node ​​positions, element lengths, etc.). Establishing constraints requires manual grouping during preprocessing, resulting in low convergence accuracy when handling high-dimensional complex topologies. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, device, equipment, medium, and product for finding the force in a cable tension structure to address the above-mentioned technical problems, improve the accuracy and automation of force finding, and enhance its versatility.

[0005] Firstly, this application provides a method for finding the force in a cable-stayed tension structure. The method includes:

[0006] The geometric topology information of the cable-stayed structure is parsed from the CAD file, and a graph network data is constructed based on the geometric topology information. The graph network data includes an incidence matrix. The nodes in the graph network data include strut nodes and boundary nodes, and the edges include strut elements and cable elements.

[0007] Input the correlation matrix into the pre-trained force distribution prediction model to obtain the predicted unit force values ​​for the strut unit and cable unit;

[0008] Among them, the loss function for training the force distribution prediction model is constructed based on physical constraints, including force distribution constraint loss and force balance residual loss;

[0009] The force distribution constraint loss uses the variance of the predicted force values ​​of corresponding units for edges of the same category as the force distribution constraint loss; the edge category is determined based on the distance from the midpoint of the edge to the origin.

[0010] The force balance residual loss is obtained by acquiring the resultant force of nodes based on the correlation matrix and the predicted value of element force. The resultant force residual of the strut node in the three coordinate axis directions of the three-dimensional coordinate system corresponding to the geometric topology information is extracted, and the L2 norm of the resultant force residual is used as the force balance residual loss in the three coordinate axis directions.

[0011] In one embodiment, the geometric topology information includes node coordinates, node type, node connection relationship, and edge type;

[0012] Among them, the node categories include strut nodes and boundary nodes. Strut nodes include upper strut nodes and lower strut nodes corresponding to the upper and lower ends of the same strut unit, respectively.

[0013] In one embodiment, constructing graph network data based on geometric topology information includes:

[0014] Convert node coordinates into floating-point tensors;

[0015] Obtain the side lengths of each strut element and each cable element based on the floating-point tensor;

[0016] Obtain the center point coordinates of each edge based on the node coordinates, obtain the Euclidean distance from the center point coordinates to the origin, classify edges with the same Euclidean distance into the same edge category, and construct a mapping relationship between edge category and edge;

[0017] Nodes are encoded according to their categories to obtain node category labels, including nodes on the strut, nodes below the strut, and boundary nodes;

[0018] Encode the relationship between edges and nodes into matrix form to obtain the association matrix;

[0019] Generate graph network data based on node coordinates, node category labels, node connection relationships, edge categories, edge lengths, and the association matrix.

[0020] In one embodiment, the force distribution prediction model includes, in sequence:

[0021] The input layer receives the correlation matrix and flattens it into a one-dimensional vector.

[0022] Several cascaded hidden layers are used to receive and process one-dimensional vectors; the hidden layers are connected by non-linear activation functions.

[0023] The output layer is used to output the predicted unit force values. The output layer includes the Softplus activation function.

[0024] In one embodiment, the Softplus activation function applies a negative Softplus activation to the hidden layer output corresponding to the strut unit, and applies a Softplus activation followed by a linear transformation of 1 to the hidden layer output corresponding to the cable unit.

[0025] In one embodiment, the predicted unit force values ​​are visualized to obtain a three-dimensional structural topology map with a prestress distribution cloud map, and graph network data is bound to the three-dimensional structural topology map.

[0026] Secondly, this application also provides a force-finding device for a cable tension structure. The device includes:

[0027] The data extraction module is used to parse the geometric topology information of the cable-stayed structure from the CAD file, construct graph network data based on the geometric topology information, and generate the association matrix of the graph network data; wherein, the nodes in the graph network data include strut nodes and boundary nodes, and the edges include strut elements and cable elements;

[0028] The prediction output module inputs the correlation matrix into the pre-trained force distribution prediction model to obtain the predicted unit force values ​​corresponding to the strut unit and cable unit;

[0029] Among them, the loss function for training the force distribution prediction model is constructed based on physical constraints, including force distribution constraint loss and force balance residual loss;

[0030] The force distribution constraint loss uses the variance of the predicted force values ​​of corresponding units for edges of the same category as the force distribution constraint loss; the edge category is determined based on the distance from the midpoint of the edge to the origin.

