Intelligent engineering supervision method and system based on artificial intelligence

Abstract

The invention belongs to the field of intelligent supervision, and discloses an intelligent engineering supervision method and system based on artificial intelligence, and the method comprises the steps: extracting three-dimensional field frequency features based on a multi-modal data package and a coordinate mapping table, carrying out the fusion through combining with a multi-modal graph neural network, and generating multi-modal fusion features; a VR complex is constructed based on the multi-modal fusion features and the coordinate mapping table, and a three-dimensional topological feature matrix is generated through continuous homology analysis; constructing a defect detection model, and obtaining defect positioning coordinates and engineering structure risk scores according to the generated three-dimensional topological feature matrix in combination with the multi-modal fusion features and the coordinate mapping table; and dynamically adjusting construction parameters according to the engineering structure risk score, and generating an engineering supervision report. According to the method, the VR complex is constructed by reducing the sampling point cloud, and the continuous homology analysis is performed, so that the topological invariant quantitative evaluation of the engineering structure microdefects is realized, and the supervision process has stronger real-time response capability.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent supervision, and specifically relates to an intelligent engineering supervision method and system based on artificial intelligence. Background Art

[0002] With the rapid development of artificial intelligence and the Internet of Things (IoT) technologies, the field of smart project supervision is undergoing a profound transformation from traditional manual oversight to intelligent and automated approaches. First, multi-source heterogeneous data acquisition technology is maturing, including BIM-based geometric data, sensor-based environmental monitoring data, and image recognition-based construction progress data. Second, machine learning algorithms have made significant progress in project quality prediction and risk assessment, with convolutional neural networks demonstrating excellent crack identification performance. Furthermore, the use of digital twin technology enables virtual simulation and real-time monitoring throughout the project lifecycle.

[0003] However, existing technologies still face critical challenges in practical engineering applications. First, ensuring spatiotemporal consistency in multi-source data fusion is difficult. While existing technologies (such as patent application CN120181586A) can achieve basic data synchronization, their reliance on static calibration and software-level time synchronization results in insufficient spatiotemporal alignment accuracy and an inability to adapt to the dynamic scenarios of mobile devices. Second, risk assessment models lack interpretability. Existing technologies often use end-to-end deep learning models for quality assessment. While these technologies offer high recognition accuracy, the decision-making process lacks transparency, making it difficult to provide a reliable basis for engineering decisions. Summary of the Invention

[0004] In view of the above problems, the present invention provides an intelligent engineering supervision method and system based on artificial intelligence to solve the problems of poor spatiotemporal consistency of data fusion and insufficient interpretability of risk assessment in the existing technology.

[0005] The present invention provides the following technical solutions:

[0006] In a first aspect, the present invention provides an artificial intelligence-based intelligent engineering supervision method, comprising the following steps: S1. Collect multimodal fusion data and perform preprocessing to generate multimodal data packets and coordinate mapping tables; S2. Based on the multimodal data packet and coordinate mapping table, extract the three-dimensional field frequency features and fuse them with the multimodal graph neural network to generate multimodal fusion features; S3, construct VR complex based on multimodal fusion features and coordinate mapping table, and generate three-dimensional topological feature matrix through continuous homology analysis; S4. Construct a defect detection model, and obtain defect location coordinates and engineering structure risk scores based on the generated three-dimensional topological feature matrix combined with multimodal fusion features and coordinate mapping table; S5. Dynamically adjust construction parameters based on the project structure risk score and generate a project supervision report.

[0007] Furthermore, in S1, the multimodal fusion data includes downsampled point cloud data, infrared thermal field distribution features, and acoustic emission frequency domain features; Based on the multimodal fusion data, multimodal data packets and coordinate mapping tables are generated through preprocessing through multimodal spatiotemporal alignment and feature fusion.

