A hole edge beam intelligent arrangement method and system based on multi-layer mask fusion

Abstract

The application relates to the technical field of engineering girder arrangement, and particularly relates to a hole edge girder intelligent arrangement method and system based on multi-layer mask fusion. The method comprises the following steps: generating an initial grid girder arrangement diagram according to size information, and performing simulated collision on the initial grid girder arrangement diagram and hole information to obtain a potential grid girder arrangement diagram; generating a potential hole edge girder arrangement diagram according to a hole position and a hole size; splicing the potential grid girder arrangement diagram and the potential hole edge girder arrangement diagram to obtain a potential secondary girder arrangement diagram; performing pixelization processing on the potential secondary girder arrangement diagram to obtain a potential secondary girder feature tensor; generating a three-layer weighted mask matrix; inputting the potential secondary girder feature tensor, structure design parameters and the three-layer weighted mask matrix into a pre-trained bidirectional interaction generation network model to generate a hole edge girder design diagram. The application can solve the problem of low design and arrangement efficiency caused by the dependence on manual operation in the prior art.

Description

Technical Field

[0001] This application relates to the field of engineering beam arrangement technology, and in particular to an intelligent arrangement method and system for opening edge beams based on multi-layer mask fusion. Background Technology

[0002] Currently, in the design of secondary beam structures in steel-framed buildings, floor slabs often require openings to accommodate the placement of equipment, staircases, or elevator shafts. When multiple openings exist within a single section or floor slab, it leads to discontinuous structural force paths, reduced local stiffness, and stress concentration. Therefore, when floor slabs have openings, simply arranging section beams is insufficient; additional edge beams are needed to reinforce the openings. However, the integrated design of section beams and edge beams heavily relies on engineers' experience, requiring manual iterative calculations and repeated adjustments. This results in insufficient automation, high manpower consumption, and low efficiency. Summary of the Invention

[0003] To address the aforementioned issues, this application provides a method and system for intelligent arrangement of opening edge beams based on multi-layer mask fusion, which can solve at least one of the problems in the prior art.

[0004] According to one aspect of the embodiments of this application, a method for intelligent arrangement of opening edge beams based on multi-layer mask fusion is proposed, the method comprising: Obtain the size information of the target grid area and the opening information located within the target grid area, wherein the opening information includes the opening location and opening size; An initial grid beam layout diagram is generated based on the size information, and the initial grid beam layout diagram is simulated to collide with the opening information to eliminate the layout lines in the initial grid beam layout diagram that collide with the opening information, thereby obtaining a potential grid beam layout diagram. A potential layout diagram of the side beams of the opening is generated based on the location and size of the opening. By splicing the potential grid beam layout diagram with the potential hole edge beam layout diagram, a potential secondary beam layout diagram is obtained. The potential secondary beam layout diagram is pixelated to obtain the potential secondary beam feature tensor; The potential secondary beam feature tensor is matrix-encoded and weighted to generate a three-layer weighted mask matrix. The potential secondary beam feature tensor, the three-layer weighted mask matrix, and the preset structural design parameters are input into a pre-trained bidirectional interactive generative network model to generate the design drawing of the opening side beam. The structural design parameters are the seismic intensity and structural design height of the target building.

[0005] In the above scheme, the step of simulating the collision between the initial grid beam layout diagram and the opening information to obtain the potential grid beam layout diagram includes: Extract each beam layout line from the initial grid beam layout diagram, where each beam layout line is composed of the coordinates of the nodes at both ends; Based on the location of the opening, determine the location of each sub-opening from the opening information; Based on the opening dimensions, determine the diameter of each of the sub-openings; For each sub-opening, the boundary line of the opening is determined according to the position and diameter of the sub-opening. If the boundary line of the opening intersects with any of the beam arrangement lines, the beam arrangement lines that intersect with the boundary line of the opening are discarded. The beam layout lines that were not removed are retained to obtain the potential grid beam layout diagram.

[0006] In the above scheme, generating a potential side beam layout diagram based on the location and size of the opening includes: The boundaries of each sub-opening are determined based on the location of each sub-opening in the opening location and the diameter of each sub-opening in the opening size; Arrange the edge beams at the boundaries of each of the sub-openings to generate the potential edge beam arrangement diagram.

[0007] In the above scheme, after obtaining the potential secondary beam layout diagram, the method further includes: Extract tangent information from the potential secondary beam layout diagram to characterize the tangential connection relationship between the hole-edge beam and the grid beam; A potential hole tangent beam layout diagram is generated based on the tangent information, and the position of the hole tangent beam is determined based on the potential hole tangent beam layout diagram.

[0008] In the above scheme, the step of pixelating the potential secondary beam layout diagram to obtain the potential secondary beam feature tensor includes: Determine the boundary lines in the potential secondary beam layout diagram; Based on a preset length-pixel conversion ratio, each boundary line of the potential grid beam layout, each boundary line of the potential hole edge beam layout, and each boundary line of the potential hole tangent beam layout are pixelated and encoded to obtain the potential secondary beam feature tensor.

