Intelligent layout system for exterior wall composite board installation

Abstract

This invention discloses an intelligent layout system for the installation of exterior wall composite panels, relating to the field of intelligent construction technology. It collects building facade information to generate a structured digital twin model of the building site; matches the site model with historical construction digital twin models based on feature fingerprint similarity, extracting an initial intelligent layout model from the graph neural network architecture of the matching cases; analyzes the differences between the site model and the historical models and encodes them as adjustment instructions for the model input graph structure; performs transfer learning on the initial model, focusing on optimizing the difference areas, to obtain an optimized intelligent layout model; acquires site environmental data, combines it with a built-in physical and process knowledge base, drives the optimized model to generate multiple candidate schemes and perform construction feasibility simulations, outputting multi-dimensional construction scores for each scheme; compares the multi-dimensional construction scores and outputs the optimal layout scheme, solving the problems of traditional layout methods being unable to handle complex structures, ignoring environmental factors, and lacking reusable experience.

Description

Technical Field

[0001] This invention relates to the field of intelligent construction technology, specifically to an intelligent layout system for the installation of exterior wall composite panels. Background Technology

[0002] In exterior wall composite panel installation projects, layout design is a crucial step affecting material waste, construction efficiency, and the final visual effect. Traditional layout mainly relies on designers' experience and manual arrangement in CAD software, which has the following prominent problems: First, for non-standard building facades that include curved window frames, irregular decorative strips, and various openings, it is difficult for humans to quickly find the optimal solution that meets multiple objectives such as minimizing waste, aesthetically pleasing joints, and facilitating construction, often getting stuck in local optima. Second, some existing automated layout tools are mostly based on rules or simple optimization algorithms, and their models are static, unable to adapt to dynamic changes in different construction sites, seasons, material properties, and worker skill levels, leading to deviations in the plan during on-site implementation, and the first-time success rate of installation decreases with long-term system use. In addition, historically successful layout solutions have not been effectively accumulated and reused, with each project starting almost from scratch, resulting in low efficiency in knowledge transfer.

[0003] Therefore, in order to address the above problems, there is an urgent need for an intelligent layout system for the installation of exterior wall composite panels. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an intelligent layout system for the installation of exterior wall composite panels, which solves the problems of traditional layout methods being unable to handle complex structures, ignoring environmental factors, and lacking the ability to reuse experience.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an intelligent layout system for installing exterior wall composite panels, comprising: a site data acquisition and modeling module based on 3D point cloud data and a pre-trained recognition model, used to acquire building facade information and semantically identify and mark layout constraint points to generate a structured digital twin model of the building site; a historical scheme matching and model extraction module, used to match the site model with historical construction digital twin models based on feature fingerprint similarity, and extract an initial intelligent layout model of the graph neural network architecture of the matching cases; a differential-driven model optimization module, used to analyze the differences between the site model and the historical model and encode them into adjustment instructions for the model input graph structure, and perform transfer learning on the initial model focusing on optimization of the difference areas to obtain an optimized intelligent layout model; a construction site perception and scheme evaluation module, used to acquire site environmental data, combine with a built-in physical and process knowledge base, drive the optimized model to generate multiple candidate schemes and perform construction feasibility deduction, and output multi-dimensional construction scores for each scheme; and an optimal scheme decision and output module, used to compare multi-dimensional construction scores and output the optimal layout scheme.

[0006] Furthermore, the intelligent layout model is a generation-evaluation integrated model based on graph neural networks; the input of the intelligent layout model is an attribute graph converted from a digital twin model, where nodes represent board units or constraint points with attached geometric attributes, and edges represent spatial adjacency or constraint relationships; the core of the intelligent layout model includes a graph encoder, a layout generation head, and an evaluation head; the graph encoder extracts features from the input attribute graph; the layout generation head outputs the spatial segmentation and arrangement sequence of the board; and the evaluation head outputs the construction score of the corresponding layout.

[0007] Furthermore, in the differential-driven model optimization module, the differential quantification is converted into modification of the input graph structure of the initial intelligent typesetting model to generate adaptive training samples; the modification includes: adding or deleting corresponding nodes and connecting edges in the input attribute graph according to the difference of constraint points, and updating node attributes; using the modified graph samples, the graph neural network model is fine-tuned with the auxiliary loss function of minimizing the non-standardization of the board material in the differential region.

