Civil engineering cost management method based on artificial intelligence

Abstract

The application discloses a civil engineering cost management method based on artificial intelligence, and relates to civil engineering cost: a BIM design model of a construction project is constructed; point cloud data of a construction site is collected by laser radar to generate an actual construction point cloud model; the actual construction point cloud model is spatially registered with the BIM design model to obtain a registered fusion model; according to the fusion model, a completed component is identified through a deep learning algorithm, and actual completed quantities are obtained through volume calculation; the actual completed quantities are compared with planned completed quantities of a corresponding time node in the BIM design model to calculate a progress deviation rate; and the construction cost is adjusted according to the progress deviation rate and combined with unit price information in a bill of quantities. In view of low dynamic monitoring precision of civil engineering cost, the application fuses laser point cloud and a BIM model, uses a deep learning algorithm to identify a component completion state and calculate actual quantities, and the like, so that the dynamic monitoring precision of the construction cost is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of engineering cost, in particular to a civil engineering cost management method based on artificial intelligence. BACKGROUND

[0002] Civil engineering projects have the characteristics of large investment scale, long construction period and many participants. Engineering cost management throughout the project life cycle is the key link to ensure project economic benefits. With the transformation of the construction industry towards digitization and intelligentization, the traditional cost management mode has been difficult to meet the needs of fine management of modern engineering projects. Especially in the construction phase, it is of great significance to control project cost and avoid investment risk to accurately grasp the project completion situation in real time and timely adjust the cost plan. The application of building information modeling (BIM) technology provides a new technical means for engineering cost management. By building a three-dimensional digital model containing geometric information, material properties and cost data, automatic extraction of engineering quantities and accurate calculation of cost are realized. However, the BIM model reflects the design intent and planned state. How to compare and analyze the actual construction state with the BIM model to realize dynamic cost management based on actual completion is still a technical problem faced by the industry.

[0003] Traditional methods rely on manual measurement using tools such as tape measures and levels. Not only is it time-consuming and labor-intensive, but the measurement accuracy is also greatly affected by human factors. For high-altitude operations and complex structures that are difficult to reach, there are safety hazards in measurement work. Manual measurement usually uses a sampling method, which is difficult to cover all components, resulting in missing engineering quantity statistics. Existing methods mostly use estimation methods to determine completed engineering quantities, such as rough estimation of completion percentage according to floor completion or reverse calculation of completed quantities based on material consumption. This approach ignores the actual completion form of components, especially for partially completed components, making it impossible to accurately calculate their completed volume. Fixed coefficients are often used for steel reinforcement deduction and masonry joint calculation of reinforced concrete components, which deviates from the actual situation.

[0004] The collection and aggregation of engineering progress data are usually done on a monthly basis, which has obvious lag. When progress deviations are found, the best adjustment opportunity has been missed. There is insufficient correlation analysis between sub-projects, making it difficult to timely identify progress risks that affect the critical path. Progress information is disconnected from cost data, making it impossible to achieve dynamic cost prediction based on actual progress. SUMMARY

[0005] To address the low accuracy of civil engineering cost dynamic monitoring, the present application provides a civil engineering cost management method based on artificial intelligence, which fuses laser point cloud and BIM model, uses deep learning algorithms to identify component completion status and calculate actual engineering quantities, etc., thereby improving the dynamic monitoring accuracy of engineering cost.

[0006] The application provides a civil engineering cost management method based on artificial intelligence, comprising the following steps: S1, constructing a BIM design model of a construction project, wherein the BIM design model comprises component information of a construction stage, a bill of quantities and plan schedule data; S2, collecting point cloud data of a construction site by using a laser radar to generate an actual construction point cloud model; S3, performing spatial registration on the actual construction point cloud model and the BIM design model to obtain a registered fusion model; S4, identifying completed components by using a deep learning algorithm according to the fusion model, and obtaining actual completed quantities by volume calculation; S5, comparing the actual completed quantities with planned completed quantities of a corresponding time node in the BIM design model to calculate a progress deviation rate; and S6, adjusting the construction cost according to the progress deviation rate and unit price information in the bill of quantities.

[0007] Further, S4, identifying completed components by using a deep learning algorithm according to the fusion model, and obtaining actual completed quantities by volume calculation, comprises the following steps: extracting point cloud feature data from the fusion model, wherein the point cloud feature data comprises spatial coordinates, reflection intensity and a normal vector; performing voxelization processing on the point cloud feature data according to the spatial coordinates to convert continuous point cloud data into a three-dimensional voxel grid, wherein each voxel unit contains reflection intensity and normal vector information of a corresponding spatial position; inputting the three-dimensional voxel grid into a pre-trained convolutional neural network model to obtain a component type label, a spatial position and a component point cloud cluster; and matching each component point cloud cluster identified according to the component type label and the spatial position information with a standard component of a corresponding position and type in the BIM design model to determine the completion state of the component by calculating a point cloud coverage rate, wherein the completion state comprises complete completion, partial completion and non-starting.