[0008] Furthermore, in S2, based on the multimodal data packet and the coordinate mapping table, the three-dimensional field frequency features are extracted and fused with the multimodal graph neural network to generate multimodal fusion features, including: Based on multimodal data packets and coordinate mapping tables, spatial node feature fusion is used to construct a cross-modal graph structure, and three-dimensional field frequency features are extracted through graph convolution aggregation; The three-dimensional field frequency features include point cloud spatial structure features, thermal field gradient features and acoustic frequency domain energy features. Based on the three-dimensional field frequency features, multimodal graph neural networks are combined to generate multimodal fusion features through neighborhood feature propagation.

[0009] Furthermore, in S3, a VR complex is constructed based on multimodal fusion features and coordinate mapping tables, and a three-dimensional topological feature matrix is ​​generated through continuous homology analysis, including: Define the connection radius parameters based on the spatial constraint parameters in the coordinate mapping table; Based on the downsampled point cloud data and the connection radius parameter, the simplex structure is determined by the neighborhood point pairs, and the VR complex is generated by dimensionally expanding it. Multi-scale filtering is performed on the VR complex to calculate the number of annular holes, and a persistence histogram is generated using a persistent homology algorithm. The persistence histogram is mapped through spatial gridding to generate a three-dimensional topological feature matrix.

[0010] Furthermore, in S4, a defect detection model is constructed to obtain defect location coordinates and engineering structure risk scores based on the generated three-dimensional topological feature matrix combined with multimodal fusion features and coordinate mapping table, including: The three-dimensional topological feature matrix and the multimodal fusion feature are aligned in a spatial grid through a coordinate mapping table to obtain the defect type and defect location coordinates of the engineering structure; Based on the defect type of the engineering structure, the positioning coordinates are converted into the engineering coordinate system through the least squares coordinate transformation matrix and the defect density is obtained through grid space division; The defect type weight is obtained by linear regression based on the engineering structure defect density and the frequency of the engineering structure defect type appearing in each area; A defect risk scoring model is constructed using an exponential function based on the engineering structure defect type, defect density, and defect type weight; Based on the defect detection model and combined with multimodal fusion features and defect location coordinates, the engineering structure risk score is calculated through the neighborhood risk propagation algorithm.

[0011] Furthermore, in S5, construction parameters are dynamically adjusted according to the engineering structure risk score, including: Based on historical structural risk event data, define low risk threshold R1 and high risk threshold R2; When the engineering structure risk score is less than R1, the engineering structure is considered to be in a low-risk state; when R1≤engineering structure risk score<R2, the engineering structure is considered to be in a medium-risk state; when the engineering structure risk score is greater than or equal to R2, the engineering structure is considered to be in a high-risk state; Based on the defect location coordinates, dynamic adjustment instructions are generated in combination with the construction strategy library, and construction parameters are adjusted.

[0012] Furthermore, a project supervision report is generated, including: Construction parameters include grouting pressure, curing cycle and equipment vibration frequency; The engineering supervision report is generated through summary logic based on the defect location coordinates, engineering structure risk score, multimodal fusion features and adjusted construction parameters.

[0013] In a second aspect, the present invention provides an artificial intelligence-based intelligent engineering supervision system, comprising a multimodal generation module, a cross-modal graph fusion module, a topology extraction module, a dual-drive module, and a supervision optimization module; The multimodal generation module is used to collect multimodal fusion data and perform preprocessing to generate a multimodal data packet and a coordinate mapping table; The cross-modal graph fusion module is used to extract three-dimensional field frequency features based on the multimodal data packet and the coordinate mapping table and fuse them with the multimodal graph neural network to generate multimodal fusion features; The topology extraction module is used to construct a VR complex based on multimodal fusion features and a coordinate mapping table, and generate a three-dimensional topological feature matrix through continuous homology analysis; The dual-drive module is used to build a defect detection model, and obtain defect location coordinates and engineering structure risk scores based on the generated three-dimensional topological feature matrix combined with multimodal fusion features and coordinate mapping tables; The report generation module is used to dynamically adjust construction parameters according to the engineering structure risk score and generate an engineering supervision report.

[0014] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the artificial intelligence-based intelligent engineering supervision method as described in the first aspect of the present invention is implemented.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the artificial intelligence-based smart engineering supervision method as described in the first aspect of the present invention.

