A bim-based fabricated building component design method

Abstract

This invention relates to the field of digital design for prefabricated buildings, and more particularly to a BIM-based design method for prefabricated building components. The method includes: extracting the three-dimensional non-convex surface boundaries and acoustic and fluid parameters of the components; defining them as two-dimensional Riemannian manifolds; minimizing the metric distortion functional to obtain a two-dimensional parameter plane and an analytical inverse mapping function; constructing a neural network with an output level set function on this plane; fusing acoustic and fluid residual terms to construct and minimize the total loss function; generating a two-dimensional topology network based on the zero isosurface of the optimal level set function; calling the inverse mapping function to pull the network back into three-dimensional space; generating a three-dimensional solid negative space containing fractal flow channels and resonant blind ends through Boolean subtraction; and finally performing distortion verification through the Gaussian curvature change rate. If the distortion exceeds the limit, the penalty multiplier is increased for closed-loop iteration; otherwise, the final building information model is output. This invention resolves the contradiction of internal flow-sound interference within the components and avoids dimensional collapse during three-dimensional reconstruction.

Description

Technical Field

[0001] This invention relates to the field of digital design of prefabricated buildings, and more particularly to a BIM-based design method for prefabricated building components. Background Technology

[0002] In the construction and service of prefabricated buildings in complex application scenarios such as urban rail transit or high-speed rail hubs, irregularly shaped prefabricated load-bearing components with non-convex curved surfaces not only need to inject non-Newtonian fluid slurry into the internal concealed cavity to achieve high-strength anchoring of the nodes, but also need to utilize the internal reserved cavity to attenuate the specific low-frequency vibration noise generated by train operation.

[0003] However, traditional irregular components mostly adopt a dimensional reduction design with isolated multi-field separation. Due to the difficulty in balancing exhaust and sound insulation and the tendency to produce metric distortion, the physical functions of internal nodes fail. Summary of the Invention

[0004] To overcome the above shortcomings, this invention provides a BIM-based prefabricated building component design method, which aims to improve the traditional dimensionality reduction design of irregularly shaped components, which mostly adopts isolated and fragmented multi-field designs. Due to the difficulty in balancing ventilation and sound insulation and the easy generation of metric distortion, the physical functions of internal nodes fail.

[0005] This invention provides the following technical solution: a BIM-based design method for prefabricated building components, comprising the following steps: S1. Extract the three-dimensional non-convex surface boundary, acoustic spectrum parameters and fluid grouting parameters of the prefabricated components; S2. Define the boundary of the three-dimensional non-convex surface as a two-dimensional Riemannian manifold and construct a mapping function. Minimize the local metric distortion functional to obtain the two-dimensional parametric plane and the analytical inverse mapping function. S3. Construct a neural network for outputting a level set function in the two-dimensional parameter plane, construct acoustic residual terms and fluid residual terms based on the acoustic spectrum parameters and fluid grouting parameters, and embed them into the neural network to construct the total loss function; S4. Assign weights to the acoustic residual term and the fluid residual term and minimize the total loss function to obtain the optimal level set function, and generate a two-dimensional topology network based on the zero isosurface of the optimal level set function. S5. Call the analytical inverse mapping function to pull the two-dimensional topology network back into the three-dimensional physical space, and generate a three-dimensional solid negative space containing fractal flow channels and resonant blind ends in the prefabricated component through Boolean subtraction. S6. Calculate the rate of change of Gaussian curvature of the three-dimensional entity negative space. If it is greater than the preset tolerance, increase the penalty multiplier of the local metric distortion functional and return to the steps of constructing the mapping function and solving the two-dimensional parameter plane. Otherwise, output the building information model containing the three-dimensional entity negative space.

[0006] By adopting the above technical solution, Riemannian manifold mapping and fusion of flow acoustic residuals are used for neural network optimization, thereby realizing the isomorphic generation of internal flow channels and resonant blind ends. This improves the problem that traditional irregular components mostly adopt isolated multi-field fragmented dimensionality reduction design, which is difficult to balance exhaust and sound insulation and is prone to metric distortion, thus causing the failure of the physical function of internal nodes.

[0007] The present invention has the following beneficial effects: 1. In this invention, the internal flow channel and the resonant blind end are generated isomorphically by Riemannian manifold mapping and fusion of flow acoustic residuals for neural network optimization. This improves the problem that traditional irregular components mostly adopt isolated multi-field fragmented dimensionality reduction design, which is difficult to balance exhaust and sound insulation and is prone to metric distortion, thus causing the failure of the physical function of internal nodes.

[0008] 2. In this invention, by assigning dominant and complementary weights to the fluid and acoustic residual terms, gradient conflicts during backpropagation of multi-physics fields are eliminated, thereby improving the problem that traditional multi-field joint solutions mostly use fixed average weights, which cause the model to diverge easily and have difficulty converging to the optimal solution due to the large differences in the magnitude of the residuals of each physics field.

[0009] 3. In this invention, by calculating the rate of change of Gaussian curvature and triggering the closed-loop iteration of the penalty multiplier, the accuracy of the solid size after reverse restoration is ensured. This improves the problem that traditional surface unfolding mostly adopts static open-loop mapping, which cannot correct the microscopic deformation caused by spatial dimension transformation, thus causing geometric collapse of the internal fine structure.

