3D Simulation and Revitalization Design System for Functional Transformation of Industrial Buildings

Abstract

This invention discloses a three-dimensional simulation and revitalization design system for the functional transformation of industrial buildings. The system includes a data acquisition module, a semantic segmentation and reconstruction module, a structural safety assessment module, a functional space parameterization generation module, a revitalization design optimization module, and a dynamic simulation and visualization module. This invention directly uses the hazardous area and modifiable area labels output from the structural safety assessment as hard constraints for the functional space parameterization generation and the optimization objectives of the multi-objective genetic algorithm in the revitalization design optimization module. This enables the proactive guidance of spatial layout by structural intervention information and solves the problem of automatic trade-off between preserving original features, structural safety, and new functional flow lines in the revitalization design of industrial buildings.

Description

Technical Field

[0001] This invention relates to the fields of computer-aided design and building information modeling, and in particular to a three-dimensional simulation and revitalization design system for the functional transformation of industrial buildings. Background Technology

[0002] Due to urban industrial restructuring and the "retreat from secondary industry and advance into tertiary industry" policy, a large number of industrial buildings (such as old factories, warehouses, and steelmaking workshops) need functional transformation and are being converted into cultural and creative industrial parks, exhibition halls, office spaces, or commercial complexes. Existing technologies include auxiliary systems that utilize 3D laser scanning technology to acquire point cloud data of existing industrial buildings and combine it with Building Information Modeling (BIM) for renovation design. Some studies have disclosed methods for generating building functional layouts based on parametric modeling, which can automatically generate preliminary room division schemes according to given spatial boundaries and functional requirements.

[0003] First, existing systems typically treat structural safety assessment and functional space design as two separate, sequential steps. Structural engineers first calculate removable or reinforced components, and then architects use this information to create the functional layout. The lack of real-time data feedback and coupled optimization between the two processes leads to long design cycles and a high likelihood of unfeasible solutions. Second, most 3D simulation systems focus solely on the transformation of building geometry, neglecting the unique structural modular logic (column grid, roof truss spacing) and cultural value of industrial relics. When automatically generating functional zones, they often destroy the original spatial characteristics, causing the renovated building to lose its original historical readability. Third, existing scheme generation methods are mostly rule-based deterministic algorithms, unable to balance and optimize the three conflicting goals of "maximizing the preservation of industrial characteristics," "minimizing structural intervention," and "meeting the requirements of new functional flow." They typically only output a single scheme, failing to meet the needs of multi-disciplinary collaborative decision-making in revitalization design. Fourth, there is currently a lack of a system architecture that can deeply integrate the three-dimensional point cloud semantic segmentation results, the modifiable area labels from finite element analysis, and the dynamic functional space reorganization. This results in the identification results of dangerous areas and modifiable areas not being able to directly drive the search process of the spatial layout algorithm, causing a disconnect between the calculation results and the actual construction design.

[0004] Therefore, in response to the problems mentioned above, this invention proposes a three-dimensional simulation and revitalization design system for the functional transformation of industrial buildings. Summary of the Invention

[0005] To overcome the problems in existing technologies where the design process for functional transformation of industrial buildings is disconnected from the 3D simulation system and cannot automatically generate revitalization schemes that balance structural safety and preservation of industrial features under multi-objective conflicts, this invention proposes a 3D simulation and revitalization design system for functional transformation of industrial buildings. By introducing a 3D basic model with labels for modifiable areas, and using this model to constrain the parameterized generation and collision detection of functional spaces, the system enables the active guidance of structural intervention information on spatial layout algorithms and Pareto optimization under non-dominated sorting.

[0006] The technical solution of this invention is: a three-dimensional simulation and revitalization design system for the functional transformation of industrial buildings, comprising:

[0007] The data acquisition module is used to acquire the original 3D point cloud data, historical image data and structural parameters of industrial buildings;

[0008] The semantic segmentation and reconstruction module performs semantic segmentation on the original 3D point cloud data based on deep neural networks, identifies the structural components and non-structural elements of the building, and generates a 3D basic model with semantic labels. Among them, structural components include columns, beams, walls and roof trusses, and non-structural elements include doors and windows, equipment foundations and industrial relics.

[0009] The structural safety assessment module calculates the remaining bearing capacity of each structural component using the finite element method based on structural parameters and semantic segmentation results, and marks the dangerous areas and areas that can be modified.

[0010] The functional space parameterization generation module receives the target function type and capacity index input by the user, automatically generates multiple candidate functional zoning schemes based on preset spatial layout rules and the modifiable area, and converts each candidate scheme into a three-dimensional spatial block. The target function type includes at least one of office, commercial, exhibition and cultural and creative industries.

[0011] The activation design optimization module performs collision detection between the three-dimensional spatial blocks corresponding to each candidate functional zoning scheme and the three-dimensional basic model with semantic tags. Combined with the dangerous areas and modifiable areas marked by the structural safety assessment module, the optimization scheme is generated by a multi-objective genetic algorithm with the optimization objectives of maximizing the preservation of original industrial features, minimizing structural intervention and meeting the target functional flow.

[0012] The dynamic simulation and visualization module generates a 3D comparison animation of the industrial building before and after the functional transformation, as well as an immersive virtual reality-based interactive interface, based on the optimized revitalization design scheme.

[0013] Preferably, the data acquisition module includes a lidar unit mounted on a UAV and a ground-based fixed 3D laser scanner. The UAV lidar is used to acquire point cloud data of the building exterior and high-level locations, while the ground-based 3D scanner is used to acquire point cloud data of the building interior and low-level locations. The two types of data are registered and fused using an iterative nearest-point algorithm to form complete original 3D point cloud data. The maximum number of iterations in the iterative nearest-point algorithm for registration and fusion is set to 50, and the root mean square error convergence threshold is 0.005m.

[0014] The semantic segmentation and reconstruction module uses PointNet++ or RandLA-LA-Net as its deep neural network. This network takes the coordinates and reflection intensity of each point in the original 3D point cloud data as input and outputs the probability of each point belonging to a structural component or a non-structural element. The probability threshold for each category is set to 0.7. At the same time, the region growing algorithm is used to cluster points of the same category into independent component instances. The growth angle threshold of the region growing algorithm is set to 15° and the distance threshold is set to 0.1m, thereby generating a 3D basic model with instance-level semantic labels.

[0015] The historical texture mapping module receives historical image data and extracts feature points from the images using the Scale Invariant Feature Transform (SIFT) algorithm. The SIFT algorithm sets a peak threshold of 0.04, an edge threshold of 10, and a nearest neighbor distance ratio threshold of 0.75. The extracted feature points are matched with their corresponding spatial locations in the original 3D point cloud data using the Random Sample Consensus (RANSAC) algorithm, with RANSAC iterations set to 2000 times and an error tolerance of 5 pixels. The successfully matched image textures are then mapped onto the corresponding surfaces of the 3D base model through perspective projection, generating a 3D model with historical appearance textures. This model is used to display the appearance comparison before and after activation in the dynamic simulation and visualization module.

