Steel bar calculation method and system based on large model of steel bar engineering

Abstract

The invention relates to a reinforcing steel bar calculation method and system based on a reinforcing steel bar engineering large model, and belongs to the technical field of constructional engineering digitization. The method comprises the steps that engineering drawings or parameter data uploaded by a user are received and subjected to standardization processing, and standardized engineering data are generated; based on the standardized engineering data, extracting reinforcing steel bar feature information, and generating a reinforcing steel bar feature data set; according to the steel bar feature data set, dynamically matching related rules from a preset calculation rule base, converting the related rules into mathematical constraint formulas, and generating a constraint condition set; inputting the steel bar feature data set and the constraint condition set into a pre-trained steel bar engineering large model for processing, and calculating a steel bar engineering amount to obtain an optimized steel bar amount result; and mapping the steel bar quantity optimization result to a three-dimensional building model to generate a visual steel bar distribution diagram. According to the method, steel bar calculation can be conveniently, rapidly and accurately carried out.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of building engineering digitization, and in particular to a steel bar calculation method and system based on a steel bar engineering large model. BACKGROUND

[0002] With the advancement of the informationization process of the construction industry, design and construction based on building information modeling (BIM) have gradually become the industry standard. Accurate calculation of steel bar quantities and reasonable optimization of steel bar arrangement are indispensable links in building engineering, and their accuracy directly affects the accuracy of project cost accounting and structural safety (such as insufficient anchorage length, which may cause seismic hazards).

[0003] Currently, in the traditional workflow, steel bar calculation needs to go through multiple intensive manual operations: first, a cost engineer interprets the building structure construction drawings (such as CAD two-dimensional drawings), identifies the geometric dimensions and reinforcement markings of components such as beams, slabs, and columns; second, the length of a single steel bar is manually calculated according to the national building standard design drawing set and the structure design specification, and construction requirements are added; and finally, a steel bar detail table and a material procurement list are generated. However, when dealing with complex engineering drawings, the traditional steel bar calculation method has a high risk of data heterogeneity and manual errors. Engineering drawings are usually in CAD / DWG format, while engineering parameters are often input in the form of Excel tables. The differences in structure and format between these two types of data make steel bar calculation more difficult.

[0004] In practical applications, due to the particularity of steel bar arrangement, different structural units such as beams, columns, and slabs need to be configured according to different rules, and these rules often involve complex specifications and standards. Therefore, the process of steel bar calculation often faces problems such as data standardization, calculation rule matching, and integration with the three-dimensional model of the building. In particular, in large-scale and complex structural building projects, it is a technical challenge to quickly and accurately complete steel bar calculation and ensure that the steel bar arrangement complies with the building design specifications. SUMMARY

[0005] In order to quickly and accurately calculate steel bar quantities, the present application provides a steel bar calculation method and system based on a steel bar engineering large model.

[0006] In a first aspect, the present application provides a steel bar calculation method based on a steel bar engineering large model, which adopts the following technical solution: A steel bar calculation method based on a steel bar engineering large model, the method comprising: receiving user-uploaded engineering drawings or parameter data and performing standardization processing to generate standardized engineering data; Based on the standardized engineering data, extract the steel bar feature information, and generate a steel bar feature data set; According to the steel bar feature data set, dynamically match related rules from a pre-design calculation rule library and convert them into mathematical constraint formulas to generate a constraint condition set; Input the steel bar feature data set and the constraint condition set into a pre-trained steel bar engineering large model for processing, calculate the steel bar engineering quantity, and obtain an optimized steel bar quantity result; Map the optimized steel bar quantity result to a three-dimensional building model to generate a visual steel bar distribution map.

[0007] By adopting the above technical solutions, the intelligent upgrading of steel bar calculation is realized, the standardized processing eliminates the heterogeneity of multi-source data, the steel bar feature data set converts engineering semantics into machine-understandable objects, and the dynamic generation of mathematical constraint conditions ensures the compliance of AI calculation; The steel bar engineering large model integrates historical experience within the rule boundary, and outputs the optimal solution balancing safety and economy; The final three-dimensional visualization mapping reduces the communication cost between design and construction, and restores abstract data to engineering context to assist construction verification and cost decision-making.

[0008] Optionally, the step of generating a steel bar feature data set based on the standardized engineering data includes: Perform spatial segmentation on the standardized engineering data, identify structural units, and obtain a structural unit set; Extract steel bar instances in the structural unit set to generate a steel bar instance set; Based on the steel bar instance set, a steel bar three-dimensional feature tensor is constructed; Based on the spatial position mapping of the structural unit set and the steel bar instance set, a steel bar-structural unit association matrix is generated; Integrate the steel bar three-dimensional feature tensor and the steel bar-structural unit association matrix to generate a steel bar feature data set.

[0009] Optionally, the step of generating a constraint condition set according to the steel bar feature data set, dynamically matching related rules from a pre-design calculation rule library, and converting them into mathematical constraint formulas includes: According to the steel bar feature data set, extract the steel bar attributes in the steel bar three-dimensional feature tensor, and analyze the structural unit types of the steel bar-structural unit association matrix to obtain a steel bar key attribute set; According to the steel bar key attribute set, retrieve associated specification items in the pre-design calculation rule library and match scenario-based rules to obtain an original rule item set; Parse the variables and operators in the original rule item set to generate a basic constraint formula set; Load the construction environment parameters and substitute into the set of basic constraint formulas to obtain a set of instantiated constraint formulas; Integrate the set of instantiated constraint formulas with the preset optimization target to obtain a set of constraint conditions.

[0010] Optionally, the set of basic constraint formulas includes an anchorage length calculation formula, specifically: La=k·d·(fy / ft); Wherein, La is the anchorage length, k is the node coefficient, d is the diameter of the reinforcing steel bar, fy is the yield strength of the reinforcing steel bar, and ft is the tensile strength of the concrete.