[0008] When the point cloud coverage rate is greater than a preset threshold Q1, the component is determined to be completely completed, the system obtains a unique identifier of the component in the BIM model through the component matching result, directly queries the design volume attribute of the component through a data interface of the BIM model, and directly assigns the design volume value obtained by the query as the actual completed volume. This processing method avoids the calculation error of point cloud reconstruction, and improves the efficiency and accuracy.

[0009] When the point cloud coverage rate is less than a preset threshold Q1 and greater than a preset threshold Q2, the component is determined to be partially completed, a three-dimensional surface of the component point cloud cluster is reconstructed according to the spatial coordinates of the point cloud cluster of the corresponding component, the volume contained by the closed surface is calculated, the volume value calculated is taken as the actual completed volume of the component, and the actual completed quantities of each subproject are accumulated according to the actual completed volume of each component and the volume calculation rule of the corresponding component type in the bill of quantities.

[0010] In particular, the present application avoids accumulated errors in point cloud data processing by directly calling the precise design volume in the BIM model for a completely completed component R>Q1. When the component is substantially completed, the deviation between the actual geometry and the design geometry is within the engineering allowable range, and directly using the design value not only ensures accuracy but also greatly improves calculation efficiency. For partially completed components Q2≤R≤Q1, the volume of the actual completed part is accurately calculated through point cloud reconstruction technology, solving the technical problem that traditional methods cannot accurately evaluate the component during construction. For components that have not started R<Q2, it avoids misjudging scattered noise points as the start of construction, improving the robustness of the system.

[0011] Further, the spatial coordinates are used to determine the spatial position and geometric shape of the component; the reflection intensity is used to distinguish components of different materials; and the normal vector is used to identify the surface orientation and boundary of the component.

[0012] Further, the three-dimensional voxel grid is input into a pre-trained convolutional neural network model to obtain component type labels, spatial positions, and component point cloud clusters, including: the convolutional neural network model includes a feature extraction layer, an instance segmentation layer, and a semantic recognition layer; the feature extraction layer extracts the geometric features and material features of the component using the reflection intensity and normal vector information; the instance segmentation layer separates adjacent components based on the extracted geometric features using a region growing algorithm, and outputs the point cloud clusters of each independent component; the semantic recognition layer outputs the component type label corresponding to each point cloud cluster based on the extracted material features and geometric features; and the center position of each component is calculated as the spatial position according to the boundary voxel coordinates of the component point cloud cluster.

[0013] Further, according to the component type label and spatial position information, each component point cloud cluster identified is matched with a standard component of the same type and position in the BIM design model, and the completion status of the component is determined by calculating the point cloud coverage rate, including:

[0014] Based on the component type label, a candidate component set of the same type is selected in the BIM design model; the centroid coordinates of the component point cloud cluster are calculated as the center position C, where n is the number of points in the component point cloud cluster, (x i ,y i ,z i ) is the three-dimensional coordinates of the i-th point; the distance D between the center position C and the geometric center of each component in the candidate component set is calculated, the component with the smallest distance D is selected as the matching component; the axis-aligned bounding box (AABB) parameters of the matching component are obtained, including the minimum vertex coordinates (x min ,y min ,z min ) and the maximum vertex coordinates (x max ,y max ,zmax ); compare each point coordinate in the component point cloud cluster with the axis-aligned bounding box (AABB) boundary, and count the number N of points falling within the axis-aligned bounding box (AABB) in ; divide the axis-aligned bounding box (AABB) into MxNxK voxel units according to a preset resolution r, count the number of points in each voxel unit, and calculate the actual distribution density

[0015]

[0016] According to the geometric information of the matched component in the BIM design model, the theoretical point cloud density p is calculated by simulating the laser radar scanning process standard ; according to the scanning parameters of the laser radar and the surface area of the component, the number of theoretical sampling points in a unit volume is calculated; the point cloud coverage rate is calculated When it is detected that the actual distribution density is obviously uneven, the point cloud is divided into multiple connected regions through a connected region analysis algorithm, the proportion e of the number of points in the largest connected region to N in is calculated, and the coverage rate is corrected: R' = R x e; when R > Q1, it is determined to be completely completed, when Q2 < R < Q1, it is determined to be partially completed, and when R < Q2, it is determined to be not started, wherein Q1 and Q2 are preset coverage rate thresholds. The value range of Q1 is 90% to 100%; the value range of Q2 is 10% to 20%.

[0017] In particular, first, based on the component type label identified by deep learning, the matching search space is reduced from thousands of components in the full model to dozens of components of the same type, and the search complexity is reduced from O(n) to O(n / k), where k is the number of component types. The introduction of this semantic prior knowledge avoids the "similar in shape but different in quality" mismatch that may occur in pure geometric matching.

[0018] Second, the centroid as the geometric feature center of the point cloud cluster, has translational invariance and noise robustness. The nearest neighbor search is performed by Euclidean distance , which in principle ensures that the components with the closest spatial positions are correctly matched. This centroid-based matching method has high computational efficiency and a complexity of only O(m), where m is the number of candidate components.

[0019] Finally, p standard is calculated by simulating the laser radar scanning process, considering the influence of scanning angle, distance, and occlusion on point cloud density. This theoretical value calculation based on the physical model is closer to the actual situation than the simple uniform distribution assumption, and solves the coverage consistency problem under different scanning conditions.