[0009] In the above scheme, the step of matrix encoding and weight fusion of the potential secondary beam feature tensor to generate a three-layer weighted mask matrix includes: Initialize a mask matrix with all three elements equal to zero; The pixel positions of the grid beam, the hole tangent beam, and the hole edge beam are extracted from the potential secondary beam feature tensor, respectively. Based on the pixel positions of the grid beams, the hole tangent beams, and the hole edge beams, pixel elements of the three mask matrices, each with zero elements, are assigned values ​​to obtain the grid beam mask matrix, the hole tangent beam mask matrix, and the hole edge beam mask matrix. The initial value of each pixel in the grid beam mask matrix, the hole tangent beam mask matrix, and the hole edge beam mask matrix is ​​set to 1. The grid beam mask matrix, the hole tangent beam mask matrix, and the hole boundary beam mask matrix are weighted respectively to obtain the three-layer weighted mask matrix.

[0010] In the above scheme, the grid beam mask matrix, the hole tangent beam mask matrix, and the hole boundary beam mask matrix are weighted respectively to obtain the three-layer weighted mask matrix, including: Determine the area of ​​each sub-hole in the target grid area and the total number of each sub-hole; For each sub-hole, the area of ​​the sub-hole is divided by the sum of the areas of all the sub-holes to obtain the area ratio of the sub-hole among all the sub-holes, and the weighting coefficient of the sub-hole is determined based on the area ratio. Determine the first-level coefficients corresponding to the grid beam mask matrix, determine the first pixel value corresponding to each sub-opening based on the first-level coefficients, the initial pixel value in each sub-opening, and the weighting coefficients, and generate the final grid beam mask matrix based on the first pixel value corresponding to each sub-opening, wherein each sub-opening corresponds to one or more first pixel values; Determine the second-level coefficients corresponding to the hole tangent beam mask matrix, determine the second pixel value corresponding to each of the sub-holes based on the second-level coefficients, the initial pixel value in each of the sub-holes and the weighting coefficients, and generate the final hole tangent beam mask matrix based on the second pixel value corresponding to each of the sub-holes, wherein each sub-hole corresponds to one or more second pixel values; The third-level coefficients corresponding to the hole boundary beam mask matrix are determined. The third pixel value corresponding to each sub-hole is determined based on the third-level coefficients, the initial pixel value in each sub-hole, and the weighting coefficient. The final hole boundary beam mask matrix is ​​generated based on the third pixel value corresponding to each sub-hole. Each sub-hole corresponds to one or more third pixel values. The final grid beam mask matrix, the final hole tangent beam mask matrix, and the final hole boundary beam mask matrix are used as the three-layer weighted mask matrix; In this context, the pixel values ​​of non-hole areas within the target grid region are all maintained at 1.

[0011] In the above scheme, the pre-trained bidirectional interactive generative network model includes a first convolutional neural network, a second convolutional neural network, a text encoder, a cross-branch interactive attention module, an adaptive interactive fusion unit, a residual neural network, and a deconvolutional neural network.

[0012] In the above scheme, the step of inputting the potential secondary beam feature tensor, the three-layer weighted mask matrix, and the preset structural design parameters into a pre-trained bidirectional interactive generative network model to generate the opening side beam design drawing includes: The image depth feature tensor is obtained by downsampling and feature extraction of the potential secondary beam feature tensor through the first convolutional neural network. The mask depth feature tensor is obtained by downsampling and feature extraction of the three-layer weighted mask matrix through the second convolutional neural network. The text encoder extracts features from the structural design parameters to obtain a text embedding feature tensor. The cross-branch interactive attention module performs correlation attention calculation and bidirectional feature calibration on the image depth feature tensor and the mask depth feature tensor to obtain a bidirectional cross feature tensor. The adaptive interactive fusion unit dynamically solves the gating coefficients and performs feature weighted fusion on the image depth feature tensor, the mask depth feature tensor, and the bidirectional cross feature tensor to obtain the fused depth feature tensor. The residual neural network and the deconvolutional neural network are used to extract and upsample the fused deep feature tensor and the text embedding feature tensor to generate the opening side beam design tensor, so as to generate the opening side beam design drawing based on the opening side beam design tensor.

[0013] According to one aspect of the embodiments of this application, a smart arrangement system for opening edge beams based on multi-layer mask fusion is proposed, the system comprising: The acquisition unit is used to acquire the size information of the target grid area and the opening information located within the target grid area, wherein the opening information includes the opening location and the opening size; The first arrangement unit is used to generate an initial grid beam arrangement diagram based on the size information, and simulate collision between the initial grid beam arrangement diagram and the opening information to eliminate the arrangement lines in the initial grid beam arrangement diagram that collide with the opening information, so as to obtain a potential grid beam arrangement diagram. The second arrangement unit is used to generate a potential hole side beam arrangement diagram based on the hole location and the hole size; The third arrangement unit is used to splice the potential grid beam arrangement diagram with the potential hole edge beam arrangement diagram to obtain the potential secondary beam arrangement diagram. The processing unit is used to pixelate the potential secondary beam layout diagram to obtain the potential secondary beam feature tensor. The fusion unit is used to perform matrix encoding and weight fusion on the potential secondary beam feature tensor to generate a three-layer weighted mask matrix; The generation unit is used to input the potential secondary beam feature tensor, the three-layer weighted mask matrix and the preset structural design parameters into the pre-trained bidirectional interactive generative network model to generate the design drawing of the opening side beam; The structural design parameters are the seismic intensity and structural design height of the target building.