[0016] The bill of quantities generation module extracts the geometric dimensions and material properties of components to be retained, demolished, and added from the 3D foundation model based on the optimized activation design scheme. According to the "Construction Engineering Quantity List Pricing Specification" GB 50500-2013 and the "Building Construction and Decoration Engineering Quantity Calculation Specification" GB 50854-2013, it automatically calculates the quantities of concrete (classified by strength grade C25 and C30), steel (classified by section steel and reinforcing steel), masonry (classified by brick type and mortar strength), and doors and windows (classified by material and opening method). The calculation accuracy is retained to two decimal places (unit: m). 3 , t, m 2 The data is then output as a structured table to assist in preparing the construction budget for the revitalization and renovation project.

[0017] Preferably, the structural safety assessment module first extracts the geometric contours and spatial positions of each structural component based on the semantic segmentation results. Then, it establishes a finite element model for each component based on the material strength, cross-sectional dimensions, and reinforcement information in the structural parameters (the finite element mesh size is 1 / 20 of the minimum component size and not greater than 0.2m). Nonlinear static pushover analysis is used to calculate the remaining bearing capacity of the component under the current load. The load increment step applied in the pushover analysis is 0.1 times the design load, and the convergence tolerance within each load step is set to 0.001. The area where the remaining bearing capacity is lower than 85% of the threshold specified in the "Code for Design of Building Structures" GB50009 is marked as a dangerous area, and the area where the non-load-bearing partition wall and equipment foundation are located with the remaining bearing capacity higher than the threshold is marked as a modifiable area.

[0018] Preferably, the preset spatial layout rules in the functional space parameterization generation module are based on the original column grid module of industrial buildings: the center distance between two adjacent rows of columns is used as the basic module unit, and the value range of this basic module unit is 4500mm-12000mm, with preferred values ​​including 6000mm, 7500mm, and 9000mm; the space enclosed by each column span is used as the basic unit, and the minimum area of ​​the basic unit is set according to the target function type: not less than 9m² for office buildings. 2 Commercial buildings should be no less than 15m 2 Exhibitions should be no less than 20m in size. 2 Cultural and creative products should be no less than 12m 2 Based on the minimum room size, evacuation passage width (not less than 1.2m), and net height requirements (net height not less than 2.6m for office buildings and not less than 4.0m for exhibition buildings) corresponding to the target functional type, spatial syntax analysis in graph theory is used to generate adjacency relationships between units. The integration threshold of spatial syntax analysis is set to 0.4-0.6. Then, the basic units are combined into a functional zoning scheme that meets the capacity index through a recursive backtracking algorithm. The maximum search depth of the recursive backtracking is set to 20 layers, and the backtracking pruning condition is that the local functional area deviation exceeds 5%.

[0019] Preferably, the multi-objective genetic algorithm used in the activation design optimization module includes the following fitness function:

[0020] Original feature retention is calculated as the ratio of the number of industrial relics point clouds that have not been demolished or obscured in the candidate scheme to the total number of original industrial relics point clouds. The value of this ratio is in the range of [0,1], and the objective is to maximize it.

[0021] The structural intervention cost is calculated by multiplying the quantities of reinforcement, demolition, or addition of structural components required in the hazardous area by a preset unit cost coefficient, based on the modifiable area output by the structural safety assessment module. The unit cost coefficient is set as follows: 450 yuan / m² for concrete components.3 Steel components: 6800 yuan / ton; Masonry components: 280 yuan / m 3 ;

[0022] Functional fit, based on spatial syntax computing theory, calculates the reciprocal of the deviation between the topological distance between the main functional spaces in the candidate scheme and the optimal path required by the target functional streamline, with the deviation penalty factor set to 1.2;

[0023] Annual energy consumption per unit area was calculated using the building energy simulation engine EnergyPlus 9.6.0, based on the wall, door, window, and roof structure information of the functional zones in the candidate schemes (heat transfer coefficient was taken as 0.45 W / (m²) for walls). 2 ·K), doors and windows 2.5W / (m 2 ·K), roof 0.35W / (m 2 ·K), combined with local annual meteorological data (hourly dry-bulb temperature, solar radiation illuminance) and the indoor environmental setpoints corresponding to the target function type (heating 20℃, cooling 26℃, lighting power density 8W / m²), 2 The annual energy consumption for heating, cooling, and lighting is calculated by dividing the total annual energy consumption by the total building area to obtain the annual energy consumption per unit area (unit: kWh / (m²)). 2 •a)), the objective is to minimize;

[0024] The algorithm employs non-dominated sorting (with a maximum sorting level of 10) and a crowding distance selection operator (with a crowding distance threshold of 0.01). The population size is 80, the crossover probability is 0.85, and the mutation probability is 0.06. After 200 iterations, the Pareto front solution set is output as the optimized activation design scheme.

[0025] Preferably, the dynamic simulation and visualization module also includes an interactive editing module. This interactive editing module allows users to directly select a 3D spatial block in the immersive roaming interface and adjust its position (0.1m movement step), size (0.05m zoom step), or functional attributes (select office / commercial / exhibition / cultural and creative) via a virtual reality controller or stylus. The modified parameters are fed back to the activation design optimization module in real time, triggering re-optimization calculations and dynamically updating the 3D comparison animation and collision detection results. The system response latency for the real-time feedback is less than 100ms.

[0026] The beneficial effects of this invention are:

[0027] 1. This invention solves the problems in the prior art where structural safety assessment and functional space design are separated, and information cannot be fed back in real time, resulting in long design cycles and the generation of infeasible solutions. By simultaneously sending the three-dimensional basic model generated by the semantic segmentation and reconstruction module to the structural safety assessment module and the functional space parameterization generation module, and directly using the dangerous area and modifiable area labels output by the structural safety assessment module as hard constraints of the functional space parameterization generation module and optimization targets of the active design optimization module, this invention solves the problems in the prior art where structural safety assessment and functional space design are separated, and information cannot be fed back in real time, resulting in long design cycles and the generation of infeasible solutions.

[0028] 2. In the activation design optimization module, this invention employs a multi-objective genetic algorithm to simultaneously perform Pareto optimization using four conflicting fitness functions: original feature retention, structural intervention cost, functional fit, and annual energy consumption per unit area. In particular, it incorporates the quantification of the number of industrial relic point clouds into the retention calculation and combines it with the labeling of modifiable areas to minimize structural intervention within dangerous areas. This solves the problem that existing technologies struggle to automatically balance the maximization of industrial feature retention, minimization of structural intervention, and satisfaction of new functional streamline requirements.