[0011] Optionally, the training step of the reinforcing steel engineering large model includes: Collect historical reinforcing steel engineering data sets and load a pre-designed calculation rule library to obtain an original training data set; Standardize and clean the original training data set and label the rules to obtain a standardized labeled data set; Divide the standardized labeled data set into a training set, a validation set, and a test set; Build an initialization model based on the Transformer architecture and configure the input layer, the encoding layer, and the output layer; Input the training set into the initialization model for training, compare the predicted value with the true value through the loss function, and update the model weight based on the back propagation algorithm; Generate a reward signal based on the rule label, and optimize the model decision through the policy gradient to obtain a trained reinforcing steel engineering large model; Verify the reinforcing steel engineering large model based on the validation set, evaluate the performance of the reinforcing steel engineering large model, and adjust the model hyperparameters according to the verification result; Test the prediction ability of the adjusted reinforcing steel engineering large model based on the test set to obtain the reinforcing steel engineering large model.

[0012] Optionally, the reinforcing steel calculation method further includes: Receive user feedback data and new engineering data, analyze and generate a to-be-corrected data set and a new feature data set; Clean and aggregate the to-be-corrected data set and the new feature data set to generate a standardized incremental training data set; Based on the standardized incremental training data set, freeze the bottom layer parameters of the reinforcing steel engineering large model, and fine-tune the output layer weight parameters; Verify the accuracy of the updated output layer weight parameters, and generate a rule change instruction set by backstepping decision logic; Based on the rule change instruction set, modify the parameters or formulas in the pre-designed calculation rule library.

[0013] In a second aspect, the present application provides a steel bar quantity calculation system based on a steel bar engineering large model, which adopts the following technical solution: A steel bar quantity calculation system based on a steel bar engineering large model, the system comprising: An engineering data receiving module for receiving engineering drawings or parameter data uploaded by a user and performing standardized processing to generate standardized engineering data; A steel bar feature extraction module for extracting steel bar feature information based on the standardized engineering data to generate a steel bar feature data set; A constraint condition generation module for dynamically matching relevant rules from a pre-design calculation rule library and converting them into mathematical constraint formulas based on the steel bar feature data set to generate a constraint condition set; A steel bar quantity calculation module for inputting the steel bar feature data set and the constraint condition set into a pre-trained steel bar engineering large model for processing to calculate steel bar quantities and obtain an optimized steel bar quantity result; A steel bar distribution map generation module for mapping the optimized steel bar quantity result to a three-dimensional building model to generate a visual steel bar distribution map.

[0014] In a third aspect, the present application provides a computer device, which adopts the following technical solution: A computer device comprising a memory, a processor, and a computer program stored on the memory, wherein the processor executes the computer program to implement the steps of the method according to the first aspect.

[0015] In a fourth aspect, the present application provides a computer-readable storage medium, which adopts the following technical solution: A computer-readable storage medium storing a computer program capable of being loaded and executed by a processor to implement any one of the methods according to the first aspect. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 is a first flowchart of a steel bar quantity calculation method based on a steel bar engineering large model according to an embodiment of the present application.

[0017] Figure 2 is a second flowchart of a steel bar quantity calculation method based on a steel bar engineering large model according to an embodiment of the present application.

[0018] Figure 3 is a third flowchart of a steel bar quantity calculation method based on a steel bar engineering large model according to an embodiment of the present application.

[0019] Figure 4 is a fourth flowchart of a steel bar quantity calculation method based on a steel bar engineering large model according to an embodiment of the present application.

[0020] Figure 5 FIG. 5 is a fifth flow diagram of a steel bar calculation method based on a large steel bar engineering model according to an embodiment of the present application. DETAILED DESCRIPTION

[0021] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and should not be used to limit the present application. Figures 1-5 In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and should not be used to limit the present application.

[0022] The embodiments of the present application disclose a steel bar calculation method based on a large steel bar engineering model.

[0023] With reference to Figure 1 The steel bar calculation method based on the large steel bar engineering model comprises the following steps. In step S101, the engineering drawing or parameter data uploaded by a user is received and standardized to generate standardized engineering data. Since there are data structure differences between the engineering drawing (such as CAD / DWG format) and the manually input parameter table (such as Excel), the key target of the standardization processing is to convert these multi-source heterogeneous data into a unified and computer-analyzable format, while ensuring data legality.

[0024] Specifically, the local coordinate system in the drawing (for example, taking the lower left corner of the drawing as the origin) is converted into a global three-dimensional coordinate system (such as the absolute coordinates in the building project), for example, the starting point coordinates (x1, y1) of a certain beam in the drawing are converted into the actual engineering coordinates (X=1250mm, Y=3400mm, Z=-500mm) to align with the component coordinates in the BIM model. Furthermore, the text description (such as “HRB400 grade steel bar”) in the parameter table can be encoded into a numerical parameter (such as the yield strength f_y=400MPa of the steel bar), and the illegal input is filtered according to the specification threshold in the calculation rule library (such as the minimum diameter d≥6mm specified in GB1499.2). For example, if the user inputs the steel bar diameter as 5mm, the system will mark it as abnormal data and refuse to process.

[0025] It can be understood that by encapsulating the cleaned data into a standardized JSON or binary format, for example, generating a table structure to store the size, material grade, spatial position and other information of the component, it is ensured that the subsequent feature extraction module can be directly called.