[0008] Furthermore, in the construction site perception and scheme evaluation module, the site environment data includes real-time temperature and humidity, daily sunshine intensity forecast, and the average technical level of currently available installation workers; the optimized intelligent layout model integrates a material parameter query interface and a process rule library; the construction feasibility deduction logic includes: querying the material expansion coefficient based on real-time temperature and humidity to determine the virtual compensation amount for the board size; querying the process rule library based on the average technical level of workers to adjust the virtual installation fault tolerance threshold for irregularly shaped boards; and evaluating the predicted values ​​of each candidate scheme on multiple indicators such as material loss, construction complexity, and joint quality based on the above virtual compensation amount and virtual installation fault tolerance threshold, and aggregating them into the construction score.

[0009] Furthermore, the matching logic of the historical scheme matching and model extraction module includes: constructing multi-dimensional feature fingerprints of the building site digital twin model and the historical construction digital twin model, wherein the multi-dimensional feature fingerprints include constraint point type distribution vectors, facade outline shape descriptors, and constraint point spatial relationship graph features; using a metric learning-based method to determine the similarity distance between the current model feature fingerprint and all historical model feature fingerprints in the database in the high-dimensional feature space; and selecting the case corresponding to the historical construction digital twin model with the smallest similarity distance as the closest historical case.

[0010] Furthermore, it also includes an incremental learning module for the construction case database; after each actual construction is completed, the incremental learning module uses the final adopted layout scheme, the corresponding digital twin model of the building site, the actual collected complete dynamic environmental data sequence, and the final acceptance quality evaluation as a feedback data pair; using the acceptance quality evaluation as a monitoring signal, the module uses the feedback data to calibrate the parameters of the optimized intelligent layout model used in this instance; and archives the calibrated model and its associated data into the construction case database for future case matching and model extraction.

[0011] Furthermore, during the construction feasibility study, the optimized intelligent layout model embeds knowledge sub-modules reflecting material physical properties and knowledge sub-modules reflecting construction technology. The knowledge injection and updating method is as follows: the initial rule base of the knowledge sub-module comes from the structured analysis of composite board material manuals and construction specification documents; through the incremental learning module of the construction case database, triple data of construction plans, environmental data, and acceptance results are continuously collected, and the parameter thresholds and rule confidence in the initial rule base are dynamically corrected and enriched by combining rule mining and parameter fitting.

[0012] Furthermore, the on-site data acquisition and modeling module specifically includes: a 3D scanning unit, used to acquire 3D point cloud data of the building facade through LiDAR or photogrammetry technology; a constraint intelligent recognition unit, which loads a pre-trained constraint point recognition model, analyzes the 3D point cloud data, and automatically identifies and semantically marks the boundaries of curved window frames, irregular decorative strips, ventilation openings, and pipe openings; and a digital twin construction unit, which converts the semantically marked 3D point cloud data into a multi-layered structured digital twin model of the building site containing geometric information and constraint semantic information.

[0013] The present invention has the following beneficial effects: This intelligent layout system for installing exterior wall composite panels shortens the solution generation time and reduces the number of modifications, avoiding repeated communication and adjustments, through digital twin modeling, historical case matching, and rapid model optimization. It controls material wastage and improves economic efficiency through global optimization, virtual compensation design, and a penalty mechanism for non-standard panels. Dynamic environmental perception and construction feasibility simulation proactively avoid problems such as joint deformation caused by high temperatures and installation errors due to worker skill mismatch, improving the pass rate compared to traditional construction methods and reducing unreasonable designs of irregularly shaped panels, thus lowering construction safety risks. An incremental learning module transforms the construction experience of each project into model parameters and a rule base; as cases accumulate, the model's adaptability and accuracy continuously improve, further shortening the optimization cycle. It supports different types of exterior wall composite panels, different building facade features, and different construction environments. Through dynamic updates of the knowledge sub-module, it can adapt to new material types and construction specifications, making it widely applicable.