[0029] 3. This invention generates candidate schemes based on the original column grid module and spatial syntax analysis of industrial buildings through a functional space parameterization generation module. Combined with an interactive editing module, it allows users to directly modify block parameters in a virtual reality environment and trigger the re-optimization of the active design optimization module. The response latency is less than 100 milliseconds. This solves the problems of existing 3D simulation systems ignoring the unique modular logic of industrial buildings and failing to achieve immersive real-time adjustments and multi-disciplinary collaboration, and greatly improves the efficiency and operability of active design. Attached Figure Description

[0030] Figure 1 The diagram shown is a schematic representation of the system framework of the present invention.

[0031] Figure 2 The diagram shown is a schematic representation of the iterative nearest point algorithm of this invention.

[0032] Figure 3 The diagram shown is a schematic representation of the region growing algorithm of this invention.

[0033] Figure 4 The diagram illustrates the process of calculating the remaining load-bearing capacity of each structural component using the finite element method of this invention. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but 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.

[0035] Please see Figure 1 The present invention provides an embodiment:

[0036] In this embodiment, the data acquisition module will be described in detail:

[0037] The data acquisition module is used to acquire raw 3D point cloud data, historical image data, and structural parameters of industrial buildings. The acquisition of raw 3D point cloud data utilizes a combination of a lidar unit mounted on a drone and a ground-based fixed 3D laser scanner. The drone lidar primarily acquires point cloud data of the building's exterior and higher elevations, such as roof trusses, skylights, high side windows, and the top of chimneys—areas difficult to scan from the ground. The ground-based 3D scanner acquires point cloud data of the building's interior and lower elevations, including the ground, lower walls, column bases, and equipment foundations. These two types of data are complementary in terms of spatial coverage. To form complete and coordinate-system-consistent raw 3D point cloud data, registration and fusion of this data are required.

[0038] Please see Figure 2 The registration algorithm used in this module is the iterative nearest point algorithm, and its detailed steps are as follows:

[0039] The first step involves designating the point cloud acquired by the UAV's LiDAR as the source point cloud P and the point cloud acquired by the terrestrial 3D scanner as the target point cloud Q. In the overlapping region of the two point clouds, points with prominent features are selected as initial matching point pairs by calculating the rate of change of the normal vector and the curvature of the point clouds. The curvature threshold is set to 0.1, and the normal vector angle threshold is set to 15°.

[0040] The second step is to process each point in the source point cloud P. Find the point in the target point cloud Q with the closest Euclidean distance. , forming a set of matching point pairs C = {( , Distance calculations are accelerated using a KD-tree, with the maximum tree depth set to 20.

[0041] The third step is to use the current set of matched point pairs to calculate the rigid body transformation (rotation matrix R and translation vector t) so that the objective function... Minimize. This minimization problem is solved using singular value decomposition.

[0042] Fourth step: Apply R and t to the source point cloud P to obtain the transformed point cloud. .

[0043] Fifth, repeat steps two through four until the convergence condition is met.

[0044] The maximum number of iterations in this module is set to 50, and the root mean square error convergence threshold is 0.005m. The iteration stops when the change in root mean square error between two consecutive iterations is less than the threshold or the maximum number of iterations is reached.

[0045] After registration, the coordinate systems of the two point clouds need to be aligned with the engineering coordinate system of the ground scanner. The fused original 3D point cloud data includes the 3D coordinates (x, y, z) and reflection intensity of each point. The average point cloud density is 200 points per square meter, and for critical structural components (beam-column joints, etc.), the density is increased to 500 points per square meter. Historical image data is obtained by collecting scanned copies of photographs and design drawings of the industrial building from different historical periods from the owner or archives department, with a resolution of no less than 1920×1080 pixels. Structural parameters include material strength grades, cross-sectional dimensions, reinforcement information, and foundation types from the original architectural design drawings. These parameters are entered into the system as structured text.

[0046] In this embodiment, the semantic segmentation and reconstruction module is described in detail:

[0047] The semantic segmentation and reconstruction module uses a deep neural network to perform semantic segmentation on the original 3D point cloud data, identifying structural components and non-structural elements of the building, and generating a 3D basic model with semantic labels. This module uses either PointNet++ or RandLA-Net as the deep neural network. This example uses RandLA-Net for illustration. Specifically:

[0048] RandLA-Net is a lightweight point cloud semantic segmentation network. Its core idea is a combination of random downsampling and local feature aggregation. The network input consists of the coordinates (x, y, z) and reflection intensity of each point in the original 3D point cloud data, forming an N×4 matrix, where N is the total number of points in the cloud (preferably a value of N). - (on a scale of magnitude), the network structure consists of multiple encoder-decoder layers:

[0049] For each encoder layer, there is a random downsampling module and several local feature aggregation modules. The random downsampling module randomly retains some points with a probability p=0.5, reducing the number of points in each layer by about half. The local feature aggregation modules use K-nearest neighbors (K=16) to find the neighbors of each point, extract local geometric features through a shared multilayer perceptron (MLP, with hidden layer dimensions of 64, 64, and 128 respectively), and then use max pooling and attention mechanisms to aggregate the features of neighboring points to the center point. After four encoder layers, the feature dimension gradually increases to 512, and the number of points decreases to about 1 / 16 of the original point cloud.

[0050] For the decoder layer, upsampling interpolation is used to restore the features of the sparse point cloud to their original density. Each decoder layer receives the skip connection features output from the corresponding encoder layer and inputs them into a shared MLP (dimensions 256, 128) and a Dropout layer (dropout rate 0.2). The final output is the probability of each point belonging to a predefined category. The predefined categories include columns, beams, walls, roof trusses, doors and windows, equipment foundations, and industrial relics, totaling 7 categories. The category probability threshold is set to 0.7, meaning that a point is only classified into that category if its maximum predicted probability exceeds 0.7; otherwise, it is marked as an unidentified point and awaits manual verification later.

[0051] The dataset used in the network training process is a pre-built industrial building point cloud database, containing 50 industrial buildings of different ages and types (brick-concrete structure, frame structure, and trunk frame structure). The point cloud of each building has been meticulously annotated manually. The loss function during training is weighted cross-entropy loss. Since the number of points for structural components (columns and beams) is usually less than that for non-structural elements (walls), columns and beams are assigned a weight coefficient of 1.5, and the other categories are assigned a weight of 1.0. The optimizer used is Adam, with an initial learning rate of 0.001, a decay factor of 0.95 every 5 epochs, a total of 200 training epochs, and a batch size of 8.

[0052] Please see Figure 3 After semantic segmentation, each point is assigned a category label. Since points belonging to the same physical component in the point cloud are typically discrete—for example, a pillar might have thousands of points—these points are spatially clustered but independent of each other. To generate instance-level semantic labels (distinguishing between independent component instances such as "first pillar" and "second pillar"), this module uses a region growing algorithm to cluster points of the same category into independent component instances. The detailed steps of the region growing algorithm are as follows:

[0053] The first step is to initially mark all points as "unvisited".