[0026] In step S102, the steel bar feature information is extracted based on the standardized engineering data to generate a steel bar feature data set. Specifically, the attribution area of the steel bar is divided according to the structural unit boundary (such as the beam-column joint) in the standardized data. For example, the 16G101-1 atlas stipulates that the anchoring area of the beam-column joint needs to be handled separately, and therefore the joint coordinate range (such as x∈[1000,1200], y∈[3000,3200]) needs to be identified. By converting the physical properties of the steel bar into parameters that can be calculated by the model, such as the bending angle θ (which determines the length of the cut-out) and the spacing between adjacent steel bars (which affects the arrangement rules), these features directly affect the accuracy of subsequent rule matching.

[0027] The steel bar feature data set includes a tensor set of geometric features (coordinates, angles), material features (strength grade f_y), and topological features (association relationship with the structural unit). For example, in the column member, the ring arrangement mode of the stirrup is identified, the stirrup diameter d, the spacing s, and the closure angle 360° are extracted, and the feature subset of the member is constructed.

[0028] In step S103, relevant rules are dynamically matched from the pre-design calculation rule library according to the steel bar feature data set and converted into mathematical constraint formulas to generate a constraint condition set; Specifically, based on the engineering parameters in the feature data (such as the structural type being “seismic frame beam”), the associated entries in the rule library are triggered. For example, “seismic grade = III level” in the feature data set automatically matches the clause of the anchoring length coefficient k = 1.15 in GB50011, and the concrete strength grade C30 corresponds to the tensile strength f_t = 2.39 MPa. Further, the text rules are converted into symbolic mathematical expressions. For example, “anchoring length ≥ laE” in the specification is converted into the inequality L_a ≥ k·d·f_y / f_t, where the parameters k, f_t, etc. are injected by dynamic lookup table. This process needs to handle the conditional branches in the rules, for example, when the steel bar diameter d > 25 mm, the correction coefficient k needs to be multiplied by 1.1. Finally, the converted formula is packaged as a structured object, including the formula type (equation / inequality), the variable list (such as d, f_y), and the calculation priority.

[0029] For example, for the distributed steel bars (diameter d = 10 mm, spacing s marked as 200 mm) of the plate member, the system matches the “plate reinforcement spacing shall not be greater than 200 mm, and shall not be greater than 15d” rule of GB50010 to generate the constraint formula s≤min(200mm,150mm)=150mm.

[0030] In step S104, the steel bar feature data set and the constraint condition set are input into the pre-trained steel bar engineering large model for processing to calculate the steel bar quantity and obtain the optimized steel bar quantity result. Specifically, the steel bar engineering large model is a neural network based on a Transformer architecture, the input layer receives a feature tensor, and the output layer generates the length, weight, and arrangement coordinates of the steel bars. The model takes a set of constraint conditions as the calculation boundary, such as taking the anchoring length formula L_a≥1.15*d*f_y / f_t as a penalty term of the loss function during back propagation to exclude solutions that do not meet the specifications.

[0031] It can be understood that the model learns the economic strategy in the historical engineering data through the pre-training process (supervised learning + reinforcement learning). For example, under the condition of meeting the minimum reinforcement ratio, the arrangement scheme with the minimum total length of the downed steel bars is selected (to reduce waste), rather than a conservative scheme that simply meets the safety factor. The network architecture of the model (such as the Transformer) can capture the spatial correlation between the steel bars (such as the spacing dependency relationship of adjacent stirrups). Moreover, the model needs to optimize multiple objectives at the same time, such as minimizing the amount of steel bars and the construction difficulty (such as reducing the number of joints) while ensuring the safety of the structure.

[0032] In step S105, the optimized steel bar amount result is mapped to the three-dimensional building model to generate a visual steel bar distribution diagram.

[0033] In the optimized steel bar amount result, the coordinates of the steel bars (such as the start and end points (x1, y1, z1) and (x2, y2, z2) of the longitudinal reinforcement of the beam) need to be accurately matched with the component coordinate system in the BIM model. For example, the end point coordinates of the steel bars are bound to the beam family instance of the Revit model to ensure the geometric accuracy of the three-dimensional visualization.

[0034] Specifically, differentiated rendering rules are defined based on the type of steel bars (such as longitudinal reinforcement, stirrup, and distribution reinforcement). For example, red spiral lines are used to represent stirrups, blue straight lines are used to represent longitudinal reinforcement, and high-light warnings are triggered at steel bar collision points (such as intersections with pipelines). At the same time, clicking on any steel bar can display detailed attributes (such as length, weight, and belonging component). By binding the steel bar engineering quantity data (such as total weight and sub-item usage) with three-dimensional geometric information, an interactive lightweight model is generated. For example, in the model displayed on the Web, the user can view the steel bar arrangement details in the stirrup encryption area by cutting the beam-column node.

[0035] In the above embodiments, the intelligent upgrading of steel bar calculation is realized, the standardized processing eliminates the heterogeneity of multi-source data, the steel bar feature data set converts engineering semantics into machine-understandable objects, and the dynamic generation of mathematical constraint conditions ensures the compliance of AI calculation; the steel bar engineering large model integrates historical experience within the rule boundary to output an optimal solution that balances safety and economy; and the final three-dimensional visualization mapping reduces the communication cost between design and construction, and restores abstract data to an engineering context to assist construction verification and cost decision-making.

[0036] Referring toFigure 2 As an embodiment of step S102, based on the standardized engineering data, the step of extracting the reinforcement feature information and generating the reinforcement feature dataset includes: Step S201, spatially segmenting the standardized engineering data to identify structural units to obtain a set of structural units; The standardized engineering data contains discrete data points (such as reinforcement endpoint coordinates, component contours) of the entire construction project, but lacks overall understanding of the building structure. The essence of spatial segmentation is to identify independent structural units (such as beams, columns, and plates) in the building through computational geometry algorithms. Based on the component contour point set (such as the flange line of the beam, the cross-sectional polygon of the column) in the standardized data, the discrete points are clustered into continuous spatial bodies using the convex hull algorithm or deep learning method (such as PointNet++). For example, the bottom four coordinates of the beam {(x1, y1, z1), (x2, y2, z2), (x3, y3, z3), (x4, y4, z4)} are determined as an independent "beam unit", and its spatial range is defined as the minimum circumscribed cuboid. Further, the clustering results are classified according to building specifications. For example, spatial bodies with a length-width ratio > 3 are labeled as "beam units", and spatial bodies with a length-width ratio ≈ 1 and vertical distribution are labeled as "column units".