[0014] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0015] Figure 1 This is a structural diagram of an intelligent layout system for installing exterior wall composite panels according to the present invention.

[0016] Figure 2 This is a flowchart of a method for an intelligent layout system for installing exterior wall composite panels according to the present invention. Detailed Implementation

[0017] This application embodiment provides an intelligent layout system for the installation of exterior wall composite panels, which enables intelligent decision-making from data perception to continuous evolution, significantly improving the accuracy, economy, and reliability of exterior wall composite panel installation.

[0018] The overall concept of this application's embodiments is as follows: Based on 3D point cloud data from the construction site, a structured digital twin model is constructed by semantically identifying layout constraints using a pre-trained model. Graph neural networks are used to extract layout logic from historical cases, and the initial model is matched using feature fingerprint similarity. For areas of difference between the on-site and historical models, graph structure adjustment instructions drive transfer learning to achieve local model optimization. Combining environmental sensors and a process knowledge base, the feasibility of solutions under different construction scenarios is simulated, generating multi-dimensional scores. Finally, the optimal layout solution, balancing efficiency, cost, and accuracy, is output through score comparison. This design, through differentiated optimization and knowledge fusion mechanisms, breaks through the dependence of traditional layout methods on fixed templates and enhances intelligent decision-making capabilities in complex scenarios.

[0019] Please see Figure 1 , Figure 2This invention provides a technical solution: an intelligent layout system for installing exterior wall composite panels, comprising: a site data acquisition and modeling module based on 3D point cloud data and a pre-trained recognition model, used to acquire building facade information and semantically identify and mark layout constraint points to generate a structured digital twin model of the building site; a historical scheme matching and model extraction module, used to match the site model with historical construction digital twin models based on feature fingerprint similarity, and extract an initial intelligent layout model of the graph neural network architecture of the matching cases; a differential-driven model optimization module, used to analyze the differences between the site model and the historical model and encode them into adjustment instructions for the model input graph structure, and perform transfer learning on the initial model focusing on optimization of the difference areas to obtain an optimized intelligent layout model; a construction site perception and scheme evaluation module, used to acquire site environmental data, combine with a built-in physical and process knowledge base, drive the optimized model to generate multiple candidate schemes and perform construction feasibility deduction, and output multi-dimensional construction scores for each scheme; and an optimal scheme decision and output module, used to compare multi-dimensional construction scores and output the optimal layout scheme.

[0020] In this implementation plan, the on-site data acquisition and modeling module obtains high-precision 3D point cloud data of the building facade to be constructed using a 3D laser scanner or UAV photogrammetry. These raw point clouds are then input into a pre-trained deep learning recognition model. This model can automatically identify and semantically label various layout constraints in the point cloud, such as the boundaries of curved window frames, the outlines of irregular decorative strips, and the precise locations of ventilation openings and pipe openings. Based on these semantically labeled point clouds, the system constructs a structured digital twin model of the building site. This model is not a simple 3D mesh, but a composite data structure containing three layers of information: Geometry layer: Stores a three-dimensional triangular mesh that describes the geometric shape of the building facade surface, reconstructed from point clouds.

[0021] Semantic layer: Associating semantic labels with each triangular facet or vertex in the geometry layer, such as the main exterior wall, the curved window boundary -1, and the 300mm diameter circular vent. This enables the computer to understand the physical meaning of each region in the diagram.

[0022] Relationship layer: This layer records the spatial topological relationships between semantic regions in the form of an attribute graph. For example, it records that the boundary of the curved window-1 is surrounded by the main exterior wall. This structured digital twin model transforms the complex building facade into standardized, semantic data objects that computers can deeply understand and process, laying a solid foundation for subsequent intelligent analysis.

[0023] The historical scheme matching and model extraction module receives the newly constructed on-site digital twin model and calculates a unique feature fingerprint for it. This fingerprint is a high-level abstraction of the model and consists of three parts: Constraint point type distribution vector: Counts the number of various constraint points in the current facade and converts them into a normalized proportional vector, reflecting the constraint composition of the facade.

[0024] Facade outline shape descriptor: Extract the two-dimensional shape of the outer outline of the building facade, encode it into a fixed-length feature vector through methods such as shape context, and describe the overall shape of the facade.