[0054] The second step is to randomly select an unvisited point as the seed point. The seed point must belong to a certain category, such as a column.

[0055] The third step is to search for all neighboring points within a radius of 0.3m centered on the seed point. If the neighboring points have the same category as the seed point, they are included in the same growth region and marked as visited.

[0056] The fourth step is to recursively execute the third step on newly included points until no new points of the same category are added.

[0057] The fifth step is to treat the point set obtained by growth as an independent component instance and assign it a unique instance ID.

[0058] Jump back to step two until all points have been visited.

[0059] During the growth process, to avoid incorrect merging caused by different components adhering in space, angle and distance thresholds need to be set. In this module, the growth angle threshold is set to 15°, and the distance threshold is set to 0.1m. The angle threshold controls the angle between the normal vectors of two points during growth. If the angle between the normal vectors of two adjacent column points exceeds 15°, it indicates that they may belong to columns in different directions and should not be merged. The distance threshold limits the growth step size, preventing the connection of different components across gaps. After the region growth algorithm, a 3D basic model with instance-level semantic labels is generated. This model is stored in file format, containing the coordinates, category ID, instance ID, and reflection intensity of each point, which can be directly read by subsequent modules.

[0060] In this embodiment, the structural safety assessment module is described in detail:

[0061] Please see Figure 4 The structural safety assessment module calculates the remaining load-bearing capacity of each structural component using the finite element method based on structural parameters and semantic segmentation results, and marks dangerous areas and areas that can be modified. Specifically:

[0062] The first step is to extract the geometric contours and spatial positions of the components. From the 3D basic model with instance-level semantic labels, all point clouds of each independent structural component are extracted based on the instance ID. The geometric contours of the components are obtained by fitting the minimum bounding rectangle or cylinder of the point clouds: for columns, the cross-sectional width, cross-sectional height, and column height are extracted; for beams, the cross-sectional width, cross-sectional height, and span are extracted; for walls, the thickness, length, and height are extracted. The measurement accuracy of these geometric dimensions is controlled within ±5mm, and the coordinate origin and orientation of the components in space are recorded simultaneously.

[0063] The second step is to establish a finite element model for each component. Based on the material strength, cross-sectional dimensions, and reinforcement information in the structural parameters, an eight-node hexahedral element (C3D8) is used to create the finite element model for each component. During mesh generation, the element size is taken as 1 / 20 of the minimum component size and not greater than 0.2m. For a typical industrial building column (cross-section 400mm × 500mm, column height 6m), the minimum size is 400mm, and 1 / 20 is 20mm. Therefore, the element size is taken as 20mm, with 300 elements along the column height direction and 20 elements along each cross-section direction, totaling 20 × 20 × 300 = 120,000 elements. For beams, the element size is also taken as 1 / 20 of the minimum cross-sectional size. Reinforcing bars are embedded in the concrete elements using bar elements (T3D2), and the main reinforcement area and stirrup spacing are set according to the reinforcement information.

[0064] The third step is to apply loads and boundary conditions. Based on the original building design drawings and the "Code for Design of Building Structures" (GB50009), determine the current dead load (self-weight, floor construction self-weight) and live load borne by the components. The live load is determined according to the target function type. For example, when converting it into an exhibition hall, the live load is taken as 3.5 kN / m. 2 The boundary conditions are: fixed constraint at the column base (Ux=Uy=Uz=URx=URy=URz=0), and the connection between the beam end and the column is set as hinged or rigid (depending on the original structural form).

[0065] The fourth step involves progressively increasing the load factor from 0 to twice the design load, with the load increment step set to 0.1 times the design load. Within each load step, the Newton-Raphson iteration is used to solve the nonlinear equations, with a convergence tolerance of 0.001 (the normalized values ​​of displacement and force residuals are less than 0.001). During the pushover analysis, the stress, strain, and plastic development of the member are recorded for each load step. When the stress at any section of the member reaches the material's ultimate strength (1.2 times the design value of the axial compressive strength of concrete, or 1.1 times the yield strength of steel), or the displacement exceeds the code limit (the displacement at the top of the column is 1 / 50 of the column height), it is considered to have reached the ultimate bearing capacity state, and the load at this point is the remaining bearing capacity of the member.

[0066] Fifth, areas containing components with remaining load-bearing capacity below 85% of the threshold specified in the "Code for Design of Building Structures" GB 50009 are marked as hazardous areas. Areas containing non-load-bearing partitions and equipment foundations with remaining load-bearing capacity above this threshold are marked as areas suitable for modification. Non-load-bearing partitions, even if demolished, will not affect the safety of the main structure, and are therefore marked as suitable for modification; equipment foundations, if no longer in use, can be retained as industrial relics or demolished depending on new functional requirements, and are also marked as suitable for modification.

[0067] In this embodiment, the function space parameterization generation module is described in detail:

[0068] The functional space parameterization generation module receives the user's input of the target functional type and capacity index. Based on preset spatial layout rules and modifiable areas, it automatically generates multiple candidate functional zoning schemes and converts each candidate scheme into a three-dimensional spatial block. The target functional type includes at least one of office, commercial, exhibition, and cultural and creative industries. Capacity indexes include, for example, "total building area of ​​office area 2000m²". 2 "Accommodates 150 workstations" and "Exhibition area 800m²" 2 "3 exhibition halls", etc.

[0069] The preset spatial layout rules of this module are based on the original column grid module of industrial buildings. Industrial buildings generally adopt a unified column grid during design and construction, using the center distance between two adjacent rows of columns as the basic modular unit. The value range of this basic modular unit is 4500mm-12000mm, with typical values ​​including 6000mm, 7500mm, and 9000mm. The space enclosed by each column span is taken as the basic unit (i.e., the rectangular area defined by four adjacent columns), and the area of ​​the basic unit varies accordingly under different column grid specifications.

[0070] The basic unit is the smallest operational object for dividing functional spaces. To accommodate the room size requirements of different functional types, this invention sets the minimum area of ​​the basic unit according to the target functional type, with office spaces not less than 9m². 2 Commercial buildings should be no less than 15m 2 Exhibitions should be no less than 20m in size. 2 Cultural and creative products should be no less than 12m 2 When the area of ​​a basic unit does not meet the minimum room size, multiple adjacent basic units may be combined for use.

[0071] When generating a functional zoning plan, the first step is to determine the organizational logic of the functional spaces based on the minimum room size, evacuation route width, and clear height requirements corresponding to the target function type. Specifically:

[0072] Minimum room size: Single office space not less than 9m² 2 The aspect ratio should not exceed 2:1; the width of commercial shops should not be less than 3.6m and the depth should not be less than 4.2m; the short side of exhibition halls should not be less than 6m and the net height should not be less than 4.0m.