[0031] The force balance residual loss is obtained by acquiring the resultant force of nodes based on the correlation matrix and the predicted value of element force. The resultant force residual of the strut node in the three coordinate axis directions of the three-dimensional coordinate system corresponding to the geometric topology information is extracted, and the L2 norm of the resultant force residual is used as the force balance residual loss in the three coordinate axis directions.

[0032] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method for finding the force in a cable tension structure.

[0033] Fourthly, this application also provides a computer-readable storage medium. This computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps in the above-described method for finding the force in a cable tension structure.

[0034] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the steps in the above-described method for finding the force in a cable tension structure.

[0035] The aforementioned method, apparatus, equipment, medium, and product for finding forces in cable-stayed tension structures include: parsing the geometric topology information of the cable-stayed structure from a CAD file; constructing graph network data based on the geometric topology information, the graph network data including an association matrix; wherein the nodes in the graph network data include strut nodes and boundary nodes, and the edges include strut elements and cable elements; inputting the association matrix into a pre-trained force distribution prediction model to obtain the predicted values ​​of the element forces corresponding to the strut elements and cable elements; wherein the loss function for training the force distribution prediction model is constructed based on physical constraints, including force distribution constraint loss and force balance residual loss; the force distribution constraint loss uses the variance of the predicted values ​​of element forces corresponding to the same type of edge as the force distribution constraint loss; the force balance residual loss obtains the resultant force of the nodes based on the association matrix and the predicted values ​​of element forces, extracts the resultant force residuals of the strut nodes in the three coordinate axes of the three-dimensional coordinate system corresponding to the geometric topology information, and uses the L2 norm of the resultant force residuals as the force balance residual losses in the three coordinate axes. The above scheme adopts the concept of dual-driven topology and physics, and proposes a GNN-PINN (Graph Neural Network-Physical Information Neural Network) hybrid architecture for force finding in cable tension structures. This greatly improves the force finding accuracy of high-dimensional complex topological structures, while enhancing versatility and universal applicability. Moreover, since the force distribution prediction model can automatically process the input and generate the phase-generated prestress distribution based on the parameters and algorithms learned during its pre-training after obtaining the input data, manual intervention is reduced and the degree of automation is improved. Attached Figure Description

[0036] Figure 1 This is a flowchart illustrating the force-finding method for a cable tension structure in one embodiment. Figure 1 ; Figure 2 This is a flowchart illustrating the force-finding method for a cable tension structure in one embodiment. Figure 2 ; Figure 3 This is a three-dimensional topological diagram of a sunflower-shaped spoke cable truss structure in one embodiment; Figure 4 Here is a three-dimensional topological diagram of the Kivitso dome structure in one embodiment; Figure 5 This is a structural block diagram of the force-finding device for the cable tension structure in one embodiment; Figure 6 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0037] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0038] This application provides a method for finding the force in a cable tension structure, such as... Figure 1 and 2 As shown, it includes the following steps:

[0039] Step 102: Parse the geometric topology information of the cable-stayed structure from the CAD file, and construct graph network data based on the geometric topology information. The graph network data includes an association matrix. The nodes in the graph network data include strut nodes and boundary nodes, and the edges include strut elements and cable elements.

[0040] Import the 3D cable-stayed structure CAD file and perform CAD geometric analysis and topology extraction. Read line segments (edges) from the CAD file and extract the start and end coordinates (node ​​coordinates) of each segment. Determine the geometric center of the 3D cable-stayed structure based on the extracted node coordinates. If the coordinates (x, y, z) of each node are known... i y i , z i The coordinates (Cx, Cy, Cz) of the geometric center can be calculated using the following formula:

[0041]

[0042]

[0043]

[0044] Here, N represents the number of nodes. Through coordinate translation, the original node coordinates are transformed to a new coordinate system based on the geometric center coordinates. Specifically, the center point coordinates are subtracted from all node coordinates, and the node coordinates are updated. Subsequent operations are all performed based on the updated node coordinates.

[0045] When extracting nodes and coordinates, key elements need to be identified, including strut elements, strut nodes, and boundary nodes. First, identify line segments from the extracted segments that satisfy |x1-x2|<0.001 and |y1-y2|<0.001, and have different z-coordinates. These line segments can be considered as approximately vertically positioned strut elements, with the upper and lower ends of each strut element being the upper and lower nodes, respectively. Using the nodes at both ends of the strut element, the node with the larger z-coordinate value is identified as the upper strut node, and the node with the smaller z-coordinate value is identified as the lower strut node. All other nodes besides the strut nodes are boundary nodes. Boundary nodes are subject to constraints, so node balance calculations are not required. All other line segments besides the strut elements are cable elements.