[0010] 4. In this invention, by constraining the neck size of the flow channel transition to the blind end to meet the fluid yield condition, a physical one-way valve utilizing the phase change of the slurry is constructed, thereby improving the problem that traditional interconnection structures mostly use conventional pipes, which lack a microscopic cutoff mechanism for rheological characteristics, resulting in grouting fluid easily overflowing and permanently damaging the acoustic cavity. Attached Figure Description

[0011] Figure 1 This is a flowchart of a BIM-based prefabricated building component design method proposed in this invention; Figure 2 This invention presents a flowchart of the multiphysics neural network construction and two-dimensional topology generation process for a BIM-based prefabricated building component design method. Figure 3 This is a flowchart illustrating the closed-loop process of three-dimensional solid negative space reconstruction and curvature distortion verification for a BIM-based prefabricated building component design method proposed in this invention. Detailed Implementation

[0012] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] Example 1: In a first embodiment of the present invention, the present invention provides a BIM-based design method for prefabricated building components, such as... Figure 1 As shown, it includes the following steps: S1. Extract the three-dimensional non-convex surface boundary, acoustic spectrum parameters and fluid grouting parameters of the prefabricated components; Furthermore, in S1, the extraction of the three-dimensional non-convex surface boundary, acoustic spectrum parameters, and fluid grouting parameters of the prefabricated components includes: Analyze the geometric data of the building information model of prefabricated components and extract the surface feature point array to reconstruct the three-dimensional non-convex surface boundary; Search the acoustic specification database of the target environment and extract the peak frequency sequence of the low-frequency noise to be attenuated as the acoustic spectrum parameter; Extract the pumping pressure threshold of the on-site grouting equipment and retrieve the fluid yield stress and dynamic viscosity properties from the grouting material database as fluid grouting parameters.

[0014] Specifically, the data input and output process is as follows: the system receives geometric data of the building information model of prefabricated components, acoustic specification database of the target environment, on-site grouting equipment control system, and grouting material database as initial data input sources; after feature point reconstruction and parameter retrieval calculation by the back-end parsing module, it outputs three-dimensional non-convex surface boundary, acoustic spectrum parameters, and fluid grouting parameters to the downstream core processing module, thereby establishing the absolute physical boundary conditions for subsequent multi-physics joint solution, forming a closed loop of data acquisition and boundary initialization.

[0015] When establishing the underlying physical mapping of fluid grouting parameters, a non-Newtonian fluid Bingham constitutive mathematical model must be introduced to characterize the true rheological properties of the grouting material. The specific rheological control formulas are set as follows: ; in This represents the shear stress experienced by the grouting fluid when pumped within a narrow cavity. This represents the fluid yield stress in the fluid grouting parameters. This represents the dynamic viscosity in the fluid grouting parameters. This represents the shear rate of the grouting fluid flowing within the cavity channel. For high-strength, non-shrinkage sleeve grouting materials specifically designed for prefabricated buildings, the fluid yield stress... A specific value must be set, and the value setting range is 20Pa to 60Pa.

[0016] When establishing physical constraints on acoustic spectrum parameters, the discrete constraint space formula for the acoustic target frequency is defined as follows: ; in This represents any single peak frequency element in the low-frequency noise peak frequency sequence to be attenuated in the acoustic spectral parameters. This indicates the lower limit of the test frequency for low-frequency noise in the target environment. This indicates the upper limit of the target environment's background low-frequency noise frequency for testing. For background noise from urban rail transit vibrations and building floor impacts, the constraint value for the peak frequency sequence of the low-frequency noise to be attenuated is set between 50Hz and 200Hz.

[0017] By analyzing the geometric data of the building information model of prefabricated components to extract surface feature point arrays and reconstruct three-dimensional non-convex surface boundaries, and by retrieving the low-frequency noise peak frequency sequence to be attenuated from the acoustic specification database, and by retrieving the fluid yield stress and dynamic viscosity properties from the grouting material database, this method improves the technical defects of traditional prefabricated component bottom-level design, which relies only on static ideal Euclidean geometric contours and has isolated and fragmented multi-physics boundary conditions in the initial model stage. This avoids the physical failure problem of the generated metamaterial structure being unable to accurately align and attenuate the real low-frequency noise peaks during the subsequent evolution of neural networks and topological networks due to the lack of real non-Newtonian rheological boundaries and strong constraints in the target frequency domain.

[0018] S2. Define the boundary of the three-dimensional non-convex surface as a two-dimensional Riemannian manifold and construct a mapping function. Minimize the local metric distortion functional to obtain the two-dimensional parametric plane and the analytical inverse mapping function. Furthermore, in S2, the boundary of the three-dimensional non-convex surface is defined as a two-dimensional Riemannian manifold, and a mapping function is constructed. Minimizing the local metric distortion functional yields the two-dimensional parametric plane and the analytical inverse mapping function, including: Introducing the first fundamental form containing the metric tensor, the discrete coordinates of the boundary of a three-dimensional non-convex surface are parameterized into a two-dimensional Riemannian manifold; A local metric distortion functional is established with the Jacobian matrix of the mapping function as the variable to measure the distortion rate of the linear transformation of the tangent space; The conjugate gradient optimization algorithm is used to iteratively minimize the local metric distortion functional until the distortion rate converges to output a two-dimensional parameter plane. Record the topological correlation matrix of the grid node mapping when the minimization solution converges, and generate the analytical inverse mapping function based on the topological correlation matrix of the grid node mapping.

[0019] Specifically, the data input / output process involves the system extracting the discrete coordinates of the 3D non-convex surface boundary as the front-end input parameters. After the underlying spatial mapping module performs manifold parameterization and functional iterative minimization, it outputs a 2D parameter plane to the downstream physical information neural network module, and separately extracts the analytical inverse mapping function and outputs it to the subsequent 3D spatial entity Boolean reduction set reconstruction module.

[0020] To address the physical characteristic that it is difficult to achieve lossless dimensionality reduction of the boundary of a three-dimensional non-convex surface, the governing equations of the local metric distortion functional are constructed as follows: ; in This represents the functional of local metric distortion. Represents the integral field of a two-dimensional Riemannian manifold. The Jacobian matrix representing the mapping function, The matrix representing the transpose of the Jacobian matrix. Represents a second-order identity matrix, with subscripts. The double vertical line symbol represents the Frobenius norm. This represents the area measure element on a two-dimensional Riemannian manifold.