[0073] Evacuation route width: According to the "Code for Fire Protection Design of Buildings", the clear width of the corridor shall not be less than 1.2m. When the function is exhibition, the width of the main evacuation route shall not be less than 1.5m.

[0074] Clear height requirements: Office space must have a clear height of no less than 2.6m, exhibition space no less than 4.0m, and cultural and creative studios no less than 3.0m. Clear height constraints need to be verified in conjunction with the original beam bottom elevation and roof truss lower chord elevation of the industrial building.

[0075] This invention employs spatial syntax analysis from graph theory to generate adjacency relationships between units. Spatial syntax abstracts architectural space as a topological graph composed of nodes (basic units) and connecting edges (openings and doorways between units). The integration degree of each basic unit is calculated, reflecting its accessibility within the entire building, with an integration degree threshold set at 0.4-0.6. Units with high integration degree are suitable for arranging public activity spaces, while units with low integration degree are suitable for arranging auxiliary spaces. Through integration degree ranking, this invention preliminarily determines the functional attribute tendencies of each basic unit.

[0076] Then, a recursive backtracking algorithm is used to combine the basic units into a functional partitioning scheme that meets the capacity index. The detailed steps of the recursive backtracking algorithm are as follows:

[0077] The first step is to sort all the basic units of the industrial building plan from highest to lowest integration level and create an empty scheme list.

[0078] The second step is to select the unit with the highest integration from the units that have not yet been assigned functions and begin assigning functions.

[0079] The third step involves recursively attempting to add adjacent, unassigned units to the current functional partition, starting from the initial unit. After each addition, the total area of ​​the current partition is calculated to determine if it exceeds the upper limit of the target functional capacity (allowing an excess of no more than 5%). If the area remains less than the lower capacity limit after addition, the recursion continues; if the lower capacity limit is reached or exceeded but the upper limit is not exceeded, it is recorded as a feasible functional partition.

[0080] Fourth, if the current partition area has exceeded 105% of the capacity limit, then backtrack and stop expanding. If a danger zone (which must not be allocated any functional space) is included, then backtrack immediately.

[0081] Fifth, after completing a functional partition, mark the basic units it occupies as allocated, and then repeat steps two through four for the remaining unallocated units until all basic units are allocated or the remaining units cannot meet the minimum area requirements for any function.

[0082] The sixth step involves generating multiple different functional partitioning schemes by changing the selection order of the starting unit and the priority of the expansion direction. Each run of this module generates no fewer than 10 candidate schemes.

[0083] The set of basic units covered by each functional zone in each candidate scheme is converted into a three-dimensional spatial block based on the floor height and the location of structural components. The three-dimensional spatial block is a hexahedral mesh model, its bottom outline defined by the column positions of the basic units, and its height determined by the required clear height for that functional type plus the structural thickness. For example, if an exhibition functional zone covers 4 basic units (each 6m × 6m), with a required clear height of 4.0m and a beam height of 0.6m, then the dimensions of the three-dimensional spatial block are 12m (length) × 12m (width) × 4.6m (height).

[0084] In this embodiment, the activation design optimization module will be described in detail:

[0085] The activation design optimization module performs collision detection between the three-dimensional spatial blocks corresponding to each candidate functional zoning scheme and the three-dimensional basic model with semantic labels. It also combines the dangerous areas and modifiable areas marked by the structural safety assessment module. The optimization objectives are to maximize the preservation of original industrial features, minimize structural intervention, and meet the target functional flow. This invention uses a multi-objective genetic algorithm to generate the optimized activation design scheme.

[0086] This example illustrates the collision detection method, specifically:

[0087] Before each fitness evaluation, it is necessary to detect whether there are spatial conflicts between the 3D space block and the original structural components. This module adopts a layered detection method based on axis-aligned bounding boxes.

[0088] The first step is to create an axis-aligned bounding box for each structural component and record its minimum and maximum x, y, z coordinates.

[0089] The second step is to create an axis-aligned bounding box for the three-dimensional space block corresponding to each functional area.

[0090] Third, if the two bounding boxes do not overlap in any dimension, then there is no collision.

[0091] The fourth step involves performing precise triangle-triangle intersection detection for cases with overlap (since the faces of structural components and blocks are represented by triangular meshes). If a solid part of a block is detected to intrude into a load-bearing structural component, and the intrusion distance exceeds 5cm, it is determined to be a collision. This candidate solution is assigned a very low fitness (penalty value) and is thus eliminated in the evolutionary process.

[0092] This example uses a multi-objective genetic algorithm to optimize four fitness functions. The specific steps and parameters of the algorithm are as follows:

[0093] Each individual (an active design) consists of a string of integer gene codes, the length of which is equal to the total number of basic units contained in the industrial building. Each gene bit takes an integer value between 0 and N, where N is the number of target functional types. The order of the genes is consistent with the index order of the basic units. For example, a factory containing 100 basic units would be encoded as an integer vector of length 100.

[0094] The population size is set to 80. The initial generation method for each individual in the population includes: first, randomly shuffling the allocation order of function types in each function partition; then, calling the recursive backtracking algorithm of the function space parameterization generation module (preserving the original column grid module constraints but randomizing the starting unit and expansion direction) to generate a complete allocation scheme, where all basic units are assigned a function type. If a random combination leads to the incorrect allocation of dangerous areas, regeneration is performed. This process is repeated until 80 valid individuals are obtained.

[0095] For each individual (scheme), the system calculates the following four fitness function values:

[0096] (1) Original feature retention (F1, maximized):

[0097] Total number of point clouds of primitive industrial relics The number of industrial relic point clouds that were not demolished or obscured in the candidate schemes was statistically obtained from the results of the semantic segmentation and reconstruction module. .

[0098] The criteria for determining whether an industrial relic site has not been demolished or obscured are as follows: For each industrial relic site, check whether its three-dimensional spatial location is covered or encroached upon by any newly added functional partition blocks. If the nearest surface of the block is more than 0.2m away from the site and does not directly obstruct the view, it is considered unobstructed; if the block directly covers the site's coordinates, it is considered demolished or obscured. The ratio is calculated as follows: The value ranges from [0,1]. The closer the value is to 1, the better the original industrial characteristics are preserved.