[0046] After the above operations, the geometric topology information of the cable-stayed structure is obtained, including node types (boundary nodes, strut upper nodes, strut lower nodes), node coordinates, and node connection relationships.

[0047] Next, graph network data is constructed using geometric topology information. Specific operations include:

[0048] 1. Coordinate Tensorization: Converts node coordinates into floating-point tensors, mathematically represented as: , This represents the coordinate vector of a node along the x, y, and z axes, where N is the number of nodes.

[0049] 2. Element attribute identifier: Side length calculation, using the Euclidean distance formula: In the formula It is the side length. and Let i and j represent the coordinates of the two ends of the edge, respectively. and ;

[0050] 3. Edge Category Encoding: Take the center point of each edge and calculate the L2 norm from the center point to the origin, i.e., obtain the Euclidean distance from the center point of each edge to the origin. Edges with the same Euclidean distance are classified as edges of the same category. Establish a mapping dictionary from each edge category set to the edge category name, such as {"k1":[0,3,5], "k2":[1,4,8], ...}, where k1 and k2 are the edge category names corresponding to the edge categories, and [0,3,5] and [1,4,8] are the edge number sets in the corresponding k1 and k2 types; the edge category name is used as the edge type identifier to define the edge category; edge_stype: puts the indices of edges of the same type in a set;

[0051] After edge category coding, struts and cables are generally classified into different edge categories. Based on the overall structure of the cable-stayed structure, edges with different positional characteristics are classified into different types. The accompanying effect of this process is to automatically distinguish strut units and cable units. Edges of the same category exhibit mechanical similarity. This coding method is applicable to both symmetrical and asymmetrical cable-stayed structures. The difference is that asymmetrical cable-stayed structures have a larger number of types.

[0052] 4. Node type coding: Definition 0: Boundary node (fixed constraint); 1: Node on the strut (free node); 2: Node below the strut (free node).

[0053] 5. Construction of the association matrix: Encode the node-edge connection relationships into matrix form and build the association matrix. Correlation matrix The row and number The elements of a column are denoted as , representing a node and edge The connection relationship between nodes. If nodes It is the edge The starting point, then =1, if node It is the edge The endpoint, then =-1, if node With edge If not connected, then =0.

[0054]

[0055] The correlation matrix is ​​used for subsequent topological feature identification and node force balance conditions.

[0056] The above data is automatically encapsulated to output a graph network data object that conforms to the PyG (PyTorch Geometric) standard, which includes: geometric features (node ​​coordinates), topological features (node ​​connection relationships), physical features (edge ​​lengths), conditional features (edge / node types), and mechanical features (incidence matrix).

[0057] Step 104: Input the correlation matrix into the pre-trained force distribution prediction model to obtain the predicted values ​​of the unit forces corresponding to the strut unit and the cable unit.

[0058] The force distribution prediction model includes, in order:

[0059] The input layer receives the correlation matrix and flattens it into a one-dimensional vector.

[0060] Several cascaded hidden layers are used to receive one-dimensional vectors for processing; the hidden layers are connected by non-linear activation functions; the function of the hidden layers is to extract and transform features from the input data; the neurons in the hidden layers are interconnected with all neurons in the previous and next layers; each hidden layer can learn features at different levels and abstractions.

[0061] The output layer is used to output the predicted unit force values. The output layer includes the Softplus activation function.

[0062] When constructing the force distribution prediction model, a multi-layer fully connected neural network is first established. The flattened vector of the correlation matrix (dimension: number of nodes N × number of edges M) is used as the input layer feature of the cable-stayed structure topology. Each hidden layer in the fully connected neural network has the same dimension, and a non-linear activation function is used between adjacent hidden layers. Connection, output layer output element force prediction value .because It contains tension cable forces and compression strut forces; therefore, the linear transformation of the nonlinear activation function of the network output using the Softplus activation function with classification labels is as follows:

[0063]

[0064] In the formula, It is the first Predicted element force values ​​for the strip edge. It is a set of strut element indices.

[0065] A negative Softplus activation is applied to the strut elements to ensure that the internal force is negative (i.e., compression); a linear transformation of Softplus activation plus 1 is applied to the non-strut elements to ensure that the minimum cable force is not less than 1. This set of linear transformations of the Softplus activation function is a key innovation that strictly follows the mechanical principle of "struts are under compression, cable elements are under tension," enabling the output of the force distribution prediction model to distinguish between cable tension and strut compression in terms of data representation.