[0021] By introducing the first fundamental form, the discrete coordinates of a non-convex surface are transformed into a two-dimensional Riemannian manifold, establishing a foundation for the transformation from a three-dimensional physical surface to a non-Euclidean geometric space. A local metric distortion functional is established using the Jacobian matrix of the mapping function to measure the distortion rate of the linear transformation of the tangent space. The conjugate gradient method is used to iteratively solve for the distortion rate until convergence, and the topological correlation matrix of the mesh node mapping at this convergence state is captured and recorded. Subsequently, the two-dimensional parameter nodes in the topological correlation matrix of the mesh node mapping are extracted as control base points, and the target three-dimensional physical coordinates are used as the fitting target values. The thin-plate spline interpolation algorithm is called to solve for the smoothing weight coefficients of the radial basis function, thereby constructing a continuously differentiable spatial mapping polynomial as the corresponding analytical inverse mapping function.

[0022] This technique abandons the unconditional absolute distance-preserving assumption relied upon by traditional building information model parametric unfolding. It performs local distance-preserving conformal mapping on non-convex surfaces with large curvature, eliminating the severe distortion and warping of geometric topology caused by conventional direct dimensionality reduction. It avoids the fatal problem of structural deformation caused by pre-defined microscopic flow-limiting necks and acoustic resonance blind ends after inverse mapping back to 3D space, leading to physical failures in fluid dynamics and acoustic properties. The 2D parameter plane output from the solution establishes a geometrically distortion-free mathematical foundation for subsequent multi-field coupled neural network solutions, while the analytical inverse mapping function ensures accurate 3D coordinate inversion from the abstract mathematical topology network to the negative space of the real engineering entity.

[0023] like Figure 2 As shown in Figure S3, a neural network for outputting the level set function is constructed in the two-dimensional parameter plane. Acoustic residual terms and fluid residual terms are constructed based on acoustic spectrum parameters and fluid grouting parameters, and embedded into the neural network to construct the total loss function. Furthermore, in S3, constructing a neural network for the output level set function in a two-dimensional parameter plane includes: Initialize a multi-layer fully connected deep neural network structure in a two-dimensional parametric plane; The input layer nodes of a multi-layer fully connected deep neural network structure receive spatial coordinate sequences and temporal variables from a two-dimensional parameter plane. Configure the output layer nodes of a multi-layer fully connected deep neural network structure to output the level set function; The entity distribution region and the internal cavity region are defined based on the positive and negative signs of the output value of the level set function, and the zero isosurface of the level set function is extracted as the physical boundary of the evolution.

[0024] In S3, the acoustic residuals and fluid residuals are constructed based on acoustic spectrum parameters and fluid grouting parameters, including: Using the fluid yield stress and dynamic viscosity in the fluid grouting parameters as physical property inputs, a Navier-Stokes equation containing the Cauchy stress tensor is constructed. Discrete fluid placement points are arranged in a two-dimensional parameter plane, and the Navier-Stokes equations are transformed into a summation of squared errors for discrete fluid placement points to generate fluid residual terms. By combining the low-frequency noise peak frequency in the acoustic spectrum parameters with the preset air static density, a Helmholtz acoustic steady-state partial differential equation is constructed. Discrete acoustic placement points are arranged in a two-dimensional parametric plane, and the Helmholtz acoustic steady-state partial differential equation is transformed into a summation of squared errors for the discrete acoustic placement points to generate acoustic residual terms.

[0025] Specifically, the data input and output process is as follows: the system receives the spatial coordinate sequence of the two-dimensional parameter plane, time variables, fluid grouting parameters, and acoustic spectrum parameters as feature input sources for the underlying operation; through the forward propagation of the multi-layer fully connected deep neural network and the residual calculation of the physical equation, it outputs the total loss function containing multi-physics field features to the subsequent weight allocation and minimization training module, and outputs the level set function used to characterize the three-dimensional spatial evolution boundary.

[0026] Initialize a multi-layer fully connected deep neural network structure in a two-dimensional parametric plane. Set the network input layer to receive a sequence of spatial coordinate vectors. With time variable The output layer generates a level set function. The regulations stipulate that when When the area is designated as the distribution area of ​​entities, The internal cavity region was defined at that time, and the extraction was performed. The isosurface serves as the physical boundary for dynamic evolution. This operation transforms the complex geometric and topological changes within prefabricated components into a process of solving for the zeros of continuous mathematical functions, avoiding the computational divergence caused by repeated meshing in traditional finite element analysis.

[0027] Substituting the fluid yield stress and dynamic viscosity into the Navier-Stokes equations, which include the Cauchy stress tensor, yields the fluid dynamics residual term. The control formula is constructed as follows: ; in This represents the total number of discrete fluid placement points arranged in the two-dimensional parametric plane. Indicates the density of the grouting fluid. Indicates the first A discrete fluid configuration point at time... The fluid velocity field vector, This represents the Cauchy stress tensor, which incorporates characteristics of fluid yield stress. This represents the gravitational acceleration vector. This operation transforms the non-Newtonian hydrodynamic partial differential equations into a summation of squared errors based on discrete spatial points, establishing the physical constraint boundaries for the pumping and venting process of high-pressure grouting fluid within complex cavities.

[0028] Combining the peak frequency of low-frequency noise with the static density of air, the acoustic frequency domain residual term The control formula is constructed as follows: ; in This represents the total number of discrete acoustic placement points arranged in the two-dimensional parametric plane. This indicates the preset static air density, specifically set to 1.225 kg per cubic meter. Indicates the first Sound pressure variables at discrete acoustic placement points This represents the target low-frequency angular frequency derived from the peak frequency of low-frequency noise. This represents the speed of sound in the air, specifically set to 340 meters per second. This step maps the Helmholtz acoustic steady-state partial differential equation into an error function for discrete points, providing a quantitative criterion for the neural network to evaluate the low-frequency sound wave reflection and attenuation performance of the cavity structure.