[0037] The structural unit refers to a building component entity with independent mechanical function, and is associated with specific specification items in the calculation rule library (such as beam unit matching anchorage length rule, column unit matching stirrup densification rule).

[0038] Step S202, extracting reinforcement instances within the set of structural units to generate a set of reinforcement instances; Specifically, the reinforcement information in the standardized data is initially discrete attributes (such as length, diameter), which need to be associated with independent entities through the membership relationship with the structural unit. According to the spatial relationship between the reinforcement coordinates and the structural unit boundary (such as the ray method to determine whether the point is inside the polygon), the reinforcement is assigned to the unit it belongs to. For example, if the reinforcement endpoint (x, y, z) is located within the three-dimensional boundary box of beam B1, it is marked as an attached reinforcement of B1. At the same time, a unique identifier is established for each independent reinforcement, and its geometric attributes (such as linear reinforcement storing start and end point coordinates (x_s, y_s, z_s), (x_e, y_e, z_e), and stirrup storing closed polygon vertex sequence) are encapsulated.

[0039] The reinforcement instance is a reinforcement entity object with a unique ID, containing fields such as type (longitudinal reinforcement / stirrup / distributed reinforcement), material parameters (f_y), geometric topology (coordinate sequence), etc., such as {rebar_id: "R-01", type: "longitudinal reinforcement", points: [[x1, y1, z1], [x2, y2, z2]], d: 20, f_y: 400}.

[0040] Step S203, based on the reinforcement instance set, the reinforcement three-dimensional feature tensor is constructed; Among them, the feature tensor is the core input of the AI model, and its construction logic depends on the quantification rules of physical properties, which reduces the spatial coordinate sequence of the reinforcement to the calculation features. For example, the straight longitudinal reinforcement can extract the feature vector [length L, horizontal projection angle θ, vertical inclination φ], and the curved stirrup can extract [inner diameter r, bending curvature β, pitch p]. All features are normalized to the interval of 0-1. Then combine the discrete parameters (such as diameter d, strength f_y) with the geometric features. For example, the diameter d=20mm is encoded as a feature value of 0.67 (when d_max=30mm), and the yield strength f_y=400MPa is encoded as 0.8 (when f_y_max=500MPa). The final three-dimensional feature tensor is a floating-point matrix with dimensions [n, k] (n is the number of reinforcement, k is the feature dimension).

[0041] Step S204, based on the spatial position mapping of the structure unit set and the reinforcement instance set, a reinforcement-structure unit association matrix is generated; Specifically, the topological relationship between the building component and the reinforcement is established, and the accurate range control of rule matching is supported. A sparse matrix A[i][j] can be constructed, where the row index i represents the reinforcement instance ID, and the column index j represents the structure unit ID. When the reinforcement instance i belongs to the structure unit j, A[i][j]=1, otherwise 0. At the same time, additional relationship labels are added to special positions (such as beam-column joint). For example, the core zone reinforcement of the beam B1 and the column C1, which passes through the joint, is labeled as A[i][B1]=1, A[i][C1]=1, tag:"node area". The final reinforcement-structure unit association matrix is a binary matrix representing the spatial attribution relationship between the reinforcement and the component, which limits the search range in the rule matching stage (such as beam longitudinal reinforcement only matches the relevant specifications of the beam).

[0042] Step S205, integrate the reinforcement three-dimensional feature tensor and the reinforcement-structure unit association matrix to generate a reinforcement feature dataset.

[0043] Among them, the data is packaged in the input format of graph neural network (GNN), where the feature tensor is the node feature and the association matrix defines the adjacency relationship. For example, each reinforcement is regarded as a graph node, and the structure unit is regarded as a super node, and the association matrix defines the node-super node edge. And add metadata (such as project seismic grade, concrete strength) to the dataset, so that the subsequent rule matching can dynamically load context parameters. The final reinforcement feature dataset contains a composite object of feature tensor (node attribute), association matrix (graph structure), and metadata (environmental variables).

[0044] In the above embodiments, spatial segmentation establishes building component models, providing semantic boundaries for rule matching, reinforcement instantiation ensures the traceability of individual computational properties of each reinforcement, and feature tensors enable the machine computability of physical parameters; the association matrix captures the mechanical role of reinforcement in the structural system, and the final dataset encapsulates both individual characteristics and system topology, enabling AI large models to accurately learn the coupled relationship between rules and experience (such as the collaborative constraints of anchorage length of longitudinal reinforcement of beams and reinforcement densification in joint areas).

[0045] Referring to Figure 3 As an embodiment of step S103, according to the reinforcement feature dataset, the step of dynamically matching relevant rules from the pre-design calculation rule library and converting them into mathematical constraint formulas to generate a constraint condition set includes: Step S301, according to the reinforcement feature dataset, extract the reinforcement attribute in the reinforcement three-dimensional feature tensor, and analyze the structure unit type of the reinforcement-structure unit association matrix to obtain a reinforcement key attribute set; Wherein, each row in the reinforcement three-dimensional feature tensor (for example, a matrix with shape [n, k]) represents the encoded features of a reinforcement, which needs to be restored to physical properties. For example, the i-th row vector [0.67, 0.42, 0.80] may correspond to a diameter d = 20 mm (normalized value 0.67), a horizontal inclination angle θ = 42°, and a yield strength f_y = 400 MPa (normalized value 0.8).