[0066] During training, the loss function of the force distribution prediction model is constructed based on physical constraints. The loss function includes force distribution constraint loss and force balance residual loss. The force distribution constraint loss uses the variance of the predicted force values ​​of the corresponding units of the same type of edge as the force distribution constraint loss. The force balance residual loss obtains the resultant force of the nodes based on the correlation matrix and the predicted force values ​​of the units, extracts the resultant force residual of the strut nodes in the three coordinate axis directions of the three-dimensional coordinate system corresponding to the geometric topology information, and uses the L2 norm of the resultant force residual as the force balance residual loss in the three coordinate axis directions.

[0067] The force distribution constraint loss is based on the physical law that the internal forces of edges of the same type should be as similar as possible. It calculates the variance of the internal forces of edges of the same type and uses this variance as the force distribution constraint loss to penalize mechanical dispersion. The specific formula is as follows:

[0068] No. Mean of category edges calculate:

[0069]

[0070] No. Variance of class edges calculate:

[0071]

[0072] Force distribution constraint loss calculate:

[0073]

[0074] In the formula, It is the number of edges in the k-th category. It represents the number of edge categories.

[0075] The force balance residual loss is constructed from the perspective of nodal static equilibrium. The force balance residual loss is based on the correlation matrix. With force vector Calculate the resultant force at the nodes, extract the resultant force residuals of the free nodes (upper and lower nodes of the strut) in the x, y, and z axes, and use the L2 norm of the resultant force residuals as the loss terms of the force balance residuals in the x, y, and z directions, respectively.

[0076] The formula for calculating the force balance residual loss is as follows:

[0077] Calculate the first Force vectors on each side :

[0078]

[0079] In the formula, and They are the first The starting and ending coordinate vectors of each edge. The resultant force matrix of all nodes. The calculation is as follows:

[0080]

[0081] In the formula, It is an N×M incidence matrix. It is the set of force vectors for all edges. M represents the total number of edges. The resultant force residuals at the free nodes (upper and lower nodes of the strut) in the x, y, and z axes are as follows:

[0082]

[0083]

[0084]

[0085] In the formula, A set of index numbers for the free nodes (the upper and lower nodes of the strut). Represents the resultant force matrix of the free nodes. Representing the resultant force matrix The components on the x-axis, Representing the resultant force matrix Components on the y-axis Representing the resultant force matrix The components on the z-axis. The force balance residual loss function is established as follows:

[0086]

[0087] The total loss function is:

[0088] T_loss=

[0089] In the formula, It is the weight of the force balance residual loss. This is the force distribution constraint loss weight. This composite loss function ensures that the cable-stayed structure simultaneously satisfies the static equilibrium condition of the nodes and the uniformity of the force distribution within the strut elements.

[0090] Before training the force distribution prediction model, initialize the call graph network data, total loss function, and optimizer, creating an Adam optimizer with an adaptive learning rate of 0.001. Begin iterative training, starting with the correlation matrix... As input features, these are fed into the neural network, passing through an input layer, multiple hidden layers, and an output layer, outputting predicted unit force values. Then The total loss function T_loss is calculated by substituting the values ​​into the total loss function; then, backpropagation of the neural network is performed to reduce the total loss function T_loss. Training stops when the total loss falls below a set tolerance limit or reaches a preset number of iterations, and the final unit force prediction value is output. .

[0091] In one embodiment, the predicted unit force values ​​are visualized to obtain a three-dimensional structural topology map with a prestress distribution cloud map, and graph network data is bound to the three-dimensional structural topology map.

[0092] For example Figure 3 Taking the sunflower-shaped spoked cable truss structure shown as an example, the specific process of finding the force in the cable tension structure is explained, including the following operations:

[0093] Import the 3D geometric topology from CAD, which contains 168 nodes and 408 edges. The nodes are divided into three categories: 24 boundary nodes, 72 strut-top nodes, and 72 strut-bottom nodes. Use the `torch_geometric` class from the PyTorch library to build the graph network structure data, including node coordinates, node connections, strut element indices (with special identifiers to distinguish strut elements from edge elements), node categories, and the association matrix B.