[0029] When arranging discrete fluid and discrete acoustic placement points, a Latin hypercube sampling strategy is employed to perform global initial point distribution within the two-dimensional parameter plane, ensuring the multidimensional uniformity of the sampling space. During the minimization training of the neural network, an adaptive resampling mechanism based on the residuals of partial differential equations is introduced. The global residual distribution is calculated in real time, and in local high-gradient regions where the absolute value of the residual is greater than a specific multiple of the overall residual average, high-density dynamic placement points are added according to a normal distribution, ensuring the strict convergence of the multiphysics coupling equations of the flow acoustics at the boundaries and in regions with severe phase transitions.

[0030] The fluid dynamics residuals and acoustic frequency domain residuals are directly embedded into the neural network as soft constraints. To ensure that the generated topological boundary satisfies the physical structure of the subsequent unidirectional fluid valve, a geometric inequality penalty term derived from the critical condition of fluid shear yield resistance is further introduced when constructing the total loss function; that is, during the evolution of the zero isosurface of the level set function, the normal cross-sectional radius of the transition region at the bifurcation end is constrained in real time. satisfy If the evolution size exceeds this critical condition, an exponentially increasing geometric penalty is applied to the total loss function, forcing the neural network to directly generate blind-end neck features that meet the physical truncation requirements. The loss function constructed from the residuals of partial differential equations can directly drive the zero isosurface of the level set function to perform spatial optimization evolution, enabling the generated geometric boundary to adaptively accommodate the two contradictory physical requirements of low-resistance non-Newtonian fluid exhaust and high-impedance low-frequency sound wave reflection. This solves the core technical problem that the hidden cavities inside complex irregular prefabricated components cannot be designed isomorphically across physical fields.

[0031] like Figure 2 As shown, S4 assigns weights to the acoustic residual term and the fluid residual term and minimizes the total loss function to obtain the optimal level set function, and generates a two-dimensional topology network based on the zero isosurface of the optimal level set function. Furthermore, in S4, weights are assigned to the acoustic and fluid residual terms, and the total loss function is minimized to obtain the optimal level set function, which includes: The irrational constant is extracted as the dominance coefficient, and the dominance coefficient is multiplied into the fluid residual term; The difference between one and the irrational constant is used as the complementary weight coefficient, and the complementary weight coefficient is multiplied into the acoustic residual term; The boundary loss term is fitted by combining the data, and the weighted fluid residual term and acoustic residual term are combined to construct the total loss function; A hybrid optimization algorithm is used to minimize the total loss function during training to update the neural network weights, and the optimal level set function is captured when the loss convergence threshold is met.

[0032] In S4, generating a two-dimensional topology network based on the zero isosurface of the optimal level set function includes: Discretize the zero isosurface continuous boundary of the optimal level set function in a two-dimensional parametric plane; The internal negative space region of the closed envelope of the continuous boundary of the zero isosurface is defined as a two-dimensional topological network; Identify and label fractal guiding topology segments with trunk-tributary connection characteristics and blind-end resonant topology segments with gradually narrowing flow-limiting characteristics in two-dimensional topology networks.

[0033] Specifically, the input and output data streams are as follows: the system receives the fluid residual term and acoustic residual term generated in the previous stage, and extracts the built-in irrational constant as the initial value for weight allocation; after processing by the weight allocation module and the neural network minimization training module, the system outputs the optimal level set function; this function is further used as the input of the topology extraction module, and finally outputs a two-dimensional topology network that defines the boundary of the three-dimensional reconstruction, including the specific spatial markings of the fractal flow guiding topology segment and the blind end resonance topology segment.

[0034] When assigning weights to the loss function, an exact irrational constant is introduced for asymmetric weighting. Total loss function The construction formula is set as follows: ; in This represents the total loss function that the neural network needs to minimize during backpropagation. This represents the extracted irrational constant, and the specific value of this constant is... As the dominance coefficient Represents the fluid residual term. Let the difference between 1 and an irrational constant be used as the complementary weight coefficient. Represents the acoustic residual term. This represents the boundary loss term for data fitting. For the physical hard boundary constraints of prefabricated building components, Dirichlet boundary conditions are used to construct the boundary loss term for data fitting. Its expansion formula is as follows: ; in This represents the total number of discrete boundary constraint points collected on the fixed boundary of the two-dimensional parametric plane. This represents the index of the corresponding discrete boundary constraint point. This represents the current node boundary prediction value output by the neural network. This represents the actual geometric boundary constraint value initially set based on the boundary of the three-dimensional non-convex surface.

[0035] In the weighting stage of the loss function, irrational numbers are used. The mechanism of asymmetric weighting is that the magnitude of the fluid phase change gradient usually decays exponentially, while the acoustic steady-state frequency domain residual oscillates periodically. By utilizing the incommensurability of this irrational constant, the symmetry stagnation of the loss surface during multi-physics backpropagation can be effectively broken, avoiding the neural network from getting trapped in a local optimum saddle point, thus eliminating gradient conflict and mode collapse from the underlying mathematical logic.