[0014] The beneficial effects of this application are as follows: First, an initial grid beam layout diagram is generated based on the size information of the target grid area. Then, a simulated collision is performed with the opening to eliminate the layout lines that conflict with the opening, thus obtaining a potential grid beam layout diagram. Next, a potential opening edge beam layout diagram is generated by combining the opening position and the opening size in the opening information. Finally, the potential grid beam layout diagram and the potential opening edge beam layout diagram are spliced ​​together to obtain a potential secondary beam layout diagram with "grid beam + opening edge beam".

[0015] Furthermore, the potential secondary beam layout diagram is pixelated to obtain the potential secondary beam feature tensor. Then, the potential secondary beam feature tensor is matrix-encoded and weighted to generate a three-layer weighted mask matrix. The potential secondary beam feature tensor, the three-layer weighted mask matrix, and structural design parameters are input into a pre-trained bidirectional interactive generative network model to automatically generate the opening side beam design diagram, thereby solving the problem of low efficiency caused by the reliance on human experience in the existing technology. Attached Figure Description

[0016] Figure 1 A flowchart illustrating the intelligent arrangement method for opening edge beams based on multi-layer mask fusion provided in this application embodiment; Figure 2 This is an overall logic diagram of the intelligent arrangement method for opening edge beams based on multi-layer mask fusion provided in the embodiments of this application; Figure 3 This is a schematic diagram illustrating the generation of the potential grid beam layout.

[0017] Figure 4 This is a schematic diagram of the generation of the potential tunnel side beam layout.

[0018] Figure 5 This is a schematic diagram of multi-layer weighted mask matrix transformation; Figure 6This is a schematic diagram of the architecture of a bidirectional interactive generative network model with multi-layer mask fusion.

[0019] Figure 7 This is a schematic diagram of the training of a bidirectional interactive generative adversarial network model with multi-layer mask fusion.

[0020] Figure 8 This is a schematic diagram of the training set preprocessing for a bidirectional interactive generative adversarial network model with multi-layer mask fusion. Figure 9 This is a block diagram of the intelligent layout system for opening edge beams based on multi-layer mask fusion provided in the embodiments of this application; Figure 10 This is a schematic diagram illustrating the frame dimensions and bay openings defined in the embodiments of this application. Detailed Implementation

[0021] To enable those skilled in the art to better understand the solutions of this application, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] It should be noted that while some processes described in the specification, claims, and accompanying drawings include multiple steps appearing in a specific order, it should be clearly understood that these steps may not be performed in the order they appear herein, or may be performed in parallel. The step numbers are merely used to distinguish different steps and do not themselves represent any execution order. Furthermore, descriptions such as "first," "second," or "objective" in this document are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. "Multiple" in this document refers to at least two.

[0023] It is worth noting that in the specific embodiments of this application, data such as size information, location information, pixel values, and layout diagrams are involved. When the above embodiments of this application are applied to specific products or technologies, permission or consent from the target object is required, and the collection, use, and processing of related data must comply with relevant laws, regulations, and standards. For example, when an embodiment of this application needs to obtain data such as size information, location information, pixel values, and layout diagrams, separate permission or consent from the target object can be obtained through pop-up windows or redirection to a confirmation page. After obtaining separate permission or consent from the target object, the relevant data such as size information, location information, pixel values, and layout diagrams used to enable the embodiments of this application to operate normally can then be obtained.

[0024] The following provides a detailed description of the specific implementation methods of the embodiments of this application: Please see Figure 1 , Figure 1 This is a flowchart illustrating the intelligent arrangement method for opening edge beams based on multi-layer mask fusion provided in this application embodiment. Figure 1 The intelligent arrangement method for opening edge beams based on multi-layer mask fusion shown includes: Step 210: Obtain the size information of the target grid area and the opening information located within the target grid area, wherein the opening information includes the opening location and opening size; Step 220: Generate an initial grid beam layout diagram based on the size information, and simulate collision between the initial grid beam layout diagram and the opening information to eliminate the layout lines in the initial grid beam layout diagram that collide with the opening information, thereby obtaining a potential grid beam layout diagram. Step 230: Generate a potential hole side beam layout diagram based on the hole location and hole size; Step 240: Combine the potential grid beam layout diagram with the potential hole edge beam layout diagram to obtain the potential secondary beam layout diagram; Step 250: Pixelate the potential secondary beam layout diagram to obtain the potential secondary beam feature tensor; Step 260: Perform matrix encoding and weight fusion on the potential secondary beam feature tensor to generate a three-layer weighted mask matrix; Step 270: Input the potential secondary beam feature tensor, the three-layer weighted mask matrix and the preset structural design parameters into the pre-trained bidirectional interactive generative network model to generate the opening side beam design drawing; The structural design parameters are the seismic intensity and structural design height of the target building.