[0020] Further, when the point cloud coverage is less than a preset threshold Q1 and greater than a preset threshold Q2, it is determined that the component is partially completed, a three-dimensional surface of the component point cloud cluster is reconstructed according to the spatial coordinates of the point cloud cluster corresponding to the component, the volume contained by the closed surface is calculated, and the calculated volume value is taken as the actual completion volume of the component, including: performing outlier processing on the point cloud cluster of the partially completed component to obtain effective point clouds; extracting construction direction information of a matching component from the BIM design model as a main construction direction vector; wherein the civil engineering construction has a clear directional feature: column components: vertical construction from bottom to top; beam components: horizontal construction along the long axis direction; plate components: layer-by-layer pouring from bottom to top. The type attribute and geometric parameters of the component are read from the BIM model; the main construction direction vector is automatically determined according to the component type; for special components, a user-defined direction can be read from the construction information attribute of the BIM model.

[0021] The three-dimensional point cloud data is projected along the main construction direction vector to simplify the spatial distribution problem into a linear distribution problem. The projection value of each point in the construction direction represents its position in the construction process. The range interval of the projection value is determined, that is, the distance from the construction starting point to the current farthest point; the interval is equally divided into m subintervals (m is adaptively adjusted according to the component size, generally 30-50); the number of points in each subinterval is counted, and the linear density value is obtained by dividing the interval length; a density distribution curve along the construction direction is formed.

[0022] In addition, at the construction interface position, the completed part of the point cloud is dense, and the uncompleted part of the point cloud is sparse or has no point cloud, so there will be a sharp drop in density at the interface. Starting from the construction starting end, the density ratio of adjacent subintervals is calculated one by one; when the density ratio of a certain position is less than a threshold T (an empirical value of 0.3), it indicates that the density has changed abruptly; the mutation position is the construction completion interface.

[0023] The cutting plane passes through the identified construction interface position point, and the plane normal vector is the main construction direction vector, ensuring that the cutting plane divides the component into completed and uncompleted parts. The built-in geometric engine (such as ACIS, Parasolid, etc.) of the BIM software is called to perform intersection operation of solid and half space, and the geometric body on the side of the construction starting end is retained. The cut geometric body is a standard B-Rep solid model, and the volume calculation function of the geometric engine is directly called to obtain the volume value as the accurate mathematical calculation result without cumulative error, which is taken as the actual completion volume of the partially completed component.

[0024] In particular, the present application converts the traditional point cloud surface reconstruction problem into a geometric cutting problem based on the BIM model. Traditional point cloud surface reconstruction, such as the AlphaShape algorithm, needs to infer the complete geometric shape from the discrete point cloud, which belongs to the "from part to whole" reverse engineering, and there is inevitably information loss and reconstruction error. While the present scheme directly uses the accurate geometry provided by the BIM model, simplifying the problem to "determine the cutting position", which belongs to the "from whole to part" forward analysis, and avoids the uncertainty of geometric reconstruction in principle. In addition, the selection of the alpha parameter of the AlphaShape algorithm is subjective, and different alpha values will produce different boundary shapes, and it cannot be guaranteed that the reconstructed boundary is consistent with the actual construction interface. The present scheme directly locates the construction interface through density analysis, and then constructs a standard cutting plane, reducing the three-dimensional boundary identification problem to a one-dimensional interface detection problem, which guarantees the uniqueness and accuracy of the boundary in principle.

[0025] Further, according to the actual completed volume of each component and the volume calculation rule of the corresponding component type in the bill of quantities, the actual completed quantities of each sub-item of the project are accumulated, including: for reinforced concrete components, the actual completed quantities are calculated according to the following rules: beam and column components: actual completed quantities = actual completed volume × (1-0.025), wherein 0.025 is the steel volume deduction coefficient; plate components: actual completed quantities = actual completed volume × (1-0.015), wherein 0.015 is the steel volume deduction coefficient; foundation components: actual completed quantities = actual completed volume, without deducting the steel volume; for steel structure components, the actual completed quantities are calculated according to the following rules: actual completed quantities = actual completed volume × 7850kg / m 3 ÷1000.

[0026] Further, the volume calculation rules also include: for masonry components, the actual completed quantities are calculated according to the following rules: standard brick masonry: actual completed quantities = actual completed volume × 0.95, wherein 0.95 is the reduction coefficient considering the mortar joint; block masonry: actual completed quantities = actual completed volume ÷ (block volume × 1.1) to obtain the number of blocks, wherein 1.1 is the mortar joint coefficient; for formwork engineering, the actual completed quantities are calculated according to the following rules: beam formwork: actual completed quantities = (beam width + 2 × beam height) × actual completed length; plate formwork: actual completed quantities = actual completed area; all component quantities under the same sub-item engineering code are accumulated according to the same measurement unit to obtain the actual completed quantities of the sub-item.