[0025] Constraint point spatial relationship graph features: Each constraint point is treated as a node in a graph, and the spatial distance and orientation relationship between nodes are used as edges to construct a relationship graph. Then, graph embedding techniques are used to compress this graph into a feature vector representing the spatial layout relationship.

[0026] The system calculates the similarity distance between the fingerprint of the current project and the fingerprints of all historical projects in the construction case database. Each historical case in the database stores its digital twin model fingerprint, the final optimized layout scheme adopted, and a trained initial intelligent layout model. The system retrieves the historical case with the most similar fingerprint and uses its associated intelligent layout model as the design starting point for the current project.

[0027] Since identical building facades are extremely rare, the matched historical model and the new site model will inevitably differ. The core task of the difference-driven model tuning module is to quantify and learn these differences. First, through geometric and semantic comparison, the difference areas are accurately identified, such as the addition of a ventilation opening or an increase in the curvature of existing window frames. These differences are then translated into adjustment instructions for the input data structure of the initial intelligent layout model. The initial intelligent layout model is a generative-evaluation integrated model based on a graph neural network, taking an attribute graph as input. The adjustment instructions modify this graph; for example, adding a ventilation opening node and deleting potential board unit nodes that overlap with it, establishing new constraint edges, thereby generating a series of training samples reflecting the new constraints. Subsequently, the system uses a transfer learning strategy focused on optimizing difference areas for fine-tuning: freezing the graph encoder parameters used for general feature extraction at the model's front end, and only using a small number of newly generated samples containing difference points to update the parameters of the layout generation head and evaluation head at the model's back end. In the training loss function, a penalty term for the regularity of board materials in difference areas is specifically added to guide the model to prioritize learning how to gracefully handle newly added or changed constraints. This process allows for the rapid acquisition of an optimized intelligent layout model that is highly adapted to the current site conditions.

[0028] The construction site perception and scheme evaluation module accesses IoT data (such as temperature and humidity sensors and weather forecast APIs) and project management data (such as worker team technical levels) at the construction site to obtain dynamic site environmental data. This data, along with the building's digital twin model, is input into the optimized intelligent layout model. This model integrates a material parameter query interface and a process rule library. During model runtime, multiple iterative simulations are performed: first, several geometrically feasible candidate layout schemes are generated; then, for each scheme, its performance under the current environment is simulated, outputting a quantitative, multi-dimensional construction score for each candidate scheme. This score integrates predicted values ​​for multiple sub-items such as material utilization rate, joint aesthetics, and estimated work hours.

[0029] The optimal solution decision and output module receives all candidate solutions and their corresponding construction scores, compares and sorts them according to preset comprehensive evaluation rules, and automatically selects the solution with the highest score as the optimal recommendation. Finally, it outputs a complete set of digital construction guidance documents, including a list of board cuttings, a numbered layout diagram, and construction sequence suggestions.

[0030] Specifically, the intelligent layout model is an integrated generation-evaluation model based on graph neural networks. The input of the intelligent layout model is an attribute graph converted from a digital twin model. Nodes represent board units or constraint points with attached geometric attributes, and edges represent spatial adjacency or constraint relationships. The core of the intelligent layout model includes a graph encoder, a layout generation head, and an evaluation head. The graph encoder extracts features from the input attribute graph. The layout generation head outputs the spatial segmentation and arrangement sequence of the board. The evaluation head outputs the construction score of the corresponding layout.

[0031] In this implementation scheme, the structured digital twin model is converted into an attribute graph. Specifically, a fine-grained virtual mesh is overlaid on the two-dimensional projection plane of the facade. Each mesh cell not marked as a constraint region by the semantic layer becomes a potential plate element node, with attributes including center coordinates, area, aspect ratio, and orientation. Each marked constraint region becomes a constraint node, with attributes including constraint type and boundary polygon coordinates. If two nodes are spatially adjacent, an edge is established between them, with edge attributes describing the type of adjacency (e.g., close adjacency, gap required).

[0032] The model employs an end-to-end architecture internally. Graph encoders consist of stacked multi-layer graph attention networks. In each layer, each node focuses on information from its neighbors and merges this information with its own to update its feature vector. After several layers of such iterative propagation, each node's feature vector contains global contextual information about its position within the entire graph structure.