[0025] First, the target area, such as a single floor slab or a single cell, needs effective protection to proceed with subsequent construction work. The overall logical diagram of this application is as follows: Figure 2 As shown, this includes the frame dimensions and bay definitions of the target grid area (the frame dimensions and bay definitions are also the size information described in this application). Figure 10As shown, the three values ​​[3200, 3600, 3200] in the bay definition specifically mean: three bay dimensions continuously arranged along the transverse direction of the frame structure. These three values ​​correspond to the distances between adjacent transverse structural axes, respectively. The widths of the first, second, and third transverse bays are 3200 mm, 3600 mm, and 3200 mm, respectively. Therefore, this transverse arrangement includes a total of four transverse structural axes (the two at the left and right boundaries are frame boundaries and are not drawn). The distances between adjacent axes from left to right are 3200 mm, 3600 mm, and 3200 mm, corresponding to a total transverse length of 10000 mm for the frame.

[0026] The three values ​​[500, 1000, 1500...] specifically represent several bay dimensions arranged continuously along the longitudinal direction of the frame structure, with each value corresponding to the distance between adjacent longitudinal structural axes. The widths of the first, second, and third longitudinal bays are 500mm, 1000mm, and 1500mm, respectively. (The top and bottom boundaries are frame boundaries and are not drawn.) The spacing between adjacent axes from bottom to top is successively 500mm, 1000mm, 1500mm, etc., corresponding to a total longitudinal length of 9000mm for the frame.

[0027] In the hole information, hole 1 [1500, 1000, 400], 1500 and 1000 are the x and y coordinates of hole 1 (i.e., hole location), and 400 is the diameter of hole 1 (i.e., hole size). The meaning of the values ​​in hole 2 is similar.

[0028] Please see Figure 3 As shown, the size information of the target grid area is used as the frame design condition to generate the initial grid beam layout drawing, and the potential grid beam layout drawing is obtained by combining the opening design conditions generated based on the opening information.

[0029] Further, please see Figure 4 As shown, the location and size parameters of each sub-opening are identified based on the location and size of the opening, thereby generating a potential side beam layout diagram. The potential grid beam layout diagram is then spliced ​​with the potential side beam layout diagram to obtain a potential secondary beam layout diagram.

[0030] Further, please see Figure 5 As shown, the potential secondary beam layout diagram is pixelated to obtain the potential secondary beam feature tensor; the potential secondary beam feature tensor is matrix-encoded and weighted to generate a three-layer weighted mask matrix; the potential secondary beam feature tensor, the three-layer weighted mask matrix, and structural design parameters (seismic intensity and structural design height) are input into a pre-trained bidirectional interactive generative network model to generate the opening side beam design diagram.

[0031] The internal architecture diagram of the bidirectional interactive generative network model is as follows: Figure 6 As shown, the architecture diagram of the three-layer weighted mask matrix is ​​as follows: Figure 7 As shown, the specific logic diagram is as follows: Figure 8 As shown.

[0032] The following provides a further explanation of the specific embodiments of this application: The specific implementation method of this application includes the following steps: S1: Obtain the size information of the target grid area and the opening information located within the target grid area, wherein the opening information includes the opening location and opening size; S2: Generate an initial grid beam layout diagram based on the size information, and simulate collision between the initial grid beam layout diagram and the opening information to eliminate layout lines in the initial grid beam layout diagram that collide with the opening information, thereby obtaining a potential grid beam layout diagram, and perform pixelation processing to generate a potential grid beam feature tensor; extract the opening information and perform pixelation processing and adaptive optimization to generate a potential opening edge beam feature tensor; superimpose the potential grid beam feature tensor and the potential opening edge beam feature tensor to obtain a potential secondary beam feature tensor; S3: Perform matrix encoding and weighted fusion on the feature tensor of potential secondary beams to generate a grid beam mask matrix, a hole tangent beam mask matrix, and a hole boundary beam mask matrix, for a total of three layers of weighted mask matrices; S4: Input the potential secondary beam feature tensor, the three-layer weighted mask matrix, and the structural design parameters into the pre-trained bidirectional interactive generative network model to generate the design drawing of the side beam of the opening.

[0033] In this application, by inputting structural geometric information and design parameters, and through a preprocessing adaptive module and a pre-trained bidirectional interactive generative network model, the design drawings of the opening side beams are quickly generated, which greatly improves the design efficiency of the opening side beam layout stage of steel frame building structures.

[0034] In step S1, the frame dimensions include the transverse and longitudinal lengths of the main grid; the opening information includes the longitudinal and transverse coordinates of the opening and the opening diameter / side length; the bay definition includes the transverse bay and the longitudinal bay. Structural design parameters include: building height and seismic fortification intensity.

[0035] In this application, according to relevant building seismic design codes, the inter-story drift ratio of the frame beam structure arrangement needs to be less than the minimum limit. The design of the opening edge beams and grid beams is controlled by the building height and seismic fortification intensity; therefore, the structural design parameters selected are the building height and seismic fortification intensity.