[0046] Further, in the reinforcement-structure unit association matrix (binary matrix A[i][j]), if reinforcement i is associated with structure unit j, then according to the structure unit type (such as beam, column, plate), an engineering semantic label is assigned. For example, when j corresponds to "frame beam", reinforcement i is labeled as "beam longitudinal reinforcement", and when j corresponds to "seismic wall", it is labeled as "edge component reinforcement". The final reinforcement key attribute set contains a composite object set of type labels (such as "column reinforcement"), physical parameters (d, f_y), and structure unit types (such as "Ⅲ grade anti-seismic frame beam").

[0047] Step S302, according to the reinforcement key attribute set, retrieve the associated specification items in the pre-design calculation rule library and match the scenario rules to obtain an original rule item set; Wherein, according to the structure unit type, seismic level, and reinforcement type in the key attribute, the applicable rules are filtered in the calculation rule library (such as SQL database). For example, "Ⅱ grade anti-seismic main beam longitudinal reinforcement" triggers "frame beam longitudinal reinforcement anchorage length should meet laE=1.15la" in "Code for Seismic Design of Buildings", and "beam longitudinal reinforcement bending anchor length ≥0.4laE" in the specification atlas.

[0048] In addition, empirical rules can be loaded in combination with construction scenarios (e.g. winter construction, node area steel bar dense). For example, in the "beam column node core area" scenario, the construction requirement of "hoop reinforcement spacing not greater than 100mm" is added. The final original rule entry set includes the specification and empirical rule list in text format.

[0049] Step S303, parse the variables and operators in the original rule entry set to generate a set of basic constraint formulas; Among them, the engineering parameters in the rule text (such as "d" mapping to steel bar diameter, "f_y" mapping to yield strength) are parsed and associated with the fields in the steel bar key attributes.

[0050] Specifically, the operators in the text (such as "≥" "≤") can be converted into inequalities, and the conditional statements (such as "when …") can be converted into piecewise functions. For example, the rule "hoop reinforcement spacing s≤min(100mm,6d)" is decomposed into two inequalities: s≤100 and s≤6*d.

[0051] In some embodiments, the set of basic constraint formulas includes the anchorage length calculation formula, specifically: L a =k·d·(f y / f t ); Where L a is the anchorage length, k is the node coefficient, d is the steel bar diameter, f y is the steel bar yield strength, and f t is the concrete tensile strength.

[0052] Step S304, load the construction environment parameters and substitute them into the set of basic constraint formulas to obtain a set of instantiated constraint formulas; Among them, the construction environment parameters are read from the engineering global configuration, for example, the concrete tensile strength f_t=2.39MPa (C30 concrete), the seismic grade coefficient α=1.0 (non-seismic area), and the environment category (two categories a control protection layer thickness). Substituting the above parameters into the basic constraint formula, a specific numerical constraint is generated, and a mathematical constraint bound to the actual parameters, i.e. the set of instantiated constraint formulas, is obtained.

[0053] Step S305, integrate the set of instantiated constraint formulas and the preset optimization target to obtain a set of constraint conditions.

[0054] Among them, the priority (such as "minimize total steel bar consumption" "maximize joint spacing uniformity") is selected from the preset target library and converted into mathematical form. For example, the minimization of total consumption can be expressed as the objective function min(ΣL_i), where L_i is the length of each steel bar.

[0055] In addition, the instantiation constraint formula can be packaged with the optimization target as an optimization problem. For example, L_a ≥ 322d is taken as an inequality constraint, and min(∑L_i) is taken as a target function, to construct a complete constraint condition set (including inequality constraints, equality constraints, and optimization problem description of the target function).

[0056] In the above embodiments, the extraction of the key attributes of the steel bar ensures the accuracy of the rule matching (such as the seismic frame beam longitudinal reinforcement only triggering the seismic code provisions), the formula instantiation process realizes the dynamic adaptation of the parameters (such as the automatic correction of the anchoring formula for the change of the concrete strength), and the integration of the optimization target balances the safety and the economy, and realizes the trade-off between the minimum consumption and the construction feasibility.

[0057] With reference to Figure 4 As an embodiment of the steel reinforcement engineering large model, the training step includes: Step S401, collecting a historical steel reinforcement engineering data set and loading a pre-designed calculation rule library to obtain an original training data set; The historical steel reinforcement engineering data set is derived from actual projects (such as a BIM model library and construction cutting list), and contains original records such as steel reinforcement specifications (diameter d, strength f_y), spatial coordinates, and consumption results. The calculation rule library is structured to store industry standards (such as GB50010 Concrete Structure Design Specification) and drawing construction requirements (such as 16G101 node detail drawing).

[0058] Specifically, the steel reinforcement instances in the historical data are mapped to the rule library entries, and economic strategies not explicitly defined in the rule library are refined from the construction records. The final original training data set contains a heterogeneous data pool of numerical engineering parameters (such as coordinates and dimensions), text specification entries, and economic indicators.

[0059] Step S402, standardizing and cleaning the original training data set and labeling the rule labels to obtain a standardized labeled data set; The standardization and cleaning includes eliminating illegal data (such as diameter d = 5mm not conforming to GB1499.2 standard), unifying dimensions (converting inch units to millimeters), and normalizing coordinates (scaling absolute coordinates to the [0, 1] interval to eliminate project size differences). Rule label labeling includes explicit rule labels (breaking down text specifications into <condition, action> pairs) and implicit experience labels (generating labels by statistical optimization patterns in historical data).