[0099] (2) Cost of structural intervention (F2, minimized):

[0100] Based on the modifiable areas output by the structural safety assessment module, the quantities of reinforcement, alteration, or addition of structural components required in the hazardous areas of the candidate solutions are multiplied by a preset unit cost coefficient and then summed. Specifically:

[0101] First, extract all functional zoning blocks within the hazardous area in the plan. For each such block, analyze whether it requires the demolition or relocation of structural components within the hazardous area, such as removing a column to open up a space. The cost of each demolished hazardous area component is calculated by multiplying its volume or weight by a unit cost factor: 450 yuan / m² for concrete components. 3 Steel components: 6800 yuan / ton; Masonry components: 280 yuan / m 3 Meanwhile, for hazardous area components that cannot be demolished but must be retained, if functional zoning blocks encroach upon them, external reinforcement or additional supports are required. The cost is calculated as 50% of the component's volume multiplied by the corresponding coefficient. Finally, the quantity of newly added structural components is also calculated at the same unit price. The sum of all costs yields F2, in yuan. A lower value is better.

[0102] (3) Functional fit (F3, maximized):

[0103] Based on spatial syntax computing theory, the reciprocal of the deviation between the topological distance between the main functional spaces in the candidate schemes and the optimal path required by the target functional streamline is calculated. Specifically:

[0104] First, a spatial topology map is constructed based on the functional zoning results of the scheme (nodes represent various functional zones, and edges indicate doorways or openings between adjacent zones (opening width ≥ 1.2m)). Then, the user or design specifications can provide the target functional flow requirements, forming a desired node access order. The sum of the shortest path distances between corresponding nodes on the actual topology map of the scheme is then calculated. The sum of the straight-line distances (or ideal distances) between nodes in the ideal order and adjacent nodes. .deviation Take the reciprocal of the deviation as If Δ > 0.5, then multiply by an additional penalty factor of 1.2. The value of F3 ranges from (0,1). Ultimately, the larger the F3, the better.

[0105] (4) Annual energy consumption per unit area (F4, minimize):

[0106] Using the building energy simulation engine EnergyPlus 9.6.0, based on the wall, door, window, and roof structure information of the functional zones in the candidate schemes, combined with local typical meteorological year data and indoor environmental setpoints corresponding to the target function type, the annual total energy consumption for heating, cooling, and lighting (unit: kWh) was calculated and divided by the total building area (unit: m²). 2 ), resulting in F4 (kWh / (m 2 ·a)). Specific structural information is taken as follows: wall heat transfer coefficient 0.45 W / (m²). 2 ·K), heat transfer coefficient of doors and windows 2.5W / (m²) 2 ·K), roof heat transfer coefficient 0.35W / (m²)2 •K). The simulation period is one full year (8760 hours), and the time step is set to 6 minutes. The lower the F4, the better.

[0107] In each generation of evolution, all individuals in the population are stratified according to Pareto dominance. Individual 1 dominates individual 2 if and only if 1 is not inferior to 2 in all four fitnesss and is superior to 2 in at least one fitness. The upper limit of the non-dominated ranking is set to 10 levels; individuals exceeding 10 levels are directly eliminated. For individuals within the same level, crowding distance is calculated. Specifically, for each fitness function, the fitness values ​​of individuals within that level are ranked. The crowding distance of individuals at both ends is set to infinity, and the crowding distance of a middle individual is the normalized sum of the differences between its two preceding and following individuals at that fitness. The crowding distance threshold is set to 0.01 (individuals with a crowding distance less than 0.01 are preferentially eliminated to ensure diversity).

[0108] The selection operator uses a binary tournament selection method, randomly selecting two individuals, prioritizing the one with the lower non-dominated layer; if the layers are the same, the one with the higher crowding degree is selected.

[0109] The crossover probability is set to 0.85. Single-point crossover is used, randomly selecting a crossover point and exchanging gene segments from two parent individuals after that point. Since gene encoding represents the functional allocation of basic units, unreasonable allocations may occur after crossover. Repair is required after crossover: for mismatched functional types at the boundary, local space parsing is invoked for fine-tuning (swapping labels of adjacent units to minimize functional conflicts).

[0110] The mutation probability is set to 0.06. For each individual, each gene locus is mutated with a probability of 0.06, randomly changing the functional type of the basic unit. After mutation, the danger region constraint is also checked: if the mutated functional partition contains a danger region, the mutation is rolled back.

[0111] The algorithm terminates after 200 iterations, the number of which is chosen based on preliminary experiments. The final output is a set of Pareto front solutions, representing multiple non-dominant optimal solutions. Each solution is accompanied by specific values ​​for four fitness functions, allowing users to select one or more solutions from the set for further refinement based on the specific project requirements. This module outputs at least five Pareto optimal solutions.

[0112] In this embodiment, the dynamic simulation and visualization module will be described in detail:

[0113] The dynamic simulation and visualization module generates a 3D comparison animation of the industrial building before and after the functional transformation, as well as an immersive virtual reality-based interactive interface, based on the optimized revitalization design scheme.

[0114] This module first loads a 3D model with historical appearance textures (from the historical texture mapping module) as the pre-conversion scene. Then, it loads the 3D spatial blocks from the optimized revitalization design scheme, applies the materials and lighting of the new functional areas to the corresponding spatial blocks, and generates the post-conversion scene. The animation shows the transition between the two through a gradual fade-out, with the original industrial building appearance fading and the new spatial blocks emerging with transparency decreasing from 1 to 0. Simultaneously, the retained parts of the structural components remain highlighted, while the demolished parts flicker and disappear. The animation duration is set to 30 seconds.

[0115] This module supports user interaction with head-mounted displays. Wearing VR headsets, users can freely walk inside the optimized building from a first-person perspective. The system calculates shadows, reflections, and collision detection (to prevent users from passing through walls) in real time based on the user's viewpoint. During the walkthrough, users can press buttons on the controllers to view information about the current functional area. The interface also provides a timeline slider, allowing users to observe the building's appearance changes over different historical periods.

[0116] The interactive editing module allows users to directly select 3D spatial blocks within the immersive walkthrough interface. They can then adjust the block's position (0.1m movement step), size (0.05m zoom step), or functional attributes (selecting from drop-down menus for office / commercial / exhibition / cultural and creative industries) using a virtual reality controller or stylus. The modified parameters are fed back to the revitalization design optimization module in real time, triggering recalculation and dynamically updating 3D comparison animations and collision detection results. The system response latency for real-time feedback is less than 100ms.

[0117] In this embodiment, the historical texture mapping module is described in detail:

[0118] The historical texture mapping module receives historical image data, extracts feature points from the images using the Scale Invariant Feature Transform (SIFT) algorithm, and matches them with the corresponding spatial locations in the original 3D point cloud data. The successfully matched image textures are then mapped onto the corresponding surfaces of the 3D base model through perspective projection, generating a 3D model with historical appearance textures. This model is used to display the appearance comparison before and after activation in the dynamic simulation and visualization module.

[0119] This example illustrates the SIFT feature extraction steps, specifically:

[0120] The first step is to perform Gaussian convolution on the input historical image (after grayscale) to generate pyramids of different scales, with each octave having 4 scales.