[0016] The beneficial effects of the present invention are as follows: the present invention constructs a global coordinate system through hardware clock synchronization and calibration parameter compensation, combines the coordinate mapping table to achieve sub-meter spatial alignment of multimodal data, and dynamically supports mobile device inspection; at the same time, in terms of model interpretability, the present invention pioneers a physical-topological dual-drive architecture to quantify thermal field gradients, acoustic energy characteristics and topological invariants; and uses downsampled point clouds to construct VR complexes and perform continuous coherent analysis, realizing the quantitative evaluation of topological invariants of microscopic defects in engineering structures, so that the supervision process has stronger real-time response capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 This is a flow chart of the smart engineering supervision method based on artificial intelligence; Figure 2 This is a schematic diagram of an AI-based smart engineering supervision system; Figure 3 Flowchart of multimodal feature fusion; Figure 4 Flowchart of topological feature extraction. DETAILED DESCRIPTION

[0019] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0020] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0021] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0022] Reference Figures 1 to 4 , is an embodiment of the present invention, which provides an intelligent engineering supervision method based on artificial intelligence, comprising the following steps:

[0023] S1. Collect multimodal fusion data and perform preprocessing to generate multimodal data packets and coordinate mapping tables; Multimodal fusion data includes downsampled point cloud data, infrared thermal field distribution characteristics, and acoustic emission frequency domain characteristics; It should be noted that the downsampled point cloud data provides a set of spatial reference points for crack location and deformation quantification by constructing a three-dimensional spatial topological geometric model of the surface of the object being measured; the infrared thermal field distribution characteristics detect the abnormal distribution of energy conduction inside the component and quantify the sudden change of thermal gradient to identify the material deterioration area; the acoustic emission frequency domain characteristics capture the frequency domain characteristics of the stress wave released in the dynamic process of microscopic damage, and identify the damage mode through the frequency band energy distribution.

[0024] Based on the multimodal fusion data, multimodal spatiotemporal alignment and feature fusion are used for preprocessing to generate multimodal data packets and coordinate mapping tables; It should be noted that the laser point cloud data, infrared thermal imaging data, and acoustic emission signals were synchronously processed in the time domain to create a spatiotemporally aligned multimodal dataset. Within the spatiotemporally aligned multimodal data, outlier filtering and voxel downsampling were performed on the laser point cloud data to generate downsampled point cloud data. Furthermore, non-uniform temperature compensation and adaptive histogram equalization were performed on the infrared thermal imaging data to enhance its quality. For the acoustic emission signals, acoustic frequency domain feature data was extracted through bandpass filtering and Mel-spectrogram conversion. Finally, the downsampled point cloud data, enhanced thermal imaging data, and acoustic frequency domain feature data were correlated through spatial coordinate projection to construct a comprehensive dataset consisting of multimodal data packages and coordinate mapping tables.

[0025] S2. Based on the multimodal data packet and coordinate mapping table, extract the three-dimensional field frequency features and fuse them with the multimodal graph neural network to generate multimodal fusion features; Based on multimodal data packets and coordinate mapping tables, spatial node feature fusion is used to construct a cross-modal graph structure, and three-dimensional field frequency features are extracted through graph convolution aggregation; It should be noted that the spatial nodes are first divided into 0.1m³ grids according to the global coordinates of the downsampled point cloud data in the coordinate mapping table. Each node is bound to the point cloud spatial structure characteristics, thermal field gradient characteristics and acoustic frequency domain energy characteristics. Then, the nodes are connected with the Euclidean distance ≤ 0.3m rule to construct weighted edges. Finally, 128-dimensional three-dimensional field frequency features are extracted through neighborhood spatial feature propagation and cross-modal frequency domain feature enhancement aggregation.

[0026] Furthermore, a cross-modal graph structure is constructed to correct the geometric neighborhood relationship through the thermal field gradient (such as expanding the connection radius by 15% in high-temperature areas) to solve the problem of missing point clouds caused by occlusion; defect features propagate faster along cross-modal edges than in single-modal graphs, realizing coordinated early warning of cracks, temperature anomalies and acoustic attenuation.