[0033] Layout generation head: Receive the updated features of all nodes. It is essentially a sequence generator that simulates the optimal layout order and determines the final destination of each sheet node in turn: whether to merge multiple small nodes into a standard board or cut them separately into special-shaped boards, and assign a unique installation number to it. Finally, an ordered sheet arrangement sequence is output.

[0034] Evaluation head: Work in parallel with or immediately after the generation head. It receives the global features of the graph and the intermediate layout representation generated by the generation head, and predicts the performance of this layout scheme on multiple objectives through a multi-layer perceptron, and outputs a comprehensive scoring vector.

[0035] Specifically, in the differential-driven model tuning module, the difference is quantified as the modification of the input graph structure of the initial intelligent layout model to generate adaptable training samples; the modifications include: in the input attribute graph, adding or deleting corresponding nodes and connecting edges according to the constraint point differences, and updating the node attributes; using the modified graph samples, taking the minimization of the non-standard degree of the sheets in the difference area as an auxiliary loss function, and fine-tuning the parameters of the graph neural network model.

[0036] In this implementation, the modification of the input graph structure is the key to creating adaptable training samples. The system will compare the semantic layer and the relationship layer of the new and old digital twin models. For example, if a new model adds a circular ventilation opening in the middle of the wall compared to the historical model, the system will perform the following operations: add a constraint node of the circular ventilation opening type at the corresponding position in the attribute graph corresponding to the historical case; traverse and delete all potential sheet unit nodes that have an intersection with the ventilation opening area; establish a rigid non-infringement constraint edge between the newly added ventilation opening node and the surviving sheet unit nodes around it. In this way, a new graph containing new constraints and available for training is generated.

[0037] To guide the model to focus on learning to handle new differences, in the loss function of the fine-tuning training, in addition to the regular global material utilization rate loss, an additional penalty term for the regularity of the sheets in the difference area is added. The calculation logic of this loss is: first, locate the local area affected by the newly added or changed constraint points in the attribute graph, and calculate the non-standard degree of all units finally divided into independent sheets in this area. Specifically, for each sheet in the area, calculate the difference ratio between its area and the area of the standard board, and the deviation degree of its aspect ratio from the aspect ratio of the ideal rectangle. Average these difference values as the penalty term.

[0038] Specifically, in the construction site perception and scheme evaluation module, the on-site environmental data includes real-time temperature and humidity, daily sunshine intensity forecast, and the average skill level of currently available installation workers; the optimized intelligent layout model integrates a material parameter query interface and a process rule library; the construction feasibility deduction logic includes: querying the material expansion coefficient based on real-time temperature and humidity to determine the virtual compensation amount for the board size; querying the process rule library based on the average skill level of workers to adjust the virtual installation fault tolerance threshold for irregularly shaped boards; based on the above virtual compensation amount and virtual installation fault tolerance threshold, evaluating the predicted values ​​of each candidate scheme on multiple indicators such as material loss, construction complexity, and joint quality, and aggregating them into a construction score.

[0039] During the feasibility study process, the optimized intelligent layout model embeds knowledge sub-modules reflecting material physical properties and construction technology. The knowledge injection and updating methods are as follows: the initial rule base of the knowledge sub-modules comes from the structured analysis of composite board material manuals and construction specification documents; through the incremental learning module of the construction case database, triple data of construction plans, environmental data, and acceptance results are continuously collected, and the parameter thresholds and rule confidence in the initial rule base are dynamically corrected and enriched by combining rule mining and parameter fitting.

[0040] In this implementation plan, the material physical properties knowledge submodule originates its initial knowledge from a structured analysis of various composite board product manuals, establishing a lookup table for "material type - temperature - coefficient of thermal expansion". An example of the logical deduction process is as follows: When the model receives information such as "current ambient temperature 35°C, using type A composite board", it first queries the knowledge base and learns that type A board expands by ΔL millimeters per meter at 35°C compared to the standard expansion at 20°C. Therefore, when calculating the gaps between boards in virtual layout, the model automatically subtracts ΔL from the theoretical gap value as a virtual compensation. This simulates the actual impact of thermal expansion on the joints of the boards.