[0036] A hybrid optimization algorithm is used to perform minimization training on the constructed total loss function to continuously update the inter-layer weights of the neural network. The specific minimization training process is as follows: First, an adaptive moment estimation optimizer is called to perform global preliminary optimization training for the first-order gradient; the descending gradient of the total loss function is monitored, and when the descending gradient is below a preset smoothing threshold for ten consecutive training epochs, an algorithm switching command is triggered, switching the optimizer to a finite-memory quasi-Newton optimizer for local precise convergence training of the second-order gradient. When the value of the total loss function decreases to finally satisfy the set absolute loss convergence threshold, training stops and the optimal level set function of the current output is captured. Subsequently, a discretization scan is performed in the two-dimensional parameter plane to extract the continuous isosurface boundary where the optimal level set function value is zero, and the internal negative space region formed by the closed envelope of this continuous boundary is defined as a two-dimensional topological network.

[0037] A geometric feature recognition algorithm is executed in the generated two-dimensional topological network. Specifically, the centerline transformation algorithm is used to extract the trajectory of the center of the largest inscribed circle of the two-dimensional topological network to generate a topological skeleton. Bifurcation skeleton branches with more than a preset level of intersecting nodes are identified on the topological skeleton, and the regions enveloping these branches are marked as fractal flow-guiding topological segments. The normal cross-sectional width of the geometric profile is calculated along the end path of the fractal flow-guiding topological segment. When the contraction rate of the normal cross-sectional width is greater than a set neck flow-limiting threshold, and a closed cavity region is connected behind it, the region is marked as a blind-end resonant topological segment. This step transforms the joint optimization results of complex fluid phase transitions and acoustic steady-state frequency domain partial differential equations into specific two-dimensional spatial geometric boundaries with clear physical functional attributes. This provides accurate boundary contour positioning and topological feature classification benchmarks for subsequently accurately pulling the mathematical topological network back into three-dimensional physical space and reconstructing the entity negative space subtraction set, eliminating functional conflicts caused by geometric interference in the generation of multi-physics structures of irregular components.

[0038] like Figure 3As shown, S5 calls the parsing inverse mapping function to pull the two-dimensional topology network back into the three-dimensional physical space, and generates a three-dimensional solid negative space containing fractal flow channels and resonant blind ends in the prefabricated component through Boolean subtraction. Furthermore, in S5, the analytical inverse mapping function is called to pull the two-dimensional topology network back into the three-dimensional physical space. Through Boolean subtraction, a three-dimensional solid negative space containing fractal flow channels and resonant blind ends is generated within the prefabricated component, including: Boundary discretization sampling is performed in a two-dimensional topology network to obtain a two-dimensional discrete coordinate array; The two-dimensional discrete coordinate array is input into the analytical inverse mapping function, and the transformation is calculated to obtain the three-dimensional physical coordinate array. The boundary of a three-dimensional spatial curved surface entity is reconstructed based on a three-dimensional physical coordinate array; In the solid model of prefabricated components, Boolean subtraction operations are performed on the boundary of the three-dimensional spatial curved surface solid to remove the corresponding volume and generate a three-dimensional solid negative space.

[0039] In S5, after performing a Boolean subtraction operation on the boundary of a 3D spatial surface solid to remove the corresponding volume and generate the negative space of the 3D solid, the following steps are also included: The region generated by pulling back the fractal flow guide topology segment in the negative space of the three-dimensional solid is defined as the fractal flow channel of the main exhaust pumping channel. The region generated by pulling back the blind-end resonant topological segment in the negative space of the three-dimensional solid is defined as the resonant blind end connected to the end of the fractal flow channel; The tapered neck dimension of the constrained fractal flow channel transitioning to the resonant blind end satisfies the critical condition of fluid shear yielding retardation, thus forming a unidirectional fluid valve geometry.

[0040] Specifically, the data input / output process involves the system extracting the two-dimensional topological network and the previously generated analytical inverse mapping function as the front-end input source. After boundary discretization sampling and inverse mapping transformation operations, a three-dimensional physical coordinate array is output to the three-dimensional reconstruction module, thereby generating the three-dimensional spatial curved surface entity boundary. The underlying modeling engine uses the three-dimensional spatial curved surface entity boundary as a basis to output Boolean reduction operation instructions to the entity model of the prefabricated component, ultimately outputting a three-dimensional entity negative space digital twin model with clearly defined physical functions.

[0041] Within a 2D topological network, boundary discrete sampling is performed to extract continuous nodes on the closed contour to obtain a 2D discrete coordinate array. This 2D discrete coordinate array is then substituted into an analytical inverse mapping function, and a 3D physical coordinate array is calculated and reconstructed according to manifold metric mapping rules. The system uses this 3D physical coordinate array to fit and generate a closed 3D spatial surface entity boundary in 3D space. Using this 3D spatial surface entity boundary as the spatial partitioning interface, a Boolean subtraction operation is directly performed within the initial solid model of the prefabricated component to remove the corresponding enclosed volume, generating a continuous and irregular 3D solid negative space with an inner surface.

[0042] The physical functional areas of the three-dimensional solid negative space are defined and constrained. The region generated by mapping and restoring the fractal flow guide topology segment in the three-dimensional solid negative space is defined as the fractal flow channel in the form of the main exhaust pumping channel, which bears the fluid transport and exhaust tasks under high pressure. The region generated by mapping and restoring the blind end resonant topology segment is defined as the resonant blind end connected to the end of the fractal flow channel, which is reserved as a sound wave reflection cavity during service.

[0043] To prevent grout from overflowing into the resonant blind end and damaging the acoustic cavity, the necking dimension of the constricted fractal flow channel transitioning to the resonant blind end must strictly satisfy the critical condition of fluid shear yield resistance. The critical control formula for the necking dimension is established as follows: ; in Indicates the radius of the tapered neck section. Indicates the effective flow length of the constricted neck. This represents the fluid yield stress of the grouting material. This indicates the local pumping pressure difference exerted on the neck by the on-site grouting equipment. For high-strength sleeve grouting material specifically designed for prefabricated joints, the specific range of fluid yield stress is fixed at 20 Pa to 60 Pa.