[0094] A four-layer neural network was constructed, consisting of an input layer, two hidden layers, and a linearly transformed output layer. The Tanh activation function was used between the input and hidden layers, while the linearly transformed activation functions -Softplus and Softplus+1 were used between the hidden layers and the output layer for the strut and cable, respectively. The total loss function (force distribution constraint loss and resultant force balance residual loss) was calculated using the output force distribution. Then, backpropagation was performed, and the Adam optimizer was used to minimize the total loss function. After 3000 iterations, the resultant force balance residual loss decreased to 8.9. e-4 The force distribution constraint loss decreased to 7.0. e-5 Ultimately, a feasible force distribution is obtained.

[0095] For example Figure 4 Taking the Kivitso dome structure shown as an example, the specific process of finding the force in a cable tension structure is explained, including the following operations:

[0096] Import a 3D geometric topology from CAD, containing 56 nodes and 145 edges. The nodes are divided into three categories: 18 boundary nodes, 19 strut-top nodes, and 19 strut-bottom nodes. Use the `torch_geometric` class from the PyTorch library to build the graph network structure data, including node coordinates, edge connections, strut element indices (special element identifiers to distinguish strut elements from edge elements), node position categories, and the association matrix B.

[0097] A four-layer neural network was constructed, consisting of an input layer, two hidden layers, and a linearly transformed output layer. The Tanh activation function was used between the input and hidden layers, while the linearly transformed activation functions -Softplus and Softplus+1 were used between the hidden layers and the output layer for the strut and cable, respectively. The total loss function (force distribution constraint loss and resultant force balance residual loss) was calculated using the output force distribution. Then, backpropagation was performed, and the Adam optimizer was used to minimize the total loss function. After 3000 iterations, the resultant force balance residual loss decreased to 6.3. e-5 The force distribution constraint loss decreased to 1.9. e-6 Ultimately, a feasible force distribution is obtained.

[0098] This invention transforms cable-stayed structures into graph network data, creating a graph neural network. Based on digitized topological relationships, it can automatically generate force balance residual conditions and force distribution constraints, explicitly learn the force transmission paths under topological constraints, and construct equilibrium equations using the correlation matrix B in the loss function. =0, ensuring the solution satisfies the fundamental laws of mechanics. This leads to the construction of a GNN-PINN hybrid architecture, which simultaneously enhances the topological adaptability and physical conformity of prestressed cable-stayed structures, significantly improving the force-finding accuracy of high-dimensional complex topologies and enhancing its versatility and universal applicability. Furthermore, this method is applicable to all prestressed cable-stayed structures with nodal equilibrium topology conditions, exhibiting extremely high convergence accuracy even in high-dimensional problems. Finally, by importing the structural topology, prestress distribution can be automatically generated, improving automation and making implementation easier.

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

[0100] Based on the same inventive concept, this application also provides an apparatus for implementing the above-mentioned cable tension structure force finding method. The solution provided by this apparatus is similar to the solution described in the above method. Therefore, the specific limitations of one or more cable tension structure force finding apparatus embodiments provided below can be found in the limitations of the cable tension structure force finding method above, and will not be repeated here.

[0101] In one embodiment, such as Figure 5 As shown, a force-finding device for a cable tension structure is provided, comprising:

[0102] The data extraction module is used to parse the geometric topology information of the cable-stayed structure from the CAD file, construct graph network data based on the geometric topology information, and generate the association matrix of the graph network data; wherein, the nodes in the graph network data include strut nodes and boundary nodes, and the edges include strut elements and cable elements;

[0103] The prediction output module inputs the correlation matrix into the pre-trained force distribution prediction model to obtain the predicted unit force values ​​corresponding to the strut unit and cable unit;

[0104] Among them, the loss function for training the force distribution prediction model is constructed based on physical constraints, including force distribution constraint loss and force balance residual loss;

[0105] The force distribution constraint loss uses the variance of the predicted force values ​​of corresponding units for edges of the same category as the force distribution constraint loss; the edge category is determined based on the distance from the midpoint of the edge to the origin.

[0106] The force balance residual loss is obtained by acquiring the resultant force of nodes based on the correlation matrix and the predicted value of element force. The resultant force residual of the strut node in the three coordinate axis directions of the three-dimensional coordinate system corresponding to the geometric topology information is extracted, and the L2 norm is used as the force balance residual loss in the three coordinate axis directions respectively.

[0107] Each module in the aforementioned cable tension structure force-finding device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the computer device's memory as software, so that the processor can call and execute the corresponding operations of each module.

[0108] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 6As shown, the computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a method for finding the force in a cable-stayed tension structure. The display unit is used to form a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0109] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer equipment on which the present application is applied.