[0027] The three-dimensional field frequency characteristics include point cloud spatial structure characteristics, thermal field gradient characteristics and acoustic frequency domain energy characteristics; It should be noted that the spatial structure feature of the point cloud is formed by extracting the average curvature and reflection intensity distribution variance of the point cloud in the grid by downsampling the point cloud data to form a 48-dimensional vector; the thermal field gradient feature is formed by calculating the mean and standard deviation of the gradient norm by associating the infrared thermal field distribution features through the coordinate mapping table to form a 48-dimensional vector; the acoustic frequency domain energy feature is formed by extracting the Mel spectrum peak frequency, 0-50kHz energy proportion and envelope attenuation slope based on the coordinate mapping table indexing the acoustic emission frequency domain features to form a 32-dimensional vector.

[0028] Based on three-dimensional field frequency features, multimodal graph neural network is combined to generate multimodal fusion features through neighborhood feature propagation; It should be noted that, first, the three-dimensional field frequency features need to be bound to the physical coordinates through a coordinate mapping table; then, in the cross-modal graph structure, weighted aggregation is performed on the three-dimensional field frequency features in the neighborhood, and linear transformation is performed through the weight matrix to output intermediate features; then, the attention mechanism is used to perform cross-modal interaction on the intermediate features; finally, based on the correlation information of the point cloud spatial structure features, thermal field gradient features and acoustic frequency domain energy features, multimodal fusion features are generated through integration.

[0029] Furthermore, intermediate features represent the high-order semantic representations generated by the cross-modal graph neural network during feature propagation. At its core, intermediate features are the transitional state formed after the original three-dimensional field frequency features undergo dual spatial and modal encoding. First, within the cross-modal graph structure, weighted aggregation is performed based on the three-dimensional field frequency features within the neighborhood to construct a neighborhood graph. Next, a weight matrix is ​​used to perform cross-modal projection on the spliced ​​multimodal features, and a 256-dimensional feature vector is activated and output.

[0030] S3, construct VR complex based on multimodal fusion features and coordinate mapping table, and generate three-dimensional topological feature matrix through continuous homology analysis; Define the connection radius parameters based on the spatial constraint parameters in the coordinate mapping table; It should be noted that the spatial constraint parameters are two core parameters defined in the coordinate mapping table: the grid size parameter and the coordinate system alignment offset, which are used to control the spatial grid division rules and the alignment benchmark of the engineering coordinate system; the connection radius parameter is used to define the maximum Euclidean distance between adjacent grids in cross-modal connections; the radius parameter can be directly defined based on the grid size parameter (for example, grid size parameter = 1.0m → connection radius parameter = 1.0m; grid size parameter is refined to 0.1m → connection radius parameter = 0.1m).

[0031] Based on the downsampled point cloud data and the connection radius parameter, the simplex structure is determined by the neighborhood point pairs, and the VR complex is generated by dimensionally expanding it. It should be noted that the coordinates of each point in the downsampled point cloud data , other points , calculate the Euclidean distance ,like ≤Connection radius parameter value, then mark point pair ( , ) is a pair of neighboring points, and then the neighboring point pairs that meet the conditions are combined into a simplex structure; the calculation formula of the Euclidean distance is: , in, for point Arrive The Euclidean distance of for point No. Coordinate components ( =1 corresponds to the x coordinate, =2 corresponds to the y coordinate); for point No. Coordinate components ( =1 corresponds to the x coordinate, =2 corresponds to the y coordinate); for point Arrive The normalized distance factor of ; for Correction term for the rate of change of local spatial distribution density of point cloud, is the density gradient vector of the local spatial distribution of the point cloud; It should be noted that the simplex structure is expanded dimensionally to generate a VR complex. First, each point in the downsampled point cloud data is expanded to a 0-dimensional simplex (100 points generate 100 0-dimensional simplexes); secondly, a 1-dimensional simplex is generated based on the triangular edges of the simplex structure (the side length is determined to be ≤ the connection radius parameter value, such as 1.0m, 162 triangular faces have 486 edges); finally, the triangular faces are merged to construct a 3-dimensional simplex, which requires that there are 4 triangular faces with common edges and the circumscribed sphere radius of the tetrahedron is ≤ the connection radius parameter value (for example, 162 triangular faces → 54 tetrahedrons). When the number of newly added 3-dimensional simplexes = 0, the expansion is terminated and the VR complex is output.