[0044] Calling the analytic inverse mapping function and performing Boolean reduction operations bridges the technical gap in converting abstract two-dimensional mathematical topology to real three-dimensional solid models of engineering, avoiding geometric mesh distortion of complex cavities inside irregularly shaped non-convex curved components during the dimensionality-upgrading process. Based on the critical condition of fluid shear yielding and retardation, a unidirectional fluid valve geometry is constructed. Utilizing the yielding and suspension physical properties of non-Newtonian fluids, during the grouting pressurization stage, the pressure difference forces residual gas inside to pass through the neck and be discharged into the resonant blind end. At the instant pumping stops and pressure is released, the local fluid shear stress rapidly drops below the fluid yield stress, causing the grout to instantaneously solidify and block at the constricted neck. This structure, from a purely geometric and topological perspective, prevents liquid grout from intruding into the acoustic cavity at the rear end, completely resolving the structural conflict between smooth fluid venting during construction and high-impedance acoustic reflection during service within the same confined space.

[0045] like Figure 3 As shown, S6 calculates the rate of change of Gaussian curvature in the negative space of the three-dimensional entity. If it is greater than the preset tolerance, the penalty multiplier of the local metric distortion functional is increased and the steps of constructing the mapping function and solving the two-dimensional parameter plane are returned. Otherwise, the building information model containing the negative space of the three-dimensional entity is output. Furthermore, in S6, the steps of calculating the rate of change of Gaussian curvature in the negative space of the three-dimensional solid, increasing the penalty multiplier of the local metric distortion functional if it exceeds a preset tolerance, and returning to the steps of constructing the mapping function and solving the two-dimensional parametric plane include: By traversing and extracting all internal surface nodes of the negative space of the three-dimensional solid, the three-dimensional Gaussian curvature matrix is ​​obtained by using a differential geometry algorithm. Extract the two-dimensional Gaussian curvature matrix of the corresponding node in the two-dimensional parametric plane by coordinate index; Calculate the extreme values ​​of the relative deviations between corresponding elements of the three-dimensional Gaussian curvature matrix and the two-dimensional Gaussian curvature matrix, and use the extreme values ​​of the relative deviations as the rate of change of Gaussian curvature; When the comparison confirms that the rate of change of Gaussian curvature is greater than the preset tolerance, the penalty multiplier of the local metric distortion functional is increased by a step size according to the deviation overshoot, and a return instruction is triggered to re-execute the calculation action of constructing the mapping function and solving the two-dimensional parameter plane.

[0046] Specifically, the data input / output process involves the system receiving the array of internal surface nodes of the 3D solid negative space and the coordinate index sequence of the 2D parametric plane generated in the previous stage as the underlying comparison data source. After completing the spatial differential geometry solution, the curvature analysis module generates a Gaussian curvature change rate value with voting rights. When this value exceeds the limit, a trigger return feedback instruction with an updated penalty multiplier is output to the source manifold mapping solution module; when the value meets the standard, the data loop terminates and a 3D building information model containing the exact physical negative space is output to the external engineering terminal.

[0047] When performing nonlinear distortion verification, the control formula for calculating the rate of change of Gaussian curvature by extracting node alignment is set as follows: ; in This represents the calculated rate of change of Gaussian curvature. This represents the total number of surface nodes inside the negative space of a three-dimensional entity. This indicates the specific surface node index number. Indicates the first The three-dimensional Gaussian curvature values ​​of each surface node are obtained by solving a differential geometry algorithm. Specifically, This represents the two-dimensional Gaussian curvature value of the corresponding node extracted through the preimage coordinate index. Within this computational framework, since the internal surface of the three-dimensional solid negative space is generated by pulling back the level set zero isosurface in the two-dimensional parametric plane through an analytical inverse mapping function, the "corresponding node" refers to the strictly mapped coordinate pair between discrete points on the surface of the three-dimensional cavity and the preimage points of the continuous boundary of the zero isosurface in the two-dimensional parametric plane. By comparing the curvature difference between the zero isosurface preimage and the surface of the three-dimensional solid negative space, the topological distortion rate caused by the spatial dimensionality upgrade process on extremely small geometric structures is accurately identified and quantified. To ensure that the physical function of the internal fluid valves of building components does not fail due to dimensional distortion, a preset tolerance threshold must be clearly defined; here, the specific value of the preset tolerance is set to 5%.

[0048] The step control formula for dynamically updating the penalty multiplier based on the deviation overshoot is set as follows: ; in This represents the updated penalty multiplier. This represents the local metric distortion functional penalty multiplier before a return instruction is triggered. This represents the step size coefficient that controls the increment magnitude. This indicates the preset tolerance.

[0049] The steps constitute a post-processing hard error correction closed loop for the entire multiphysics topology reconstruction algorithm. In the preceding step, when performing a non-Euclidean geometric inverse mapping from a 2D parametric plane to a 3D non-convex surface, the local mesh is prone to uncontrollable curvature distortion due to the spatial dimension transformation. The fractal exhaust channels and narrow acoustic resonant necks within prefabricated components are extremely sensitive to spatial dimensions. By traversing and comparing the extreme value differences between the 3D and 2D Gaussian curvature matrices before and after mapping, the system can accurately identify weak nodes on the hidden cavity surface where deformation amplification has occurred. Once the distortion rate is confirmed to exceed the safety margin, the algorithm immediately uses the penalty multiplier of step expansion to force the surface dimensionality reduction expansion in the first step to apply more stringent physical deformation constraints and recalculate the entire chain. This completely eliminates the technical risk of the internal unidirectional fluid valve's physical shut-off function collapsing due to microscopic morphology collapse during the geometric inverse transformation process, establishing absolute accuracy assurance in the transformation process from a mathematically virtual topology network to an industrial-grade manufacturable entity.