[0041] Construction Technology Knowledge Submodule: The initial rules of this module are derived from construction specifications and expert experience, defining virtual installation fault tolerance thresholds for workers of different skill levels (e.g., junior, intermediate, and senior). For example, the rule might stipulate: "Intermediate workers can reliably install trapezoidal panels with a side length ratio within 1:2; for pentagonal and larger irregular-shaped panels, the probability of installation failure increases by 30%." Logical Deduction Process: For each non-rectangular panel in the scheme, the model calculates its number of sides and interior angle complexity. If its complexity exceeds the threshold corresponding to the current specified worker level, the estimated installation time and estimated installation risk coefficient for that panel are adjusted upwards according to the rules.

[0042] The knowledge submodule is updated as follows: After each project, the incremental learning module of the construction case database collects a triplet of data: construction plan, environmental data, and acceptance results. For example, records might show that at 30°C, following a certain construction plan, a specific joint treatment method received an excellent acceptance rating. Using a large amount of such triplet data, the knowledge base is dynamically corrected and enriched through a combination of rule mining and parameter fitting. For instance, it might discover that the estimated coefficient of thermal expansion for a certain type of board material at high temperatures is too low in the original knowledge base, thus requiring calibration. This makes the system's experience increasingly rich and accurate.

[0043] The system ultimately outputs a comprehensive score. The calculation logic is as follows: for each candidate solution, it calculates the original predicted values ​​for three core indicators: material wastage rate, joint aesthetics, and ease of construction. Then, using a normalization function, these original values ​​are converted into scores between 0 and 1. Finally, based on the specific project requirements (e.g., whether the client prioritizes aesthetics or cost), appropriate weights are assigned to the three indicators, and a weighted average is calculated to obtain the comprehensive score. The weights can be flexibly adjusted by the project manager in the system interface to adapt to different project priorities.

[0044] Specifically, the matching logic of the historical scheme matching and model extraction module includes: constructing multi-dimensional feature fingerprints of the building site digital twin model and the historical construction digital twin model. The multi-dimensional feature fingerprints include constraint point type distribution vectors, facade outline shape descriptors, and constraint point spatial relationship graph features; using a metric learning-based method, determining the similarity distance between the current model feature fingerprint and the feature fingerprints of all historical models in the database in the high-dimensional feature space; and selecting the case corresponding to the historical construction digital twin model with the smallest similarity distance as the closest historical case.

[0045] Specifically, it also includes an incremental learning module for the construction case database. After each actual construction is completed, the incremental learning module uses the final layout scheme, the corresponding digital twin model of the building site, the complete dynamic environmental data sequence collected, and the final acceptance quality evaluation as a feedback data pair. Using the acceptance quality evaluation as a monitoring signal, the module uses the feedback data to calibrate the parameters of the optimized intelligent layout model used in this case. The calibrated model and its associated data are archived into the construction case database for future case matching and model extraction.

[0046] In this implementation plan, after project construction is completed, the system initiates calibration. Using the actual environmental data from construction and the final adopted solution (which may have undergone on-site fine-tuning) as the standard answer, the environmental data is input again into the optimized intelligent layout model used in this project, allowing the model to run again and obtain its post-construction prediction. Next, the system calculates the difference between the post-construction prediction and the on-site fine-tuned solution, for example, comparing the positional deviations on the dividing lines of each board. This deviation value is used as a loss signal, and the model parameters are updated slightly using a backpropagation algorithm.

[0047] The calibrated model, the digital twin model of the project's construction site, the complete sequence of site environmental data, the final layout scheme, and the acceptance quality evaluation will all be packaged into a new case study package. The core index of this case study package is its feature fingerprint, calculated in the historical scheme matching and model extraction module. The system will generate and store this feature fingerprint for the new case study package, and then archive it into the construction case database. In the future, when new projects are matched, the fingerprint of this new case study will also participate in similarity calculations, and the successful experiences it contains may be retrieved and reused.