[0027] Further, S5, comparing the actual completed work quantity with the planned completed work quantity of the corresponding time node in the BIM design model to calculate the progress deviation rate, including: obtaining the current construction date, extracting the planned completed work quantity of the corresponding time node from the planned progress data of the BIM design model; comparing the actual completed work quantity of each sub-project with the corresponding planned completed work quantity to calculate the progress deviation rate of each sub-project: progress deviation rate = (actual completed work quantity - planned completed work quantity) / planned completed work quantity x 100%; calculating the weighted average progress deviation rate according to the weight coefficient of each sub-project in the total project, wherein the weight coefficient is determined according to the contract amount proportion of each sub-project; generating progress warning information when the progress deviation rate of a certain sub-project exceeds the preset warning threshold, the warning information including the sub-project name, deviation rate value and deviation cause analysis; judging the progress state according to the positive and negative values of the progress deviation rate: when the progress deviation rate is positive, it means that the project progress is ahead of schedule; when the progress deviation rate is negative, it means that the project progress is behind schedule; generating a progress deviation analysis report containing the planned work quantity, actual work quantity, deviation rate and cumulative completion percentage of each sub-project.

[0028] Compared with the prior art, the application has the advantages that:

[0029] The application realizes accurate calculation of the work quantity by precisely registering and fusing the high-density point cloud data (resolution up to millimeter level) collected by the laser radar with the BIM design model, automatically identifying the components and calculating the point cloud coverage rate by using a three-dimensional convolutional neural network.

[0030] For completely completed components (coverage rate R > 90%), the accurate design volume in the BIM model is directly extracted, avoiding point cloud reconstruction errors; for partially completed components (10%≤R≤90%), the construction interface is accurately identified through density gradient analysis, and the BIM geometric engine is used for Boolean cutting operation to obtain the accurate volume.

[0031] Compared with the traditional manual measurement and estimation method, the error of the work quantity calculation is reduced from 10% to 15% to 2% to 3%, especially for complex special-shaped components and hard-to-reach high-altitude parts, realizing full coverage and high-precision automatic measurement. At the same time, fine work quantity calculation rules are established for different material types, such as differentiated deduction coefficients of reinforced concrete components (2.5% for beams and columns, 1.5% for plates) and accurate conversion of masonry mortar joints, ensuring accurate conversion from the actual completed volume to the work quantity and providing a reliable data basis for cost dynamic monitoring. BRIEF DESCRIPTION OF DRAWINGS

[0032] The present application will be further described in the manner of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting, and in these embodiments, the same numbers represent the same structures, wherein:

[0033] Figure 1 is an exemplary flowchart of a civil engineering cost management method based on artificial intelligence according to some embodiments of the present application;

[0034] Figure 2 is an exemplary flowchart of obtaining actual completed quantities according to some embodiments of the present application;

[0035] Figure 3 is an exemplary flowchart of generating component point cloud clusters according to some embodiments of the present application;

[0036] Figure 4 is an exemplary flowchart of calculating point cloud coverage according to some embodiments of the present application. DETAILED DESCRIPTION

[0037] The method and system provided by the embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0038] As Figure 1 shown, S1, a BIM design model of a construction project is constructed, the BIM design model including component information of a construction stage, a bill of quantities and planned progress data; S2, point cloud data of a construction site is collected by a laser radar to generate an actual construction point cloud model; S3, the actual construction point cloud model is spatially registered with the BIM design model to obtain a registered fusion model; S4, completed components are identified by a deep learning algorithm according to the fusion model, and actual completed quantities are obtained by volume calculation; S5, the actual completed quantities are compared with planned completed quantities at a corresponding time node in the BIM design model to calculate a progress deviation rate; S6, the engineering cost is adjusted according to the progress deviation rate in combination with unit price information in the bill of quantities.

[0039] Specifically, S1, a BIM design model of a construction project is constructed, the BIM design model including component information of a construction stage, a bill of quantities and planned progress data.

[0040] S2, point cloud data of a construction site is collected by a laser radar to generate an actual construction point cloud model; a laser radar scanner is used to scan the construction site 360 degrees in all directions according to a preset time interval; the collected raw point cloud data is preprocessed, including denoising, registration and segmentation; a semantic segmentation algorithm based on deep learning is used to classify the point cloud data to identify different materials and component types; and an actual construction point cloud model with semantic labels is generated.

[0041] S3, space registration is performed between the actual construction point cloud model and the BIM design model to obtain a registered fusion model, including: extracting feature geometric elements of the building from the BIM design model, including wall corner points, column center points and floor boundary lines, to generate a BIM feature point set; from the actual construction point cloud model, detecting planar features by using a RANSAC algorithm to identify wall surfaces, floor surfaces and column surfaces, and extracting planar intersection lines and intersection points as a point cloud feature point set; performing preliminary matching on the BIM feature point set and the point cloud feature point set, and using a nearest point iteration algorithm to calculate an initial transformation matrix T0, including a rotation matrix R0 and a translation vector t0; performing coarse registration on the actual construction point cloud model based on the initial transformation matrix T0 to convert the point cloud coordinate system to the BIM coordinate system; on the basis of coarse registration, performing fine registration by using a point-to-plane ICP algorithm: projecting each point in the point cloud to the nearest component surface in the BIM model; calculating the distance from the point to the plane to construct an error function wherein d i is the distance from the ith point to the corresponding plane; the transformation parameters are iteratively optimized by using a least squares method until the error function converges; and a final transformation matrix T final is applied to transform the actual construction point cloud model to obtain a registered point cloud in the same coordinate system as the BIM design model.