[0036] Figure 3 As shown, step S22 includes: Based on the longitudinal and transverse bay positions, grid beam layout lines are generated. Collision detection is performed between the grid beam layout lines and the openings, and all conflicting grid beam layout lines are removed. The generated grid beam layout lines are pixelated and encoded by converting the length to pixels (the scale is specifically: 1 pixel equals the actual size of 9.765mm) to obtain the potential grid beam feature tensor.

[0037] Figure 4 As shown, steps S23-S24 include: Based on the potential grid beam feature tensor and the opening location information, the sub-grids where all sub-openings are located are identified. The arrangement lines of the opening edge beams are generated along the longitudinal and transverse tangent directions of the boundaries of all sub-openings, and are orthogonal to the boundaries of the sub-grids. The arrangement lines of the opening edge beams are adaptively regularized and optimized in combination with the spatial geometric position of the openings. Then, the generated arrangement lines of the opening edge beams are pixelated and encoded by the length-to-pixel ratio conversion to obtain the potential opening edge beam feature tensor. The potential grid beam feature tensor and the potential opening edge beam feature tensor are superimposed to obtain the potential secondary beam feature tensor.

[0038] The adaptive optimization of potential opening edge beams includes the following optimizations: Optimization 1: For all sets of staggered parallel secondary beams generated at all sub-openings, determine the net spacing between the beams. If the spacing is less than 100mm, align them uniformly to the outer reference secondary beam axis to regulate the beam arrangement; Optimization 2: Detect the relative distance between all generated opening edge beams and the nearest potential grid beam. When the net spacing is less than 200mm, remove the potential opening edge beam to automatically avoid the problem of overly dense beam arrangement; Optimization 3: When the diameter or side length of the sub-opening is greater than 1500mm, the impact of inter-story stiffness reduction is significant, so an additional octagonal beam is arranged inside the opening edge beam; Optimization 4: When the diameter or side length of the sub-opening is less than 300mm, the impact of inter-story stiffness reduction is minor, so no potential opening edge beams are arranged.

[0039] Figure 5 As shown, step S3 includes: S31: Initialize a mask matrix with all three elements being zero; S32: Extract the pixel positions of the grid beams, hole tangent beams, and hole boundary beams from the potential secondary beam feature tensor, respectively; S33: Based on the extracted pixel positions, the three mask matrices are indexed and located respectively, and the pixel elements at the corresponding positions are assigned the value of one, thereby obtaining the mask matrices of the grid beam, the hole tangent beam and the hole boundary beam; S34: Combining the hierarchical coefficient and the proportion parameters of each opening, dynamically calculate the heterogeneous weighted matrix corresponding to the grid beam, the tangent beam of the opening, and the boundary beam of the opening. Then, fuse the obtained heterogeneous weighted matrix with the initial mask matrix pixel by pixel to generate a weighted mask matrix.

[0040] Among them, the grid beam mask matrix identifies the potential areas where grid beams can be arranged; the tunnel tangent beam and tunnel boundary beam mask matrices identify the potential areas where tunnel edge beams can be arranged. The physical meaning of separating the two is that the tunnel boundary beam is used to locate the position of the tunnel entrance and is distinguished from the tunnel boundary beam.

[0041] Differentiated weighted matrices for beams in each zone, tangential beams of tunnels, and boundary beams of tunnels. The calculation is expressed as follows: ; ; in, The coefficients are: grid beam (corresponding to the first level coefficient, taken as 1), tunnel tangent beam (corresponding to the second level coefficient, taken as 2), and tunnel boundary beam (corresponding to the third level coefficient, taken as 3). Let be the area of ​​the j-th opening, n be the total number of openings, and e be the Euler number. In terms of hierarchical priority, the boundary beams of the openings have the highest importance, followed by the tangent beams, and the grid beams have the lowest. The larger the opening area within the same grid, the higher the attention weight assigned to the three types of beams within that area; if there are no openings within a grid, the attention weight of the secondary beams within that grid is taken as the lowest value.

[0042] Figure 6 The diagram shows the architecture of a bidirectional interactive generative network model with multi-layer mask fusion. This model includes: Convolutional Neural Network 1 (first convolutional neural network), Convolutional Neural Network 2 (second convolutional neural network), a text encoder, a cross-branch interactive attention module, an adaptive interactive fusion unit, a residual neural network, and a deconvolutional neural network. Specifically, it includes: The potential secondary beam feature tensor, the three-layer weighted mask matrix, and the structural design parameters are input into a pre-trained bidirectional interactive generative network model. The potential secondary beam feature tensor is downsampled and feature extracted through convolutional neural network 1 to obtain an image depth feature tensor. The three-layer weighted mask matrix is ​​downsampled and feature extracted through convolutional neural network 2 to obtain a mask depth feature tensor. The structural design parameters are feature extracted through a text encoder to obtain a text embedding feature tensor. The image and mask depth feature tensors are correlated and attentionally calibrated through a cross-branch interactive attention module to obtain a bidirectional cross feature tensor. The image depth feature tensor, mask depth feature tensor, and bidirectional cross feature tensor are dynamically solved for gating coefficients and fused with feature weights through an adaptive interactive fusion unit to obtain a fused depth feature tensor. The fused feature tensor and the text embedding feature tensor are then upsampled using a residual neural network and a deconvolutional neural network to generate the opening side beam design tensor.