[0121] The second step is to compare each pixel with the surrounding 26 pixels (8 in the same layer + 9 in the upper and lower layers) in the adjacent scale space to find local maxima or minima as candidate keypoints.

[0122] The third step is to remove low-contrast points by fitting a three-dimensional quadratic function, with the contrast threshold set to 0.04.

[0123] The fourth step is to calculate the trace and determinant of the Hessian matrix, set the ratio (principal curvature ratio) edge threshold to 10, and remove edge response points.

[0124] The fifth step is to calculate the gradient direction histogram of the neighborhood of key points (36 bins), with the peak direction as the main direction.

[0125] The sixth step involves dividing the 16×16 neighborhood centered on the keypoint into 4×4 sub-blocks. For each sub-block, an 8-directional gradient histogram is calculated, forming a 128-dimensional feature vector. The nearest neighbor distance ratio threshold is set to 0.75 for matching verification.

[0126] SIFT feature points extracted from historical images are matched with 2D feature points obtained by rendering projection from 3D point clouds. The number of iterations of the random sampling consensus algorithm is set to 2000. Each time, 4 sets of matching point pairs are randomly selected to calculate the homography matrix H. The error tolerance is set to 5 pixels. Finally, H with the most interior points is selected as the optimal transformation to establish the mapping relationship between the image and the 3D model.

[0127] Based on the homography matrix obtained from RANSAC and the camera intrinsic parameters, each pixel in the image is mapped to a corresponding surface in the 3D point cloud space. Surface orientation must be considered during mapping: only surfaces with a normal vector angle less than 90° to the viewing direction will accept texture. This invention uses multiple images to perform texture fusion on the same surface, selecting the median color to avoid overexposure or shadow areas.

[0128] In this embodiment, the bill of quantities generation module will be described in detail:

[0129] The bill of quantities generation module extracts the geometric dimensions and material properties of components to be retained, demolished, and added from the 3D basic model based on the optimized revitalization design scheme. According to the "Construction Engineering Quantity List Pricing Specification" GB 50500-2013 and the "Building and Decoration Engineering Quantity Calculation Specification" GB 50854-2013, it automatically calculates the quantities of concrete (classified by strength grade C25 and C30), steel (classified by section steel and reinforcing steel), masonry (classified by brick type and mortar strength), and doors and windows (classified by material and opening method). The calculation accuracy is retained to two decimal places (unit: m). 3 , t, m 2 The calculation logic is as follows:

[0130] For components to be retained, this module directly calculates their volume or area from the components marked as "retained" in the original model.

[0131] For components to be demolished, extract their geometric dimensions and calculate the demolition volume based on the components marked "demolished" in the revitalization design scheme. The amount of waste generated after demolition is not calculated; only the physical quantity of the demolition work is calculated.

[0132] For newly added components, the calculation of new walls, floors, stairs, doors, and windows corresponding to the new functional area blocks shall be based on the standard calculation rules. New doors and windows shall be calculated by number or area (m²). 2 ) Classification statistics.

[0133] The module automatically summarizes the project quantities into Table 1 format (example):

[0134] Table 1 Example of Bill of Quantities Generation Module

[0135] Sub-projects Strength / Material Work volume Remark <![CDATA[Demolition of concrete components (m 3 )]]> C25 45.30 Original equipment foundation <![CDATA[Demolition of masonry (m 3 )]]> MU10 brick 120.50 Non-load-bearing partition wall <![CDATA[Retain concrete column (m 3 )]]> C30 28.80 24 pieces <![CDATA[New concrete floor slab (m 3 )]]> C30 86.40 120mm thickness <![CDATA[New light steel keel partition wall (m 2 )]]> - 345.00 Double-sided gypsum board New steel beams (t) Q235 4.56 Hot-rolled H-beams <![CDATA[New broken bridge aluminum alloy window (m 2 )]]> Insulating glass 128.00 Flat opening

[0136] This invention provides a comparative example:

[0137] This example selects three industrial buildings as test cases, specifically:

[0138] Case 1 is a single-story, frame-structured textile workshop built in the 1950s (column grid 6m × 9m, building area 4800m²). 2 Case 2 is a multi-story frame structure machine shop built in the 1970s (column grid 6m×6m, 3 stories, total building area 7200m²). 2 Case 3 is a brick-concrete warehouse built in the 1980s (column grid 4.5m × 4.5m, single-story, area 2000m²). 2 Each case study employed traditional manual design methods, a control system using only parametric generation modules (without multi-objective optimization), and the present invention for functional conversion design. The conversion objective was a combined exhibition and cultural and creative function, with capacity indicators of 40% exhibition area, 30% cultural and creative studios, 20% commercial facilities, and 10% office space. The various testing indicators and experimental results are shown in Table 2.

[0139] Table 2 Performance comparison of different methods in three cases

[0140] index Case 1 Case 2 Case 3 Design cycle (days) for manual methods 45 62 28 Compare with the system design cycle (days). 7 9 5 System design cycle (hours) 2.5 3.2 1.8 Original feature retention rate (F1) of manual methods 0.62 0.55 0.71 Comparison System F1 0.58 0.52 0.68 This system F1 0.86 0.79 0.89 Cost of structural intervention using artificial methods (ten thousand yuan) 86.2 112.5 34.7 Cost of intervention in the comparative system structure (ten thousand yuan) 79.5 105.3 31.2 Cost of system structural intervention (ten thousand yuan) 52.3 68.7 22.6 Functional fit of artificial methods (F3) 0.71 0.68 0.74 Comparison System F3 0.69 0.65 0.73 This system F3 0.91 0.88 0.93 <![CDATA[Annual energy consumption by artificial method (kWh / m 2 ·a)]]> 95.2 102.3 88.6 annual energy consumption of the comparison system 92.7 98.5 85.4 Annual energy consumption of this system 78.4 82.1 75.3

[0141] As shown in Table 2, the design cycle of this invention is shortened from several weeks to several hours, improving efficiency by more than 50 times. The average original feature retention (F1) reaches 0.85, which is 35% higher than the 0.63 of the manual method. This is because the invention directly incorporates the number of industrial relic point clouds into the optimization objective and uses semantic segmentation to accurately identify the objects to be retained. The average cost of structural intervention is reduced by 38%. This is because the multi-objective genetic algorithm can actively select layout schemes that avoid dangerous areas, while manual design is forced to reinforce dangerous areas due to poor communication. The average functional fit is improved by 28%. The combination of spatial syntactic analysis and the streamline deviation penalty of the genetic algorithm makes the topological distance of the final scheme very close to the ideal streamline. The average annual energy consumption per unit area is reduced by 16%. Since energy consumption is an additional optimization objective, the algorithm will prioritize combinations with good orientation and minimal impact from the envelope structure, while the control system does not consider energy consumption, so the improvement is limited.