[0032] Furthermore, VR complex is a method for constructing topological structures based on point cloud data. Its core idea is to connect the distance point set in Euclidean space into a simplex (such as point, edge, triangle, tetrahedron, etc.) by setting the distance threshold parameter, thereby forming an abstract structure that reflects the topological characteristics of the data.

[0033] Furthermore, the simplex structure is a triangular topological patch composed of three points in the downsampled point cloud data that satisfy "the distance between any two points ≤ the connection radius parameter value (such as 1.0m)" and are "non-collinear".

[0034] Multi-scale filtering is performed on the VR complex to calculate the number of annular holes, and a persistence histogram is generated using a persistent homology algorithm. It should be noted that based on the input VR complex and connection radius parameters, the scale sequence is first divided and valid annular holes are filtered within each scale sequence. Persistent holes are defined as those detected in ≥3 consecutive scales (e.g., the scale sequence 0.6m, 0.8m, and 1.0m). The final number of annular holes is the sum of persistent holes and valid annular holes. Based on the input total number of annular holes, the persistence value of each hole is first obtained. Then, the holes are unconditionally classified into fixed intervals: holes with persistence values ​​of 0m-0.2m are assigned to the first interval, holes with persistence values ​​of 0.2m-0.4m to the second interval, and holes with persistence values ​​of 0.4m-0.6m to the third interval. Finally, a persistence histogram is generated.

[0035] Furthermore, annular holes are a common defect type in engineering structures. These defects primarily include standard concentric holes (regular circles that cause stress concentration and reduced fatigue life), eccentric and isocentric holes (center offsets that cause asymmetric failure and crack initiation), and multiply connected nested holes (multiple holes intersecting, significantly reducing structural stability and load-bearing capacity). Annular holes can severely impact the mechanical properties and service life of structures through stress concentration effects, asymmetric load distribution, and topological weakening. Targeted repairs require nondestructive testing and topological optimization.

[0036] According to the persistence histogram, a three-dimensional topological feature matrix is ​​generated through spatial grid mapping; It should be noted that the interval index, interval median and number of holes of the persistence histogram are mapped to the X-axis, Y-axis and Z-axis of the three-dimensional space and distributed to a three-dimensional grid of fixed size to generate a three-dimensional topological feature matrix, which is used to quantify the geometric distribution law of the topological features.

[0037] S4. Construct a defect detection model, and obtain defect location coordinates and engineering structure risk scores based on the generated three-dimensional topological feature matrix combined with multimodal fusion features and coordinate mapping table; The three-dimensional topological feature matrix and the multimodal fusion feature are aligned in a spatial grid through a coordinate mapping table to obtain the defect type and defect location coordinates of the engineering structure; It should be noted that the multimodal fusion features are mapped to the X, Y, and Z axes of three-dimensional space through linear mapping, and grid coordinates are obtained. The grid coordinates are directly converted into defect location coordinates through a coordinate mapping table. The grid topology eigenvalues ​​can be directly obtained based on the three-dimensional topological feature matrix and grid coordinates. Then, a defect type feature library is established based on the grid topology eigenvalues. For example, 0.3 < grid topology eigenvalue < 0.5 indicates slag inclusion, 0.5 < grid topology eigenvalue < 0.7 indicates pores, and 0.7 < grid topology eigenvalue < 1.0 indicates cracks. By mapping the topological eigenvalues ​​to the defect type feature library, the defect type of the engineering structure can be obtained.

[0038] Based on the defect type of the engineering structure, the positioning coordinates are converted into the engineering coordinate system through the least squares coordinate transformation matrix and the defect density is obtained through grid space division; It should be noted that the defect location coordinates are first converted into homogeneous coordinate form, and then the coordinate transformation is performed through the pre-calibrated least squares transformation parameters. After the transformation, it is necessary to verify that the Z-axis coordinate does not exceed the structure height limit and ensure that the rotation matrix meets the orthogonality condition, and finally obtain the engineering coordinate system; according to the defect location coordinates, the engineering coordinates are mapped to all similar defects in the preset grid, the number of defects in each grid is counted, and the density calculation is performed based on the number of defects in the unit volume to obtain the defect density.