[0060] Step S403, dividing the standardized labeled data set into a training set, a validation set, and a test set; Among them, the training set can account for 70%, covering common structural types (beams, plates, columns) and regular parameter ranges, for basic rule learning; the validation set can account for 15%, containing boundary scenarios (such as super-long span beams, dense node areas), for tuning the sensitivity of the model to complex conditions; the test set can account for 15%, introducing new project data (such as special-shaped structures) that did not participate in training, to evaluate the industrial-level generalization ability of the model.

[0061] Step S404, an initialization model based on the Transformer architecture is constructed, and the input layer, the encoding layer and the output layer are configured; Among them, the input layer includes geometric feature input and rule label input, the geometric feature input includes a sequence of reinforcement coordinates (such as [x1, y1, z1, x2, y2, z2,...]), which retains the spatial relationship through position encoding; the rule label input is to embed the annotated <condition, action> pair into a 256-dimensional vector. In the configuration process of the encoding layer, the multi-head attention mechanism of the Transformer captures the mechanical correlation between the reinforcements. For example, the spacing constraint between adjacent stirrups is automatically identified; at the same time, the residual connection prevents the degradation of the deep network, ensuring the effective propagation of complex rule chains (such as "node area → seismic grade → anchorage length"). The output layer includes regression output and decision output, the regression output includes reinforcement usage (length L, weight W), and the decision output includes arrangement scheme score (economy, constructability).

[0062] Step S405, the training set is input to the initialization model for training, and the predicted value is compared with the true value through the loss function, and the model weight is updated based on the back propagation algorithm; Specifically, the loss function design includes basic loss and rule violation penalty, the basic loss is the mean square error (MSE) of the predicted usage (L_pred) and the true value (L_true); the rule violation penalty refers to the additional penalty term λ·max(0, laE-L_a) if the output scheme violates the annotated rule (such as L_a<laE). The back propagation mechanism updates the weight through gradient descent (such as Adam optimizer) to reduce the usage error while meeting the hard constraint. For example, the gradient value of the scheme with insufficient anchorage length increases in a certain iteration, driving the model to adjust the weight to meet L_a≥laE.

[0063] Step S406, generate a reward signal based on the rule label, and optimize the model decision through policy gradient to obtain a trained reinforcement engineering large model; Specifically, the reward signal includes explicit rule reward and implicit economic reward, the explicit rule reward is compliance output reward + R_rule (such as anchorage length meeting the standard), and the rule output penalty is -P_rule; the implicit economic reward can be set according to historical experience (such as total material usage being 10% lower than the average value + R_economy).

[0064] In addition, the policy gradient optimization can employ a PPO (Proximal Policy Optimization) algorithm to adjust the decision-making policy guided by the reward signal. For example, when the model attempts a solution of "reducing the number of joints but increasing the length of a single rod", if historical data proves that this solution reduces construction costs, the probability of this decision is increased through policy gradient.

[0065] Step S407, verify the reinforcement engineering large model based on the verification set, evaluate the performance of the reinforcement engineering large model and adjust the model hyperparameters according to the verification results; Among them, the key evaluation indicators include rule compliance rate and economic gain, and the rule compliance rate is the proportion of output schemes meeting the specifications (target ≥ 98%); the economic gain represents the percentage of savings compared to the historical average material consumption (target ≥ 8%). The hyperparameter adjustment strategy is, for example: if the verification set rate is low, increase the rule penalty coefficient λ; if the economic gain is insufficient, increase the economic reward weight.

[0066] Step S408, test the prediction ability of the adjusted reinforcement engineering large model based on the test set, and obtain the reinforcement engineering large model.

[0067] Among them, the test set contains untrained structure types, and the model capability is tested by introducing noise data (such as fuzzy labeled drawings). The acceptance criteria can be: the key indicators (compliance rate, economy) fluctuate by ≤5% compared to the verification set; and still be able to output safe schemes under extreme conditions (such as concrete strength C15).

[0068] In the above embodiments, the industry specifications are converted into machine-optimizable objective functions, the policy gradient mechanism gives the model the ability to learn economic strategies, and the long-distance dependency processing characteristics of the Transformer architecture accurately capture the mechanical correlation of the reinforcement system, ensuring the model's generalization ability from common scenarios to complex boundaries, and improving the efficiency and accuracy of reinforcement calculation.

[0069] Reference Figure 5 As a further embodiment of the reinforcement calculation method, it further comprises: Step S501, receive user feedback data and new engineering data, and parse to generate a to-be-corrected data set and a new feature data set; Specifically, a closed-loop feedback mechanism is established between actual engineering applications and AI models. User feedback data contains two types of key information: one is the correction record of calculation results (such as manual adjustment of anchor length value by construction units), and the other is the rule deviation scene marked in the operation log (such as "node area stirrup spacing violates GB50011 Article 6.3.3"). New engineering data refers to the complete data package of new projects (such as BIM models of special-shaped structures).

[0070] Specifically, the system identifies the engineering semantics in the feedback through a semantic parsing engine: for example, parsing "insufficient anchorage at beam end" as the error type ERROR_ANCHOR_LENGTH and associating a specific reinforcement instance ID; identifying "spatial curved surface reinforcement net" in the new project as the feature type FEAT_CURVED_SURFACE. This parsing converts the original discrete information into structured data objects, forming a corrected data set (recording model decision defects) and a new feature data set (extending the model cognitive boundary), providing precise target points for subsequent optimization.

[0071] Step S502, clean and aggregate the corrected data set and the new feature data set to generate a standardized incremental training data set; Wherein, the key technical principle of solving the quality alignment problem of incremental data is the cross-scene data same distribution constraint. The cleaning operation first eliminates invalid feedback (such as corrections marked as "user misoperation"), and then detects data bias through adversarial validation (Adversarial Validation): for example, if the feature distribution of the new special-shaped structure data differs from the original training set by more than a threshold (KL divergence > 0.1), feature remapping is started, and the adversarial generation network (GAN) is used to project the curved reinforcement features of the special-shaped structure to the original feature space.