[0050] Example 2: In the construction of prefabricated buildings for urban rail transit (such as underground hubs or high-speed rail viaducts), irregularly shaped prefabricated spatial load-bearing nodes with hyperboloid or non-convex surfaces are widely used. During the construction period, these components require high-pressure injection of high-strength non-Newtonian fluid grout to achieve structural anchoring, while during service, they face strong low-frequency vibrations and structural transmitted noise generated by train operation. The following mutually constraining technical problems exist in this application scenario: First, the internal concealed cavity structure of such irregularly shaped components is extremely complex. Conventional flow channel designs are prone to gas-liquid entrainment when pumping high-viscosity slurry, leading to poor exhaust and the formation of fatal structural voids. Second, the internal acoustic resonance blind ends reserved to attenuate specific low-frequency noise directly interfere with the smooth grouting main exhaust flow channel in spatial topology, resulting in a zero-sum game between physical requirements during construction and service. Finally, existing BIM parametric design inevitably suffers from geometric metric distortion when dealing with the dimensional reduction and unfolding of large-curvature non-convex surfaces. This causes the precise geometric structures such as microscopic flow-limiting short necks planned by engineers in the two-dimensional domain to be severely distorted and dimensionally collapsed after being mapped to the three-dimensional solid model, resulting in the complete failure of the expected fluid dynamic phase change interception and high-impedance acoustic reflection functions at the physical manufacturing level. To solve the above problems, this invention provides a BIM-based prefabricated building component design method, the structure of which is as follows: Figure 1 As shown. The specific implementation process of this method is as follows: The system resolves the physical interference contradiction in the spatial distribution of the grouting and exhaust channels within the irregular curved surface component and the acoustic resonant blind cavity from the underlying mathematical logic. After obtaining the initial surface boundary and multi-physics parameters, the system establishes a low-distortion two-dimensional parameter computation domain using the principle of minimizing local metric distortion of Riemannian manifolds. A neural network architecture is then implanted on this mathematical foundation, transforming the fluid dynamics evolution and acoustic frequency response characteristics into residual terms and fusing them into a total loss function. Through backpropagation and weighted optimization, the level set function autonomously generates an isomorphic topological network within the reduced-dimensional plane, balancing low-resistance exhaust of non-Newtonian fluids with high-resistance reflection of low-frequency noise. Subsequently, an analytical inverse mapping is invoked to precisely pull the two-dimensional geometric boundaries back to the physical space, performing three-dimensional entity Boolean subtraction to reconstruct an endogenous negative space containing complex fractal and blind-end structures. Ultimately, a nonlinear distortion correction mechanism was constructed based on the Gaussian curvature change rate. The penalty multiplier was adaptively adjusted according to the extreme value of micro-deformation and iterative recalculation was triggered. This completely eliminated the hidden danger of structural neck size collapse caused by topological dimensional transformation and established a complete technical closed loop from solving abstract multi-field coupled equations to high-precision output of industrial-grade building information models.

[0051] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A BIM-based design method for prefabricated building components, characterized in that, Includes the following steps: S1. Extract the three-dimensional non-convex surface boundary, acoustic spectrum parameters and fluid grouting parameters of the prefabricated components; S2. Define the boundary of the three-dimensional non-convex surface as a two-dimensional Riemannian manifold and construct a mapping function. Minimize the local metric distortion functional to obtain the two-dimensional parametric plane and the analytical inverse mapping function. S3. Construct a neural network for outputting a level set function in the two-dimensional parameter plane, construct acoustic residual terms and fluid residual terms based on the acoustic spectrum parameters and fluid grouting parameters, and embed them into the neural network to construct the total loss function; S4. Assign weights to the acoustic residual term and the fluid residual term and minimize the total loss function to obtain the optimal level set function, and generate a two-dimensional topology network based on the zero isosurface of the optimal level set function. S5. Call the analytical inverse mapping function to pull the two-dimensional topology network back into the three-dimensional physical space, and generate a three-dimensional solid negative space containing fractal flow channels and resonant blind ends in the prefabricated component through Boolean subtraction. S6. Calculate the rate of change of Gaussian curvature of the three-dimensional entity negative space. If it is greater than the preset tolerance, increase the penalty multiplier of the local metric distortion functional and return to the steps of constructing the mapping function and solving the two-dimensional parameter plane. Otherwise, output the building information model containing the three-dimensional entity negative space.

2. The BIM-based prefabricated building component design method according to claim 1, characterized in that, In S1, the extraction of the three-dimensional non-convex surface boundary, acoustic spectrum parameters, and fluid grouting parameters of the prefabricated component includes: The geometric data of the building information model of the prefabricated component is analyzed, and the surface feature point array is extracted to reconstruct the three-dimensional non-convex surface boundary. Search the acoustic specification database of the target environment and extract the low-frequency noise peak frequency sequence to be attenuated as the acoustic spectrum parameter; Extract the pumping pressure threshold of the on-site grouting equipment, and retrieve the fluid yield stress and dynamic viscosity properties from the grouting material database as the fluid grouting parameters.

3. The BIM-based prefabricated building component design method according to claim 1, characterized in that, In S2, defining the three-dimensional non-convex surface boundary as a two-dimensional Riemannian manifold and constructing a mapping function, minimizing the local metric distortion functional to obtain the two-dimensional parametric plane and the analytical inverse mapping function includes: Introducing a first fundamental form containing a metric tensor, the discrete coordinates of the boundary of the three-dimensional non-convex surface are parameterized into the two-dimensional Riemannian manifold; A local metric distortion functional is established with the Jacobian matrix of the mapping function as the variable to measure the distortion rate of the linear transformation of the tangent space; The local metric distortion functional is iteratively minimized using the conjugate gradient optimization algorithm until the distortion rate converges to output the two-dimensional parameter plane. Record the grid node mapping topological correlation matrix when the minimization solution converges, and generate the analytical inverse mapping function based on the grid node mapping topological correlation matrix.