[0048] Specifically, the on-site data acquisition and modeling module includes: a 3D scanning unit, used to acquire 3D point cloud data of the building facade through LiDAR or photogrammetry technology; a constraint intelligent recognition unit, which loads a pre-trained constraint point recognition model, analyzes the 3D point cloud data, and automatically identifies and semantically marks the boundaries of curved window frames, irregular decorative strips, ventilation openings, and pipe openings; and a digital twin construction unit, which converts the semantically marked 3D point cloud data into a multi-layered structured digital twin model of the building site containing geometric information and constraint semantic information.

[0049] In this implementation plan, the 3D scanning unit uses a ground-based 3D laser scanner or a drone equipped with a lidar and oblique photography module to perform multi-station and multi-angle scanning of the target building facade, obtain complete and high-density 3D point cloud data, and accurately record every geometric detail of the facade.

[0050] The constraint-based intelligent recognition unit loads a pre-trained deep learning model. This model has been trained on massive amounts of labeled building point cloud data and can automatically and accurately classify the input point cloud into categories such as walls, windows, doors, decorative strips, and ventilation louvers, and further identify their precise boundaries.

[0051] The digital twin building block receives point clouds that have been semantically segmented and labeled. First, surface reconstruction is performed on the point clouds of each category to generate corresponding 3D mesh models. Then, the semantic labels output by the recognition model are associated with each mesh facet or vertex. Finally, based on spatial relationships, a topological connection graph between the semantic objects is automatically analyzed and established, thereby outputting the final structured digital twin model.

[0052] In summary, this application has at least the following effects: High-precision acquisition of building facade information and intelligent labeling of constraint points are achieved through 3D point cloud and semantic recognition technologies, significantly improving data acquisition efficiency and accuracy. A historical scheme matching mechanism based on graph neural networks can quickly locate similar cases and extract initial models, avoiding the resource consumption of training from scratch. A differentiation-driven transfer learning strategy focuses on optimizing the differences between the site and historical models, effectively solving the problem of excessive reliance on global data in traditional methods and improving model adaptability. The construction site perception module integrates environmental data and a process knowledge base, achieving quantitative assessment of scheme feasibility through multi-dimensional construction scoring, reducing the subjectivity of human experience judgment. An optimal scheme decision-making mechanism ensures that the output scheme achieves comprehensive optimization in terms of material utilization, construction efficiency, and error control. This system realizes intelligent processing from data acquisition to scheme output, significantly shortening the layout cycle, reducing material waste, and improving layout accuracy and construction reliability in complex building scenarios.

[0053] Those skilled in the art will understand that embodiments of the present invention can be provided as methods. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0054] This invention is described with reference to a flowchart of a method according to embodiments of the invention. It should be understood that the combination of each step in the flowchart can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing device, generate instructions for implementing the process. Figure 1 A device for a function specified in one or more processes.

[0055] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 The function specified in one or more processes.

[0056] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 Steps of a specified function in one or more processes.

[0057] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0058] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. An intelligent layout system for installing exterior wall composite panels, characterized in that, include: The on-site data acquisition and modeling module based on 3D point cloud data and pre-trained recognition model is used to collect building facade information and semantically identify and mark layout constraint points to generate a structured digital twin model of the building site. The historical scheme matching and model extraction module is used to match the on-site model with the historical construction digital twin model based on feature fingerprint similarity, and extract the initial intelligent layout model of the graph neural network architecture of the matching case. The differentiation-driven model tuning module is used to analyze the differences between the on-site model and the historical model and encode them into adjustment instructions for the model input graph structure. It performs transfer learning on the initial model with a focus on optimizing the difference areas to obtain the tuned intelligent layout model. The construction site perception and scheme evaluation module is used to acquire on-site environmental data, combine it with the built-in physical and process knowledge base, drive the optimized model to generate multiple candidate schemes and conduct construction feasibility simulation, and output multi-dimensional construction scores for each scheme. The optimal solution decision and output module is used to compare multi-dimensional construction scores and output the optimal layout solution.