[0042] As shown in Figure 2 S4, according to the fusion model, a deep learning algorithm is used to identify completed components, and the actual completed quantities are obtained by volume calculation; point cloud feature data is extracted from the fusion model, including spatial coordinates, reflection intensity and normal vectors; the spatial coordinates are used to determine the spatial position and geometric shape of the component; the reflection intensity is used to distinguish components of different materials; and the normal vector is used to identify the surface orientation and boundary of the component. The spatial coordinates of each point P i are represented as (x i , y i , z i ), wherein i ∈ [1, N] and N is the total number of point clouds; the reflection intensity is represented as I , wherein P r is the received power, P e is the transmitted power, and the reflection intensity characteristic values of different materials are as follows: concrete: I i ∈ [0.3, 0.5], steel: I i ∈ [0.7, 0.9], and brick masonry: I i ∈ [0.4, 0.6]; the reflection intensity is clustered by using a Gaussian mixture model (GMM) to realize automatic material identification.

[0043] The normal vector is calculated using the Principal Component Analysis (PCA) algorithm to compute the normal vector n at each point. i =(n x ,n y ,n z Select point P. i The set of k nearest neighbors N k (P i Given k=20, construct the covariance matrix. Where, p j ∈N k (P i ), normal vector n i Let C be the eigenvector corresponding to the smallest eigenvalue of C. When the angle θ between the normal vectors of adjacent points is greater than 30°, it is determined to be a boundary.

[0044] The point cloud feature data is voxelized based on spatial coordinates, converting continuous point cloud data into a 3D voxel mesh. Each voxel unit contains the reflection intensity and normal vector information corresponding to its spatial location. Specifically, the voxel resolution v is set. r =0.05m (5cm), calculate the point cloud bounding box: [x min ,x max ]×[y min ,y max ]×[z min ,z max Voxel mesh size generated:

[0045] A 3D voxel mesh is input into a pre-trained convolutional neural network model to obtain component type labels, spatial locations, and component point cloud clusters. The convolutional neural network model includes a feature extraction layer, an instance segmentation layer, and a semantic recognition layer. The feature extraction layer uses reflection intensity and normal vector information to extract the geometric and material features of the components. The instance segmentation layer separates adjacent components based on the extracted geometric features using a region growing algorithm and outputs point cloud clusters for each independent component. The semantic recognition layer outputs the component type label corresponding to each point cloud cluster based on the extracted material and geometric features. The center position of each component is calculated based on the boundary voxel coordinates of the component point cloud clusters, which serves as the spatial location.

[0046] like Figure 3 and Figure 4 As shown, based on component type labels and spatial location information, the identified component point cloud clusters are matched with standard components of corresponding locations and types in the BIM design model. The completion status of the components is determined by calculating the point cloud coverage rate, which includes fully completed, partially completed, and not yet started. Based on the component type labels, a set of candidate components of the same type is filtered in the BIM design model. The centroid coordinates of the component point cloud clusters are calculated and used as the center position C. Where n is the number of points in the component point cloud cluster, (x i ,y i ,z i Let C be the three-dimensional coordinates of the i-th point; calculate the distance D between the center position C and the geometric centers of all components in the candidate component set.

[0047] Select the component with the smallest distance D as the matching component; obtain the axis-aligned bounding box (AABB) parameters of the matching component, including the minimum vertex coordinates (x, y, y). min ,y min ,z min ) and maximum vertex coordinates (x max ,y max ,z max The coordinates of each point in the component point cloud cluster are compared with the axis-aligned bounding box (AABB) boundary, and the number N points falling into the AABB is counted. in Divide the axis-aligned bounding box (AABB) into M×N×K voxel units according to a preset resolution r, count the number of points in each voxel unit, and calculate the actual distribution density. Based on the geometric information of the matching components in the BIM design model, the theoretical point cloud density ρ is calculated by simulating the lidar scanning process. standard Based on the LiDAR scanning parameters and the component surface area, calculate the theoretical number of sampling points per unit volume; calculate the point cloud coverage rate. When a significant unevenness in the actual distribution density is detected, the point cloud is divided into multiple connected regions using a connected component analysis algorithm. The proportion ε of the number of points in the largest connected region relative to Nin is calculated, and the coverage is corrected: R' = R × ε. When R > Q1, it is considered fully completed; when Q2 ≤ R ≤ Q1, it is considered partially completed; and when R < Q2, it is considered not started. Here, Q1 and Q2 are preset coverage thresholds. The value of Q1 ranges from 90% to 100%; the value of Q2 ranges from 10% to 20%.

[0048] When the point cloud coverage R > Q1 (Q1 = 0.95), it is determined to be a fully completed component, and the unique identifier of the component is set as CID (ComponentID); the design volume is extracted through the BIM data interface function: V actual =BIM.GetVolume(CID), where V actual This indicates the actual completed volume, in meters (m). 3 The designed volume is directly used as the actual completed volume, thus avoiding point cloud reconstruction errors.