[0043] The adaptive interactive fusion unit determines the structural complexity of the region based on the number and area of ​​openings, and dynamically calculates the gating coefficient. Then, it uses two sets of coefficients to perform weighted fusion of image features, mask features, and bidirectional cross features, outputting a fused feature tensor. The expression is as follows: ; For image depth feature tensor, For the mask depth feature tensor, This is a bidirectional cross-feature tensor, where L and W are the transverse and longitudinal lengths of the frame, respectively. This unit emphasizes the design of the side beams when the distribution of openings is complex, while prioritizing the layout of the grid beams when the number of openings is small.

[0044] Figure 7 The diagram illustrates the training process of a generative adversarial network (GAN) model, which is divided into a generative network model and a discriminative network model. The multi-layered alienated masking-based generation network model for the opening side beam is trained based on mask matrix sample data, potential secondary beam feature tensors, and label data of the opening side beam design feature tensors. Specifically, it includes: For any given iteration, the potential secondary beam feature tensor, the three-layer weighted mask matrix, and structural design parameters are input into the bidirectional interactive generative network model, which outputs the opening side beam design tensor. The generated and target opening side beam design tensors are then input into the discriminative network model, which outputs a authenticity label. Based on the obtained authenticity labels, the loss function is calculated, and through game-theoretic optimization, the parameters of the generative and discriminative network models are adjusted to minimize the difference between the generated and target design tensors.

[0045] Figure 8 The diagram shows the dataset preprocessing process, which is divided into the construction of the input pixel image, the target pixel image, and the mask mapping image, specifically including: By scaling the engineering design drawings to a uniform scale and encoding them into image tensors of size 3×1024×1024, pixel elements containing key image information are assigned RGB values ​​based on their attributes. Pixels with attributes representing the actual arrangement of grid beams and opening edge beams are assigned red (255, 0, 0), while those representing potential secondary beam areas are assigned gray (132, 132, 132). Elements without key image information are assigned white (255, 255, 255). This results in an image pair: the input potential secondary beam layout and the target opening edge beam layout. The input image represents the potential arrangeable area, while the target image represents the actual opening edge beam layout.

[0046] The pixel positions of the grid beams, tangent beams of the opening, and boundary beams of the opening are extracted from the potential secondary beam layout diagram. Three image tensors of size 3×1024×1024 are initialized. The three image tensors are indexed and located according to the extracted pixel positions. The pixel elements at the corresponding positions are assigned a value of 1, and the remaining elements are assigned a value of zero. The mask mapping diagrams of the grid beams, tangent beams of the opening, and boundary beams of the opening are obtained in sequence.

[0047] The bidirectional interactive generative network model for the opening side beam in this application is an improvement upon the pix2pixHD model. (Generative network model) With discriminative network model The loss expressions are shown below: ; ; This represents the adversarial loss of the generator, used to constrain the generated results to approximate the real samples in terms of distribution. This represents the feature matching loss, used to align the representations of the generated image and the real image in the discriminator's intermediate feature space; For perceptual loss based on pre-trained networks; and This is the heterogeneity weighting matrix for each loss term. For the discriminator, and These represent the discrimination loss for real samples and generated samples, respectively.

[0048] This application also provides a smart arrangement system 400 for opening edge beams based on multi-layer mask fusion, the system 400 comprising: The acquisition unit 401 is used to acquire the size information of the target grid area and the opening information located in the target grid area, wherein the opening information includes the opening location and the opening size; The first arrangement unit 402 is used to generate an initial grid beam arrangement diagram based on the size information, and simulate collision between the initial grid beam arrangement diagram and the opening information to eliminate the arrangement lines in the initial grid beam arrangement diagram that collide with the opening information, so as to obtain a potential grid beam arrangement diagram. The second arrangement unit 403 is used to generate a potential hole side beam arrangement diagram based on the hole location and the hole size; The third arrangement unit 404 is used to splice the potential grid beam arrangement diagram with the potential hole edge beam arrangement diagram to obtain the potential secondary beam arrangement diagram. Processing unit 405 is used to perform pixelation processing on the potential secondary beam layout diagram to obtain the potential secondary beam feature tensor. The fusion unit 406 is used to perform matrix encoding and weight fusion on the potential secondary beam feature tensor to generate a three-layer weighted mask matrix. The generation unit 407 is used to input the potential secondary beam feature tensor, the three-layer weighted mask matrix and the preset structural design parameters into the pre-trained bidirectional interactive generative network model to generate the design drawing of the opening side beam; The structural design parameters are the seismic intensity and structural design height of the target building.