4. The BIM-based prefabricated building component design method according to claim 1, characterized in that, In S3, constructing the neural network that outputs the level set function in the two-dimensional parameter plane includes: Initialize a multi-layer fully connected deep neural network structure in the two-dimensional parameter plane; The input layer nodes of the multi-layer fully connected deep neural network structure are configured to receive the spatial coordinate sequence and time variable of the two-dimensional parameter plane. Configure the output layer nodes of the multi-layer fully connected deep neural network structure to output the level set function; The entity distribution region and the internal cavity region are defined based on the positive and negative signs of the output value of the level set function, and the zero isosurface of the level set function is extracted as the physical boundary of the evolution.

5. The BIM-based prefabricated building component design method according to claim 1, characterized in that, In S3, constructing the acoustic residual term and the fluid residual term based on the acoustic spectrum parameters and the fluid grouting parameters includes: Using the fluid yield stress and dynamic viscosity in the fluid grouting parameters as physical property inputs, a Navier-Stokes equation including the Cauchy stress tensor is constructed. Discrete fluid placement points are arranged in the two-dimensional parameter plane, and the Navier-Stokes equations are transformed into a summation of squared errors for the discrete fluid placement points to generate the fluid residual terms. By combining the low-frequency noise peak frequency in the acoustic spectrum parameters with a preset air static density, a Helmholtz acoustic steady-state partial differential equation is constructed. Discrete acoustic placement points are arranged in the two-dimensional parameter plane, and the Helmholtz acoustic steady-state partial differential equation is transformed into a summation of squared errors for the discrete acoustic placement points to generate the acoustic residual term.

6. The BIM-based prefabricated building component design method according to claim 1, characterized in that, In S4, assigning weights to the acoustic residuals and fluid residuals and minimizing the total loss function to obtain the optimal level set function includes: An irrational constant is extracted as the dominant weight coefficient, and the dominant weight coefficient is multiplied into the fluid residual term; The difference between one and the irrational constant is used as the complementary weight coefficient, and the complementary weight coefficient is multiplied into the acoustic residual term; The total loss function is constructed by combining the data fitting boundary loss term with the weighted fluid residual term and the acoustic residual term; A hybrid optimization algorithm is used to minimize the total loss function during training to update the neural network weights, and the optimal level set function is captured when the loss convergence threshold is met.

7. The BIM-based prefabricated building component design method according to claim 1, characterized in that, In S4, generating a two-dimensional topology network based on the zero isosurface of the optimized level set function includes: Discretize the zero isosurface continuous boundary of the optimal level set function in the two-dimensional parameter plane; The internal negative space region of the closed envelope of the continuous boundary of the zero isosurface is defined as the two-dimensional topological network; In the two-dimensional topology network, fractal guiding topology segments with trunk-branch connection characteristics and blind-end resonant topology segments with tapering and flow-limiting characteristics are identified and marked.

8. The BIM-based prefabricated building component design method according to claim 7, characterized in that, In S5, the step of calling the analytical inverse mapping function to pull the two-dimensional topology network back into the three-dimensional physical space, and generating a three-dimensional solid negative space containing fractal flow channels and resonant blind ends within the prefabricated component through Boolean subtraction includes: Boundary discretization sampling is performed in the two-dimensional topology network to obtain a two-dimensional discrete coordinate array; The two-dimensional discrete coordinate array is input into the analytical inverse mapping function to calculate and transform it into a three-dimensional physical coordinate array. The three-dimensional spatial surface entity boundary is reconstructed based on the aforementioned three-dimensional physical coordinate array; In the solid model of the assembled component, the Boolean subtraction operation is performed using the boundary of the three-dimensional spatial curved surface to remove the corresponding volume and generate the negative space of the three-dimensional solid.

9. A BIM-based prefabricated building component design method according to claim 8, characterized in that, In S5, after performing the Boolean subtraction operation using the boundary of the three-dimensional spatial surface entity to remove the corresponding volume and generate the negative space of the three-dimensional entity, the method further includes: The region generated by pulling back the fractal flow guide topology segment in the negative space of the three-dimensional entity is defined as the fractal flow channel of the main exhaust pumping channel shape. The region in the negative space of the three-dimensional entity generated by pulling back the blind-end resonant topology segment is defined as the resonant blind end connected to the end of the fractal flow channel; The neck dimension of the fractal flow channel transitioning to the resonant blind end is constrained to meet the critical condition of fluid shear yielding retardation, so as to form a unidirectional fluid valve geometry.

10. A BIM-based prefabricated building component design method according to claim 1, characterized in that, In S6, the step of calculating the rate of change of the Gaussian curvature of the negative space of the three-dimensional entity, increasing the penalty multiplier of the local metric distortion functional if it exceeds a preset tolerance, and returning to construct the mapping function and solve the two-dimensional parameter plane includes: All internal surface nodes of the negative space of the three-dimensional entity are extracted by traversing the system, and the three-dimensional Gaussian curvature matrix is ​​obtained by using a differential geometry algorithm. The two-dimensional Gaussian curvature matrix of the corresponding node in the two-dimensional parameter plane is extracted by coordinate indexing; Calculate the extreme values ​​of the relative deviations between the corresponding elements of the three-dimensional Gaussian curvature matrix and the two-dimensional Gaussian curvature matrix, and use the extreme values ​​of the relative deviations as the rate of change of the Gaussian curvature; When the comparison confirms that the rate of change of Gaussian curvature is greater than the preset tolerance, the penalty multiplier of the local metric distortion functional is increased by a step size according to the deviation overshoot, and a return instruction is triggered to re-execute the calculation action of constructing the mapping function and solving the two-dimensional parameter plane.

Images