[0049] When Q2≤R≤Q1 (Q2=0.3, Q1=0.95), the refined processing flow is executed.

[0050] First, outliers are removed. For a point cloud cluster P = {p1, p2, ..., p...} n Each point p in} i Calculate p i The average distance to its k nearest neighbors: Where k = 20 (empirical value); calculate the statistical characteristics of the average distance of all points: Outlier detection criteria: If Then p i For outliers, remove all outliers to obtain the valid point cloud P. valid .

[0051] Extract the main construction direction vector. Based on the component type T, extract the main construction direction vector v from the BIM model: If T = "Column": v = (0,0,1) T If T = "beam": If T = "board": v = (0, 0, 1) T If T = "wall": v = n wall (Wall normal vector); where p start ,p end These are the coordinates of the beam's start and end points, respectively.

[0052] For the effective point cloud P valid Each point p in i =(x i ,y i ,z i ) T Projection: t i =p i ×v=x i v x +y i v y +z i v z Determine the projection interval: [t] min ,t max ];t min =min{t i |i=1,2,.....,n};t max =max{t i |i=1,2,.....,n}.

[0053] Divide the projection interval into m equal sub-intervals (m=50): The j-th subinterval: I j =[t min +(j-1)Δt,t min +jΔt], j=1,2,.....,m. Count the number of points in each subinterval: n j =|{p i|t i ∈I j}|, calculate linear density: (Unit: points / meter)

[0054] Identify the location of the construction interface and calculate the density ratio of adjacent sub-intervals: Construction interface determination: If r j <T (T=0.3) and ρ j+1 <ρ min Then the interface is located at the end of the j-th subinterval, where ρ min =0.1×max{ρ j |j=1,2,.....,m} is the minimum density threshold, and the construction interface location is: h cut =t min +j×Δt.

[0055] Cut and calculate the partially completed volume, cut point: P cut =p origin +h cut ×v, where p origin Let v be the starting point of the component. Equation of the cutting plane: v × (PP) cut ) = 0, expanded form: v x (xx cut )+v y (yy cut )+v z (zz cut ) = 0.

[0056] Boolean cut operation: Define half-space H: H = {P|v×(PP)} cut )≤0}. Perform the intersection operation:

[0057] G partial =G BIM ∩H, where G BIM For component geometry in BIM, G partial This is the portion after cutting. Volume calculation: V actual =∫∫∫ Gpartial dV is actually obtained through the Volume() function of the geometry engine.

[0058] Based on the actual completed volume of each component and the volume calculation rules for the corresponding component type in the bill of quantities, the actual completed volume of each sub-item of the project is obtained by accumulating the volumes according to the project sub-items.

[0059] For reinforced concrete components, the following rules are used for calculation: beam and column components: actual completed engineering quantity = actual completed volume x (1-0.025), wherein 0.025 is the steel volume deduction coefficient; plate components: actual completed engineering quantity = actual completed volume x (1-0.015), wherein 0.015 is the steel volume deduction coefficient; foundation components: actual completed engineering quantity = actual completed volume, without deduction of steel volume.

[0060] For steel structure components, the following rules are used for calculation: actual completed engineering quantity = actual completed volume x 7850 kg / m 3 ÷1000.

[0061] For masonry components, the following rules are used for calculation: standard brick masonry: actual completed engineering quantity = actual completed volume x 0.95, wherein 0.95 is the reduction coefficient considering mortar joints; block masonry: actual completed engineering quantity = actual completed volume ÷ (block volume x 1.1) to obtain the number of blocks, wherein 1.1 is the mortar joint coefficient.

[0062] For formwork engineering, the following rules are used for calculation: beam formwork: actual completed engineering quantity = (beam bottom width + 2 x beam height) x actual completed length; plate formwork: actual completed engineering quantity = actual completed area; all component engineering quantities under the same sub-item engineering code are added up according to the same measurement unit to obtain the actual completed engineering quantity of the sub-item;

[0063] S5, comparing the actual completed engineering quantity with the planned completed engineering quantity of the corresponding time node in the BIM design model, calculating the progress deviation rate; obtaining the current construction date, extracting the planned completed engineering quantity of the corresponding time node from the planned progress data of the BIM design model;

[0064] The actual completed engineering quantity of each sub-item engineering is compared with the corresponding planned completed engineering quantity to calculate the progress deviation rate of each sub-item engineering: progress deviation rate = (actual completed engineering quantity - planned completed engineering quantity) / planned completed engineering quantity x 100%;

[0065] According to the weight coefficient of each sub-item engineering in the total engineering, the weighted average progress deviation rate is calculated, wherein the weight coefficient is determined according to the contract amount proportion of each sub-item engineering; when the progress deviation rate of a sub-item engineering exceeds a preset warning threshold, progress warning information is generated, and the warning information includes the name of the sub-item engineering, the deviation rate value and the deviation cause analysis;

[0066] According to the positive and negative values of the progress deviation rate, the engineering progress state is judged: when the progress deviation rate is positive, it means that the engineering progress is ahead of schedule; when the progress deviation rate is negative, it means that the engineering progress is behind schedule; a progress deviation analysis report is generated, which contains the planned engineering quantity, actual engineering quantity, deviation rate and cumulative completion percentage of each sub-item engineering.