[0072] And the aggregation process uses priority weighted fusion: rule violation type feedback is given 3 times weight, and experience optimization type feedback is given 1.5 times weight, to ensure that key problems are optimized first. Finally, a standardized incremental data set is generated, maintaining the same tensor dimension and normalized scale (such as all coordinates normalized to the [0, 1] interval) as the initial training set, so that the model can seamlessly access incremental learning.

[0073] Step S503, based on the standardized incremental training data set, freeze the bottom layer parameters of the reinforcement engineering large model, and fine-tune the output layer weight parameters; Wherein, since the basic rules of reinforcement engineering (such as minimum anchorage length, hoop reinforcement densification area requirements) have strong stability, freezing the Transformer encoding layer (including 12 layers of self-attention modules) can maintain the learned core knowledge (such as reinforcement stress transfer law). Fine-tuning focuses on the fully connected network of the output layer: using a selective parameter update algorithm, only the output nodes strongly related to new features (such as curved reinforcement usage prediction layer) are updated.

[0074] The specific process is as follows: first, through correlation analysis (such as calculating the mutual information between features and output nodes), the weight subset to be updated is selected; then fine-tuning training is performed with a small learning rate (η = 0.0001), so that the model adapts to incremental features (such as curved reinforcement arrangement rules) without destroying the existing knowledge framework.

[0075] Step S504, accuracy verification is performed on the updated output layer weight parameters, and a decision logic generation rule change instruction set is backstepped; Among them, the accuracy verification adopts three-order verification: (1) compliance test: input test cases with known rule boundary (such as the limit anchorage length of C20 concrete beam); (2) generalization test: use feature perturbation data (such as ±10% reinforcement coordinate offset); (3) economic verification: compare the reinforcement consumption index before and after optimization.

[0076] In addition, the decision logic backstepping relies on class activation mapping technology, which locates the decision basis by calculating the gradient contribution of the output layer weight to the input feature. For example, if the weight of a certain node significantly increases the curved reinforcement length prediction value, the rule change requirement "when the bending curvature β>60°, the minimum lap length increases by 1.2 times" is backstepped, and is coded as a structured rule change instruction set.

[0077] Step S505, based on the rule change instruction set, modify the parameters or formulas in the pre-design calculation rule library.

[0078] Among them, the rule change execution adopts a double-channel mechanism, the explicit rule update includes the parameter / formula change defined in the instruction set (such as correcting the anchorage length coefficient), directly modifying the SQL record of the rule library; the implicit rule generation includes the backstepped empirical strategy (such as curved reinforcement lap optimization), which is converted into a rule with confidence ("when curvature>0.05, the lap length is recommended to increase [15%-20%], confidence 92.3%").

[0079] In the above embodiment, user feedback is converted into standardized incremental training data, ensuring that model optimization is based on real engineering scenarios; bottom layer parameter freezing maintains the stability of core rules to prevent forgetting; decision backstepping technology converts "black box" model output into interpretable rule changes to ensure that technical solutions comply with engineering logic, and finally realizes continuous evolution through version management of the rule library.

[0080] The application also discloses a reinforcement calculation system based on a reinforcement engineering large model.

[0081] A reinforcement calculation system based on a reinforcement engineering large model, the system comprising: An engineering data receiving module for receiving user-uploaded engineering drawings or parameter data and performing standardized processing to generate standardized engineering data; A reinforcement feature extraction module for extracting reinforcement feature information based on the standardized engineering data to generate a reinforcement feature data set; A constraint condition generation module for dynamically matching related rules from a pre-design calculation rule library based on the reinforcement feature data set and converting them into mathematical constraint formulas to generate a constraint condition set; The steel bar quantity calculation module is configured to input the steel bar feature data set and the constraint condition set into the pre-trained steel bar engineering large model to process and calculate the steel bar quantity, and obtain an optimized steel bar quantity result. The steel bar distribution map generation module is configured to map the optimized steel bar quantity result to the three-dimensional building model to generate a visual steel bar distribution map.

[0082] The steel bar quantity calculation system based on the steel bar engineering large model can implement any of the steel bar quantity calculation methods described above, and the specific working processes of the modules in the steel bar quantity calculation system can refer to the corresponding processes in the method embodiments described above.

[0083] In the several embodiments provided in the present application, it should be understood that the provided method and system can be implemented in other manners. For example, the system embodiments described above are merely schematic; for example, the division of a certain module is merely a logical function division, and there can be another division manner in actual implementation; for example, a plurality of modules or features can be combined or integrated into another system, or some features can be ignored or not executed.

[0084] The present application also discloses a computer device.

[0085] The computer device comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steel bar quantity calculation method based on the steel bar engineering large model when executing the computer program.

[0086] The present application also discloses a computer readable storage medium.

[0087] The computer readable storage medium stores a computer program capable of being loaded and executed by a processor to implement any of the steel bar quantity calculation methods based on the steel bar engineering large model described above.

[0088] The computer readable storage medium can be any tangible medium containing or storing a program, which can be used by or in combination with an instruction execution system, device or apparatus; the program code contained in the computer readable medium can be transmitted by any appropriate medium, including but not limited to wireless, wire, optical cable, RF, etc., or any appropriate combination of the above.

[0089] It should be noted that in the above embodiments, the description of each embodiment has its own emphasis, and the parts not described in detail in a certain embodiment can refer to the related description of other embodiments.

[0090] The above are only preferred embodiments of the present application, not intended to limit the protection scope of the present application, any one feature disclosed in the specification (including the abstract and the drawings) can be replaced by other equivalent or similar features, unless specifically stated otherwise. That is, unless specifically stated, each feature is only an example of a series of equivalent or similar features.