[0049] In summary, this application designs an automated layout module for potential grid beams and opening edge beams. It directly reads the opening and frame information from structured data to autonomously plan the beam layout area (target grid area). Its layout logic aligns with engineering design principles, solving the problem that existing generative design methods rely on the conversion of complex architectural drawings to determine the pre-defined beam layout area and are subject to form limitations. The potential grid beams and opening edge beams are represented in a matrix format and hierarchically differentiated to construct a multi-layered weighted mask matrix. The introduction of prior location information forms spatial constraints, and weights enhance the focus on key beam features, improving the accuracy and controllability of the design scheme. The generated network... The network model is designed with a cross-branch coding interaction module and an adaptive fusion unit to construct a dual-branch interaction architecture of image geometric features and engineering mask constraint features. Through the cross-branch cross-attention mechanism and the adaptive fusion unit, the deep association of the two types of heterogeneous features is realized. The features of the two branches are dynamically adjusted according to the structural complexity of the openings in the region, so as to realize the adaptive control of the feature fusion strategy. When the distribution of openings is complex, the design of the side beams of the openings is emphasized, while when the number of openings is small, the layout of the grid beams is taken into consideration. This can improve the generation quality and stability, thereby solving the problem of low efficiency in the arrangement of side beams of openings in existing steel frame structures and realizing efficient and reliable intelligent structural design.

[0050] The above is a detailed description of the embodiments of this application. However, this application is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this application. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.

Claims

1. A method for intelligent arrangement of edge beams at openings based on multi-layer mask fusion, characterized in that, The method includes: Obtain the size information of the target grid area and the opening information located within the target grid area, wherein the opening information includes the opening location and opening size; An initial grid beam layout diagram is generated based on the size information, and the initial grid beam layout diagram is simulated to collide with the opening information to eliminate the layout lines in the initial grid beam layout diagram that collide with the opening information, thereby obtaining a potential grid beam layout diagram. A potential layout diagram of the side beams of the opening is generated based on the location and size of the opening. By splicing the potential grid beam layout diagram with the potential hole edge beam layout diagram, a potential secondary beam layout diagram is obtained. The potential secondary beam layout diagram is pixelated to obtain the potential secondary beam feature tensor; The potential secondary beam feature tensor is matrix-encoded and weighted to generate a three-layer weighted mask matrix. The potential secondary beam feature tensor, the three-layer weighted mask matrix, and the preset structural design parameters are input into a pre-trained bidirectional interactive generative network model to generate the design drawing of the opening side beam. The structural design parameters refer to the seismic intensity structural design height of the target building.

2. The intelligent arrangement method for opening edge beams based on multi-layer mask fusion according to claim 1, characterized in that, The step of simulating the collision between the initial grid beam layout and the opening information to obtain a potential grid beam layout includes: Extract each beam layout line from the initial grid beam layout diagram, where each beam layout line is composed of the coordinates of the nodes at both ends; Based on the location of the opening, determine the location of each sub-opening from the opening information; Based on the opening dimensions, determine the diameter of each of the sub-openings; For each sub-opening, the boundary line of the opening is determined according to the position and diameter of the sub-opening. If the boundary line of the opening intersects with any of the beam arrangement lines, the beam arrangement lines that intersect with the boundary line of the opening are discarded. The beam layout lines that were not removed are retained to obtain the potential grid beam layout diagram.

3. The intelligent arrangement method for opening edge beams based on multi-layer mask fusion according to claim 2, characterized in that, The step of generating a potential side beam layout diagram based on the location and size of the opening includes: The boundaries of each sub-opening are determined based on the location of each sub-opening in the opening location and the diameter of each sub-opening in the opening size; Arrange the edge beams at the boundaries of each of the sub-openings to generate the potential edge beam arrangement diagram.

4. The intelligent arrangement method for opening edge beams based on multi-layer mask fusion according to claim 3, characterized in that, After obtaining the potential secondary beam layout, the method further includes: Extract tangent information from the potential secondary beam layout diagram to characterize the tangential connection relationship between the hole-edge beam and the grid beam; A potential hole tangent beam layout diagram is generated based on the tangent information, and the position of the hole tangent beam is determined based on the potential hole tangent beam layout diagram.

5. The intelligent arrangement method for opening edge beams based on multi-layer mask fusion according to claim 4, characterized in that, The pixelation process of the potential secondary beam layout diagram to obtain the potential secondary beam feature tensor includes: Determine the boundary lines in the potential secondary beam layout diagram; Based on a preset length-pixel conversion ratio, each boundary line of the potential grid beam layout, each boundary line of the potential hole edge beam layout, and each boundary line of the potential hole tangent beam layout are pixelated and encoded to obtain the potential secondary beam feature tensor.

6. The intelligent arrangement method for opening edge beams based on multi-layer mask fusion according to claim 5, characterized in that, The step of matrix encoding and weight fusion of the potential secondary beam feature tensor to generate a three-layer weighted mask matrix includes: Initialize a mask matrix with all three elements equal to zero; The pixel positions of the grid beam, the hole tangent beam, and the hole edge beam are extracted from the potential secondary beam feature tensor, respectively. Based on the pixel positions of the grid beams, the hole tangent beams, and the hole edge beams, pixel elements of the three mask matrices, each with zero elements, are assigned values ​​to obtain the grid beam mask matrix, the hole tangent beam mask matrix, and the hole edge beam mask matrix. The initial value of each pixel in the grid beam mask matrix, the hole tangent beam mask matrix, and the hole edge beam mask matrix is ​​set to 1. The grid beam mask matrix, the hole tangent beam mask matrix, and the hole boundary beam mask matrix are weighted respectively to obtain the three-layer weighted mask matrix.