[0067] S6, according to the progress deviation rate, combining the unit price information in the bill of quantities, adjusting the project cost.

[0068] The above description of the application and its embodiments is illustrative and not restrictive, and the application can be practiced in other specific forms without departing from the spirit or essential characteristics thereof. The embodiments shown in the drawings are only one of the embodiments of the application, and the actual structure is not limited thereto. Therefore, if a person of ordinary skill in the art is inspired by it, without departing from the spirit of the application, similar structural forms and embodiments can be designed without creativity, which should belong to the protection scope of the application. In addition, the word "comprising" does not exclude other elements or steps, and the word "one" before the element does not exclude the inclusion of "multiple" elements. The words "first", "second" and the like are used to indicate names, not any specific order.

Claims

1. A civil engineering cost management method based on artificial intelligence, characterized by, Comprise: S1, constructing a BIM design model of a construction project, the BIM design model comprising component information of a construction stage, a bill of quantities and schedule data; S2, collecting point cloud data of a construction site by laser radar to generate an actual construction point cloud model; S3, spatially registering the actual construction point cloud model with the BIM design model to obtain a registered fusion model; S4, identifying completed components from the fusion model by a deep learning algorithm and obtaining actual completed quantities of work by volume calculation; S5, comparing the actual completed quantities of work with planned completed quantities of work at a corresponding time node in the BIM design model to calculate a progress deviation rate; S6, adjusting the construction cost according to the progress deviation rate in combination with unit price information in the bill of quantities; S4, identifying completed components from the fusion model by a deep learning algorithm and obtaining actual completed quantities of work by volume calculation, comprising: extracting point cloud feature data from the fusion model, the point cloud feature data comprising spatial coordinates, reflection intensity and normal vector; performing voxelization processing on the point cloud feature data according to the spatial coordinates to convert continuous point cloud data into a three-dimensional voxel grid, each voxel unit containing reflection intensity and normal vector information of the corresponding spatial position; inputting the three-dimensional voxel grid into a pre-trained convolutional neural network model to obtain component type labels, spatial positions and component point cloud clusters; matching each component point cloud cluster identified according to the component type labels and spatial position information with a standard component of the corresponding position and type in the BIM design model to determine the completion status of the component by calculating the point cloud coverage rate, the completion status comprising complete completion, partial completion and non-starting; When the point cloud coverage is greater than a preset threshold When the point cloud coverage is greater than a preset threshold When the point cloud coverage is greater than a preset threshold When the point cloud coverage is less than a preset threshold and greater than a preset threshold , it is determined that the component is partially completed, a three-dimensional surface of the component point cloud cluster is reconstructed according to the spatial coordinates of the point cloud cluster of the corresponding component, the volume contained by the closed surface is calculated, and the calculated volume value is taken as the actual completed volume of the component; according to the actual completion volume of each component and the volume calculation rule of the corresponding component type in the bill of quantities, accumulating the actual completed quantities of work of each sub-item of work according to the engineering sub-items; inputting the three-dimensional voxel grid into a pre-trained convolutional neural network model to obtain component type labels, spatial positions and component point cloud clusters, comprising: the convolutional neural network model comprises a feature extraction layer, an instance segmentation layer and a semantic recognition layer; the feature extraction layer extracts geometric features and material features of the component using reflection intensity and normal vector information; the instance segmentation layer separates adjacent components based on the extracted geometric features by a region growing algorithm to output point cloud clusters of each independent component; the semantic recognition layer outputs the component type labels corresponding to each point cloud cluster based on the extracted material features and geometric features; the center position of each component is calculated as the spatial position according to the boundary voxel coordinates of the component point cloud cluster.

2. The civil engineering cost management method based on artificial intelligence according to claim 1, characterized in that: the spatial coordinates are used to determine the spatial position and geometric shape of the component; the reflection intensity is used to distinguish components of different materials; the normal vector is used to identify the surface orientation and boundary of the component.

3. The civil engineering cost management method based on artificial intelligence according to claim 1, characterized in that: According to the component type label and the spatial position information, each identified component point cloud cluster is matched with a standard component of the corresponding position and type in the BIM design model, the completion state of the component is determined by calculating the point cloud coverage rate, including: Based on the component type label, a candidate component set of the same type is screened in the BIM design model; a center position C is calculated as the centroid coordinate of the component point cloud cluster, where n is the number of points in the component point cloud cluster, is the three-dimensional coordinate of the i-th point; calculating distances D between the center position C and the geometric centers of the components in the candidate component set, selecting the component with the smallest distance D as the matching component; obtaining an axis-aligned bounding box (AABB) parameter of the matching component, the AABB parameter comprising minimum vertex coordinates and maximum vertex coordinates ; Compare each point coordinate in the component point cloud cluster with the axis-aligned bounding box (AABB) boundary, and count the number of points falling into the axis-aligned bounding box (AABB) ; The axis-aligned bounding box (AABB) is divided into MxNxK voxel units according to a preset resolution r, the number of points in each voxel unit is counted, and the actual distribution density is calculated ; According to the geometric information of the matching component in the BIM design model, the theoretical point cloud density is calculated by simulating the laser radar scanning process ; According to the scanning parameters of the laser radar and the surface area of the component, the theoretical sampling point number in unit volume is calculated; Calculate point cloud coverage When a significant unevenness in the actual distribution density is detected, the point cloud is divided into multiple connected regions using a connected component analysis algorithm, and the percentage of points in the largest connected region is calculated. proportion To correct the coverage: ; When is determined to be fully completed, when is determined to be partially completed, and when is determined to be not started, wherein, and are preset coverage rate thresholds.