Claims

1. A steel bar calculation method based on a large steel bar engineering model, characterized by, The method comprises: receiving user-uploaded engineering drawings or parameter data and performing standardization processing to generate standardized engineering data; based on the standardized engineering data, extracting reinforcement feature information to generate a reinforcement feature dataset; according to the reinforcement feature dataset, dynamically matching related rules from a pre-design calculation rule library and converting them into mathematical constraint formulas to generate a constraint condition set; inputting the reinforcement feature dataset and the constraint condition set into a pre-trained reinforcement engineering large model for processing to calculate reinforcement quantities and obtain an optimized reinforcement quantity result; mapping the optimized reinforcement quantity result to a three-dimensional building model to generate a visualized reinforcement distribution diagram.

2. The steel calculation method based on the large model of reinforcement engineering according to claim 1, characterized in that, The step of extracting reinforcement feature information based on the standardized engineering data to generate a reinforcement feature dataset comprises: spatially segmenting the standardized engineering data to identify structural units to obtain a structural unit set; extracting reinforcement instances within the structural unit set to generate a reinforcement instance set; based on the reinforcement instance set, constructing a reinforcement three-dimensional feature tensor; based on the spatial position mapping of the structural unit set and the reinforcement instance set, generating a reinforcement-structural unit association matrix; integrating the reinforcement three-dimensional feature tensor and the reinforcement-structural unit association matrix to generate a reinforcement feature dataset.

3. The steel calculation method based on the large model of reinforcement engineering according to claim 2, characterized in that, The step of dynamically matching related rules from a pre-design calculation rule library and converting them into mathematical constraint formulas based on the reinforcement feature dataset to generate a constraint condition set comprises: based on the reinforcement feature dataset, extracting reinforcement attributes in the reinforcement three-dimensional feature tensor and analyzing structural unit types in the reinforcement-structural unit association matrix to obtain a reinforcement key attribute set; based on the reinforcement key attribute set, retrieving associated specification items in the pre-design calculation rule library and matching scenario-based rules to obtain an original rule item set; analyzing variables and operators in the original rule item set to generate a basic constraint formula set; loading construction environment parameters and substituting them into the basic constraint formula set to obtain an instantiated constraint formula set; integrating the instantiated constraint formula set with a pre-set optimization target to obtain a constraint condition set.

4. The steel calculation method based on the large model of reinforcement engineering according to claim 3, characterized in that, The basic constraint formula set includes an anchorage length calculation formula, specifically: L a = k · d · (f y / f t ); wherein L a is the anchorage length, k is the nodal coefficient, d is the diameter of the reinforcement, f y is the yield strength of the reinforcement, f t is the tensile strength of the concrete.

5. The steel calculation method based on the large model of steel reinforcement engineering according to claim 1, characterized in that, The training steps of the reinforcement engineering large model comprise: collecting historical reinforcement engineering datasets and loading pre-design calculation rule libraries to obtain an original training dataset; standardizing and cleaning the original training dataset and labeling rule labels to obtain a standardized labeled dataset; dividing the standardized labeled dataset into a training set, a validation set, and a test set; constructing an initialization model based on a Transformer architecture and configuring an input layer, an encoding layer, and an output layer; inputting the training set into the initialization model for training, comparing predicted values with true values through a loss function, and updating model weights based on a backpropagation algorithm; generating a reward signal based on rule labels and optimizing model decisions through a policy gradient to obtain a trained reinforcement engineering large model; verifying the reinforcement engineering large model based on the validation set, evaluating the performance of the reinforcement engineering large model, and adjusting model hyperparameters according to the verification result; Test the prediction ability of the adjusted steel reinforcement engineering large model based on the test set, to obtain the steel reinforcement engineering large model.

6. The steel calculation method based on the large model of steel engineering according to any one of claims 1 to 5, characterized in that, The steel reinforcement quantity calculation method further comprises: Receiving user feedback data and new engineering data, and analyzing to generate a to-be-corrected data set and a new feature data set; Cleaning and aggregating the to-be-corrected data set and the new feature data set to generate a standardized incremental training data set; Based on the standardized incremental training data set, freezing the bottom layer parameters of the steel reinforcement engineering large model, and fine-tuning the output layer weight parameters; Accuracy verification is performed on the updated output layer weight parameters, and a decision logic generation rule change instruction set is generated by backstepping; Based on the rule change instruction set, modify the parameters or formulas in the pre-design calculation rule library.

7. A steel bar calculation system based on a large steel bar engineering model, characterized by, The system comprises: An engineering data receiving module for receiving user-uploaded engineering drawings or parameter data and performing standardized processing to generate standardized engineering data; A steel reinforcement feature extraction module for extracting steel reinforcement feature information based on the standardized engineering data to generate a steel reinforcement feature data set; A constraint condition generation module for dynamically matching related rules from a pre-design calculation rule library based on the steel reinforcement feature data set and converting them into mathematical constraint formulas to generate a constraint condition set; A steel reinforcement quantity calculation module for inputting the steel reinforcement feature data set and the constraint condition set into a pre-trained steel reinforcement engineering large model for processing to calculate steel reinforcement quantities and obtain an optimized steel reinforcement quantity result; A steel reinforcement distribution map generation module for mapping the optimized steel reinforcement quantity result to a three-dimensional building model to generate a visual steel reinforcement distribution map.

8. A computer device, comprising: A computer program stored on a memory and executable on a processor, wherein the processor implements the method of any one of claims 1 to 6 when executing the program.

9. A computer-readable storage medium, characterized in that: A computer program stored on a memory and executable on a processor, wherein the processor implements the method of any one of claims 1 to 6 when executing the program.

Images