[0142] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A three-dimensional simulation and revitalization design system for the functional transformation of industrial buildings, characterized in that, include: The data acquisition module is used to acquire the original 3D point cloud data, historical image data and structural parameters of industrial buildings; The semantic segmentation and reconstruction module performs semantic segmentation on the original 3D point cloud data based on deep neural networks, identifies the structural components and non-structural elements of the building, and generates a 3D basic model with semantic labels. The structural components include columns, beams, walls and roof trusses, while the non-structural elements include doors and windows, equipment foundations and industrial relics. The structural safety assessment module calculates the remaining bearing capacity of each structural component using the finite element method based on structural parameters and semantic segmentation results, and marks the dangerous areas and areas that can be modified. The functional space parameterization generation module receives the target function type and capacity index input by the user, generates multiple candidate functional zoning schemes based on preset spatial layout rules and modifiable areas, and converts each candidate scheme into a three-dimensional spatial block. The target function type includes at least one of office, commercial, exhibition and cultural and creative industries. The activation design optimization module performs collision detection between the three-dimensional spatial blocks corresponding to each candidate functional zoning scheme and the three-dimensional basic model with semantic labels. Combined with the dangerous areas and modifiable areas marked by the structural safety assessment module, the optimization scheme is generated by a multi-objective genetic algorithm with the optimization objectives of maximizing the preservation of original industrial features, minimizing structural intervention and meeting the target functional flow. The dynamic simulation and visualization module generates a 3D comparison animation of the industrial building before and after the functional transformation, as well as an immersive virtual reality-based interactive interface, based on the optimized revitalization design scheme.

2. The three-dimensional simulation and revitalization design system for industrial building function transformation according to claim 1, characterized in that: The data acquisition module includes a lidar unit mounted on a drone and a ground-based fixed 3D laser scanner. The drone lidar is used to acquire point cloud data of the exterior and upper parts of the building, while the ground-based 3D scanner is used to acquire point cloud data of the interior and lower parts of the building. The two types of data are registered and fused using an iterative nearest-point algorithm to form complete original 3D point cloud data.

3. The three-dimensional simulation and revitalization design system for industrial building function transformation according to claim 1, characterized in that: The semantic segmentation and reconstruction module uses PointNet++ or RandLA-Net as deep neural networks. This network takes the coordinates and reflection intensity of each point in the original 3D point cloud data as input, outputs the probability of each point belonging to a structural component or a non-structural element, and further uses a region growing algorithm to cluster points of the same category into independent component instances, thereby generating a 3D basic model with instance-level semantic labels.

4. The three-dimensional simulation and revitalization design system for functional transformation of industrial buildings according to claim 1, characterized in that, The structural safety assessment module is used to perform the following operations: First, extract the geometric contours and spatial positions of each structural component based on the semantic segmentation results. Then, establish a finite element model for each component based on the material strength, cross-sectional dimensions, and reinforcement information in the structural parameters. Use nonlinear static pushover analysis to calculate the remaining bearing capacity of the component under the current load. Mark the area where the remaining bearing capacity is lower than the design specification threshold as a dangerous area, and mark the area where the non-load-bearing partition wall and equipment foundation are located with the remaining bearing capacity higher than the threshold as an area that can be modified.

5. The three-dimensional simulation and revitalization design system for functional transformation of industrial buildings according to claim 1, characterized in that, The preset spatial layout rules in the functional space parameterization generation module are based on the original column grid module of industrial buildings. Specifically, the center distance between two adjacent rows of columns is used as the basic module unit, and the space enclosed by each column span is used as the basic unit. According to the minimum room size, evacuation passage width and net height requirements corresponding to the target functional type, the adjacency relationship between the units is generated by spatial syntax in graph theory. Then, the basic units are combined into a functional zoning scheme that meets the capacity index through a recursive backtracking algorithm.

6. The three-dimensional simulation and revitalization design system for functional transformation of industrial buildings according to claim 1, characterized in that, The multi-objective genetic algorithm used in the activation design optimization module includes the following fitness functions: Original feature retention is calculated as the ratio of the number of industrial relics point clouds that have not been demolished or obscured in the candidate scheme to the total number of original industrial relics point clouds. Structural intervention cost is calculated by multiplying the amount of work required for reinforcement, demolition, or addition of structural components in the hazardous area in the candidate scheme by a preset unit cost coefficient, based on the modifiable area output by the structural safety assessment module. Functional fit, based on spatial syntax computing theory, calculates the reciprocal of the deviation between the topological distance between the main functional spaces in the candidate scheme and the optimal path required by the target functional streamline; The algorithm employs non-dominated sorting and crowding distance selection operators, and outputs the Pareto front solution set as the optimized activation design scheme after a preset algebraic iterations.

7. The three-dimensional simulation and revitalization design system for industrial building function transformation according to claim 1, characterized in that: The dynamic simulation and visualization module also includes an interactive editing module, which allows users to directly select a 3D spatial block in the immersive roaming interface, adjust the position, size or functional attributes of the 3D spatial block, and feed the modified parameters back to the activation design optimization module in real time to trigger re-optimization calculation, while dynamically updating the 3D comparison animation and collision detection results.

8. The three-dimensional simulation and revitalization design system for industrial building function transformation according to claim 1, characterized in that: The system also includes a historical texture mapping module, which receives the historical image data, extracts feature points in the image using a scale-invariant feature transform algorithm, matches them with the corresponding spatial positions in the original 3D point cloud data, and maps the successfully matched image textures onto the corresponding surfaces of the 3D base model through perspective projection to generate a 3D model with historical appearance textures, which is used to display the appearance comparison before and after activation in the dynamic simulation and visualization module.

9. The three-dimensional simulation and revitalization design system for functional transformation of industrial buildings according to claim 1, characterized in that: The system also includes an energy consumption simulation module. Based on the wall, door, window and roof structure information of the functional areas in the optimized revitalization design scheme, combined with local meteorological parameters and indoor environmental settings corresponding to the target function type, the energy consumption simulation module uses a building energy consumption simulation engine to calculate the annual heating, cooling and lighting energy consumption of the scheme. The energy consumption results are fed back to the revitalization design optimization module as an additional optimization objective, so that the multi-objective genetic algorithm can optimize the energy consumption index at the same time.

10. The three-dimensional simulation and revitalization design system for industrial building function transformation according to claim 1, characterized in that: The system also includes a bill of quantities generation module. Based on the optimized revitalization design scheme, the bill of quantities generation module extracts the geometric dimensions and material properties of the components to be retained, the components to be demolished, and the components to be added from the three-dimensional basic model. According to the current national or local building engineering quantity calculation standards, the module calculates the quantities of the sub-items and outputs them as structured tables to assist in the preparation of the construction budget for revitalization and renovation. The sub-items include at least concrete, steel, masonry, and doors and windows.

Images