4. The artificial intelligence-based civil engineering cost management method according to claim 3, characterized in that: a value range of 90% to 100%; the range of 10% to 20%.

5. The artificial intelligence-based civil engineering cost management method according to claim 1, characterized in that: When the point cloud coverage is less than a preset threshold and greater than a preset threshold , it is determined that the component is partially completed, and a three-dimensional surface of a point cloud cluster of the component is reconstructed according to the spatial coordinates of the point cloud cluster of the corresponding component, a volume contained by the closed surface is calculated, and a calculated volume value is taken as an actual completed volume of the component. Outlier points are processed for the point cloud cluster of the partially completed component to obtain effective point clouds; The construction direction information of the matched component is extracted from the BIM design model as a main construction direction vector; The effective point clouds are projected along the main construction direction vector, the projection interval is equally divided into m sub-intervals, the number of points in each sub-interval is counted, and the density distribution of the effective point clouds in the main construction direction is calculated; The density ratio of adjacent sub-intervals is calculated, and when the density ratio is less than a preset threshold T, the end position of the i-th sub-interval is determined as the construction interface position. The density ratio of adjacent sub-intervals is calculated, and when the density ratio is less than a preset threshold T, the end position of the i-th sub-interval is determined as the construction interface position. According to the construction interface position, a cutting plane is set on the geometric model of the matched component in the BIM design model, the cutting plane passes through the construction interface position and is perpendicular to the main construction direction vector; The geometric engine in the BIM is used to perform a Boolean cutting operation on the geometric model of the matched component, the part between the construction starting end and the cutting plane is retained, the volume of the retained part is calculated as the actual completed volume of the partially completed component.

6. The artificial intelligence-based civil engineering cost management method according to claim 5, characterized in that: According to the actual completed volume of each component and the volume calculation rule of the corresponding component type in the bill of quantities, the actual completed quantities of each sub-item of the project are accumulated according to the engineering sub-items, including: For reinforced concrete components, the actual completed quantities are calculated according to the following rules: Beam and column components: actual completed quantities = actual completed volume × (1-0.025), where 0.025 is the steel volume deduction coefficient; Plate components: actual completed quantities = actual completed volume × (1-0.015), where 0.015 is the steel volume deduction coefficient; Foundation components: actual completed quantities = actual completed volume, without deduction of steel volume; For steel structure components, the actual completed quantities are calculated according to the following rules: Actual completed quantities = actual completed volume × 7850 kg / m³ ÷ 1000.

7. The artificial intelligence-based civil engineering cost management method according to claim 6, characterized in that: The volume calculation rule includes: For masonry components, the actual completed quantities are calculated according to the following rules: Standard brick masonry: actual completed quantities = actual completed volume × 0.95, where 0.95 is the reduction coefficient considering the mortar joint; Block masonry: actual completed quantities = actual completed volume ÷ (block volume × 1.1) to obtain the number of blocks, where 1.1 is the mortar joint coefficient; For formwork engineering, the actual completed quantities are calculated according to the following rules: Beam formwork: actual completed quantities = (beam bottom width + 2 × beam height) × actual completed length; Plate formwork: actual completed quantities = actual completed area; The quantities of all components under the same sub-item engineering code are accumulated according to the same measurement unit to obtain the actual completed quantities of the sub-item.

8. The artificial intelligence-based civil engineering cost management method according to any one of claims 1 to 7, characterized in that: S5, comparing the actual completed quantities with the planned completed quantities of the corresponding time nodes in the BIM design model to calculate the progress deviation rate, including: obtaining the current construction date, extracting the planned completed quantities of the corresponding time nodes from the planned progress data of the BIM design model; comparing the actual completed quantities of the respective sub-projects with the corresponding planned completed quantities to calculate the progress deviation rate of each sub-project: Progress deviation rate = (actual completed quantity - planned completed quantity) / planned completed quantity x 100%; calculating the weighted average progress deviation rate according to the weight coefficients of each sub-project in the total project, wherein the weight coefficients are determined according to the contract amount proportion of each sub-project; when the progress deviation rate of a sub-project exceeds the preset warning threshold, generating progress warning information, the warning information including the sub-project name, deviation rate value and deviation cause analysis; judging the progress state according to the positive and negative values of the progress deviation rate: when the progress deviation rate is positive, it means that the project progress is ahead of schedule; when the progress deviation rate is negative, it means that the project progress is behind schedule; generating a progress deviation analysis report containing the planned quantities, actual quantities, deviation rates and cumulative completion percentages of each sub-project.

Images