2. The intelligent layout system for installing exterior wall composite panels according to claim 1, characterized in that, The intelligent typesetting model is a generation-evaluation integrated model based on graph neural networks; The input to the intelligent typesetting model is an attribute graph converted from a digital twin model. Nodes represent board units or constraint points and have associated geometric attributes, while edges represent spatial adjacency or constraint relationships. The core of the intelligent layout model includes a graph encoder, a layout generation head, and an evaluation head; the graph encoder extracts features from the input attribute graph; the layout generation head outputs the spatial segmentation and arrangement sequence of the board material; and the evaluation head outputs the construction score of the corresponding layout.

3. The intelligent layout system for installing exterior wall composite panels according to claim 1, characterized in that, In the differential-driven model tuning module, the differential is quantified as a modification to the input graph structure of the initial intelligent typesetting model in order to generate adaptive training samples. The modifications include: adding or deleting corresponding nodes and connecting edges in the input attribute graph based on the differences in constraint points, and updating node attributes; Using the modified graph samples, the parameters of the graph neural network model are fine-tuned by minimizing the degree of non-standardization of the board material in the difference region as the auxiliary loss function.

4. The intelligent layout system for installing exterior wall composite panels according to claim 1, characterized in that, The construction site perception and scheme evaluation module includes on-site environmental data such as real-time temperature and humidity, daily sunshine intensity forecast, and the average technical level of currently available installation workers. The optimized intelligent typesetting model integrates a material parameter query interface and a process rule library. The construction feasibility deduction logic includes: determining the virtual compensation amount for the board size based on the material expansion coefficient queried according to real-time temperature and humidity. Based on the average skill level of workers, the process rule library is consulted to adjust the virtual installation fault tolerance threshold of irregularly shaped panels; Based on the aforementioned virtual compensation amount and virtual installation fault tolerance threshold, the predicted values ​​of each candidate scheme in multiple indicators such as material loss, construction complexity, and joint quality are evaluated and aggregated into the construction score.

5. The intelligent layout system for installing exterior wall composite panels according to claim 1, characterized in that, The matching logic of the historical scheme matching and model extraction module includes: Construct a multi-dimensional feature fingerprint of the building site digital twin model and the historical construction digital twin model. The multi-dimensional feature fingerprint includes constraint point type distribution vector, facade outline shape descriptor, and constraint point spatial relationship graph features. A metric learning-based approach is used to determine the similarity distance between the current model's feature fingerprint and all historical model feature fingerprints in the database in a high-dimensional feature space. The case corresponding to the historical construction digital twin model with the smallest similarity distance is selected as the closest historical case.

6. The intelligent layout system for installing exterior wall composite panels according to claim 1, characterized in that, It also includes an incremental learning module for the construction case database; After each actual construction is completed, the incremental learning module of the construction case database will use the final layout scheme, the corresponding digital twin model of the building site, the complete dynamic environmental data sequence actually collected, and the final acceptance quality evaluation as feedback data pairs. Using the acceptance quality evaluation as a monitoring signal, the parameters of the optimized intelligent typesetting model used in this trial are calibrated using the feedback data. The calibrated model and its associated data are archived into the construction case database for future case matching and model extraction.

7. The intelligent layout system for installing exterior wall composite panels according to claim 4, characterized in that, During the feasibility study process, the optimized intelligent layout model embeds knowledge sub-modules reflecting material physical properties and construction technology. The knowledge injection and updating methods are as follows: The initial rule base of the knowledge submodule is derived from the structured analysis of composite panel material manuals and construction specification documents; The incremental learning module of the construction case database continuously collects triplet data of construction plans, environmental data, and acceptance results. By combining rule mining and parameter fitting, the parameter thresholds and rule confidence in the initial rule base are dynamically corrected and enriched.

8. The intelligent layout system for installing exterior wall composite panels according to claim 1, characterized in that, The on-site data acquisition and modeling module specifically includes: A 3D scanning unit is used to acquire 3D point cloud data of building facades using LiDAR or photogrammetry technology. The constraint intelligent recognition unit loads a pre-trained constraint point recognition model, analyzes the three-dimensional point cloud data, and automatically identifies and semantically marks the boundaries of the curved window frame, irregular decorative strip, ventilation opening, and pipe opening. The digital twin building block transforms semantically labeled 3D point cloud data into a multi-layered structured digital twin model of the building site, containing geometric and constraint semantic information.

Images