[0110] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in all of the above method embodiments.

[0111] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in all of the above method embodiments.

[0112] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in all of the above method embodiments.

[0113] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0114] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0115] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0116] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for finding the force in a cable-stayed tension structure, characterized in that, The method includes: The geometric topology information of the cable-stayed structure is parsed from the CAD file, and a graph network data is constructed based on the geometric topology information. The graph network data includes an association matrix. The nodes in the graph network data include strut nodes and boundary nodes, and the edges include strut elements and cable elements. The correlation matrix is ​​input into a pre-trained force distribution prediction model to obtain the predicted unit force values ​​corresponding to the strut unit and the cable unit; The loss function for training the force distribution prediction model is constructed based on physical constraints, including force distribution constraint loss and force balance residual loss. The force distribution constraint loss uses the variance of the predicted force value of the unit corresponding to the edge of the same category as the force distribution constraint loss; the edge category is divided according to the distance from the midpoint of the edge to the origin of the coordinate system. The force balance residual loss is obtained based on the correlation matrix and the predicted unit force to obtain the resultant force of the nodes. The resultant force residual of the strut node in the three coordinate axis directions of the three-dimensional coordinate system corresponding to the geometric topology information is extracted, and the L2 norm of the resultant force residual is used as the force balance residual loss in the three coordinate axis directions respectively.

2. The method according to claim 1, characterized in that: The geometric topology information includes node coordinates, node categories, and node connection relationships; The node categories include the strut nodes and the boundary nodes. The strut nodes include upper strut nodes and lower strut nodes corresponding to the upper and lower ends of the same strut unit, respectively.

3. The method according to claim 2, characterized in that, The process of constructing graph network data based on the geometric topology information includes: Convert the node coordinates into floating-point tensors; The side lengths of each strut unit and each cable unit are obtained based on the floating-point tensor. The center point coordinates of each edge are obtained based on the node coordinates. The Euclidean distance from the center point coordinates to the origin is obtained. Edges with the same Euclidean distance are classified into the same edge category. A mapping relationship is constructed between the edge category and the edge. The nodes are encoded according to the node category to obtain node category tags including nodes on the strut, nodes below the strut, and boundary nodes; The connection relationship between edges and nodes is encoded into matrix form to obtain the association matrix; Graph network data is generated based on the node coordinates, node category labels, node connection relationships, edge categories, edge lengths, and the association matrix.

4. The method according to claim 1, characterized in that, The force distribution prediction model includes, in sequence: An input layer is used to receive the correlation matrix and flatten the correlation matrix into a one-dimensional vector. A series of cascaded hidden layers are used to receive the one-dimensional vector for processing; the hidden layers are connected by non-linear activation functions. An output layer is used to output the predicted unit force value, and the output layer includes a Softplus activation function.

5. The method according to claim 4, characterized in that, The Softplus activation function applies a negative Softplus activation to the hidden layer output corresponding to the strut unit, and applies a Softplus activation to the hidden layer output corresponding to the cable unit followed by a linear transformation with 1.

6. The method according to claim 1, characterized in that, The method further includes: The predicted unit force values ​​are visualized to obtain a three-dimensional structural topology map with a prestress distribution cloud map, and the graph network data is bound to the three-dimensional structural topology map.

7. A force-finding device for a cable tension structure, characterized in that, The device includes: The data extraction module is used to parse the geometric topology information of the cable-stayed structure from the CAD file, construct graph network data based on the geometric topology information, and generate the association matrix of the graph network data; wherein, the nodes in the graph network data include strut nodes and boundary nodes, and the edges include strut elements and cable elements; The prediction output module inputs the correlation matrix into the pre-trained force distribution prediction model to obtain the predicted unit force values ​​corresponding to the strut unit and the cable unit; The loss function for training the force distribution prediction model is constructed based on physical constraints, including force distribution constraint loss and force balance residual loss. The force distribution constraint loss uses the variance of the predicted force values ​​of the units corresponding to edges of the same category as the force distribution constraint loss; the edge category is divided according to the distance from the midpoint of the edge to the origin of the coordinate system. The force balance residual loss is obtained based on the correlation matrix and the predicted unit force to obtain the resultant force of the nodes. The resultant force residual of the strut node in the three coordinate axis directions of the three-dimensional coordinate system corresponding to the geometric topology information is extracted, and the L2 norm of the resultant force residual is used as the force balance residual loss in the three coordinate axis directions respectively.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images