7. The intelligent arrangement method for opening edge beams based on multi-layer mask fusion according to claim 6, characterized in that, The grid beam mask matrix, the hole tangent beam mask matrix, and the hole boundary beam mask matrix are weighted respectively to obtain the three-layer weighted mask matrix, including: Determine the area of ​​each sub-entrance within the target grid region and the total number of each sub-entrance; For each sub-hole, the area of ​​the sub-hole is divided by the sum of the areas of all the sub-holes to obtain the area ratio of the sub-hole among all the sub-holes, and the weighting coefficient of the sub-hole is determined based on the area ratio. Determine the first-level coefficients corresponding to the grid beam mask matrix, determine the first pixel value corresponding to each sub-opening based on the first-level coefficients, the initial pixel value in each sub-opening, and the weighting coefficients, and generate the final grid beam mask matrix based on the first pixel value corresponding to each sub-opening, wherein each sub-opening corresponds to one or more first pixel values; Determine the second-level coefficients corresponding to the hole tangent beam mask matrix, determine the second pixel value corresponding to each of the sub-holes based on the second-level coefficients, the initial pixel value in each of the sub-holes and the weighting coefficients, and generate the final hole tangent beam mask matrix based on the second pixel value corresponding to each of the sub-holes, wherein each sub-hole corresponds to one or more second pixel values; The third-level coefficients corresponding to the hole boundary beam mask matrix are determined. The third pixel value corresponding to each sub-hole is determined based on the third-level coefficients, the initial pixel value in each sub-hole, and the weighting coefficient. The final hole boundary beam mask matrix is ​​generated based on the third pixel value corresponding to each sub-hole. Each sub-hole corresponds to one or more third pixel values. The final grid beam mask matrix, the final hole tangent beam mask matrix, and the final hole boundary beam mask matrix are used as the three-layer weighted mask matrix; In this context, the pixel values ​​of non-hole areas within the target grid region are all maintained at 1.

8. The intelligent arrangement method for opening edge beams based on multi-layer mask fusion according to claim 1, characterized in that, The pre-trained bidirectional interactive generative network model includes a first convolutional neural network, a second convolutional neural network, a text encoder, a cross-branch interactive attention module, an adaptive interactive fusion unit, a residual neural network, and a deconvolutional neural network.

9. The intelligent arrangement method for opening edge beams based on multi-layer mask fusion according to claim 8, characterized in that, The step of inputting the potential secondary beam feature tensor, the three-layer weighted mask matrix, and preset structural design parameters into a pre-trained bidirectional interactive generative network model to generate a design drawing of the opening side beam includes: The image depth feature tensor is obtained by downsampling and feature extraction of the potential secondary beam feature tensor through the first convolutional neural network. The mask depth feature tensor is obtained by downsampling and feature extraction of the three-layer weighted mask matrix through the second convolutional neural network. The text encoder extracts features from the structural design parameters to obtain a text embedding feature tensor. The cross-branch interactive attention module performs correlation attention calculation and bidirectional feature calibration on the image depth feature tensor and the mask depth feature tensor to obtain a bidirectional cross feature tensor. The adaptive interactive fusion unit dynamically solves the gating coefficients and performs feature weighted fusion on the image depth feature tensor, the mask depth feature tensor, and the bidirectional cross feature tensor to obtain the fused depth feature tensor. The residual neural network and the deconvolutional neural network are used to extract and upsample the fused deep feature tensor and the text embedding feature tensor to generate the opening side beam design tensor, so as to generate the opening side beam design drawing based on the opening side beam design tensor.

10. A smart arrangement system for opening edge beams based on multi-layer mask fusion, characterized in that, The system includes: The acquisition unit is used to acquire the size information of the target grid area and the opening information located within the target grid area, wherein the opening information includes the opening location and the opening size; The first arrangement unit is used to generate an initial grid beam arrangement diagram based on the size information, and simulate collision between the initial grid beam arrangement diagram and the opening information to eliminate the arrangement lines in the initial grid beam arrangement diagram that collide with the opening information, so as to obtain a potential grid beam arrangement diagram. The second arrangement unit is used to generate a potential hole side beam arrangement diagram based on the hole location and the hole size; The third arrangement unit is used to splice the potential grid beam arrangement diagram with the potential hole edge beam arrangement diagram to obtain the potential secondary beam arrangement diagram. The processing unit is used to pixelate the potential secondary beam layout diagram to obtain the potential secondary beam feature tensor. The fusion unit is used to perform matrix encoding and weight fusion on the potential secondary beam feature tensor to generate a three-layer weighted mask matrix; The generation unit is used to input the potential secondary beam feature tensor, the three-layer weighted mask matrix and the preset structural design parameters into the pre-trained bidirectional interactive generative network model to generate the design drawing of the opening side beam; The structural design parameters are the seismic intensity and structural design height of the target building.

Images