[0039] Furthermore, the least squares coordinate transformation matrix is ​​a method used to accurately transform the local coordinates collected by defect detection equipment into the global coordinate system of the project (such as the architectural drawing or BIM model coordinate system). By selecting a set of control points with known corresponding positions, the optimal rotation, translation, and scaling parameters are calculated to minimize the matching error between the two coordinate systems. For example, in steel structure inspection, the least squares coordinate transformation matrix can ensure that the deviation between the scanned crack position and the coordinates on the CAD model does not exceed 2 mm, thereby ensuring the accuracy of defect positioning and providing a reliable spatial reference for subsequent grid analysis and risk assessment. The least squares coordinate transformation matrix is ​​often used in engineering for multi-sensor data alignment and construction deviation correction.

[0040] Furthermore, the preset grid first divides the global space according to the maximum envelope size of the structural parts, adopts a dynamic voxelization strategy to set high-resolution grids at welds and bolt connections, uses sparse grids in non-critical areas, and adaptively encrypts the grid through the probability distribution of the welding path to ensure that each grid cell contains at least 3 or more detection point clouds, and finally generates a topologically consistent grid.

[0041] The defect type weight is obtained by linear regression based on the engineering structure defect density and the frequency of the engineering structure defect type appearing in each area; It should be noted that a multivariate data set is established in combination with engineering parameters. First, key engineering variables (such as material yield strength, welding current, and vibration frequency) are extracted as static features and spatially and temporally aligned with dynamic detection data (such as three-dimensional coordinates of defects, thermal field gradients, and acoustic spectra). Then, a variable correlation matrix is ​​established based on physical constraints, and principal component analysis is used to eliminate multicollinearity between parameters. The defect density is used as the target variable, and the engineering parameters are used as the characteristic variables. The regression coefficients are obtained through the linear regression equation, and the weight values ​​of each defect type are extracted from the regression coefficients.

[0042] Construct a defect detection model based on the engineering structure defect type, defect density and defect type weight through an exponential function; It should be noted that, first, the measured defect density of the engineering structure defect type is normalized to eliminate the dimensional differences of different engineering structure defect types; second, the weighted defect density is obtained by normalizing the defect density and the defect type weight to reflect the difference in risk contribution of various types of engineering structure defects; then, the weighted defect density of all defect types is summed to obtain a comprehensive risk index; the sensitivity can be adjusted according to the engineering type, so that the defect detection model can dynamically amplify the cumulative effect of high-risk defects, thereby constructing a defect detection model.

[0043] Based on the defect detection model and combined with multimodal fusion features and defect location coordinates, the neighborhood risk propagation algorithm is used to calculate the engineering structure risk score. The expression is: , in, The final output of the engineering structure risk score is Cumulative risk score of the largest continuous high-risk area Benchmark normalization constant, used to adjust the scoring scale, Reflects the grid risk value of the criticality of the material location, No. The original risk value of each grid is calculated by the neighborhood risk propagation algorithm. For the The volume of the grid, in cubic meters, No. The distance from the grid to the nearest key area, Indicates the total number of preset grids.

[0044] It should be noted that the risk propagation factor of each grid is first obtained based on the defect detection model combined with the multimodal fusion features, and then iterative calculation is performed through the 6-neighborhood (up, down, left, right, front, and back) propagation algorithm. Each grid receives the risk value propagated by the neighborhood; the propagation steps are repeated until the maximum risk change between two adjacent times reaches the upper limit of 20 times and the engineering structure risk score is output.

[0045] Furthermore, the risk propagation factor is a dynamic weight parameter that quantifies the risk diffusion intensity of grid unit defects. Its calculation integrates multimodal characteristics and engineering constraints: it is weighted by the original risk probability, thermal field gradient amplitude and acoustic energy attenuation rate output by the defect detection model, and then multiplied by the material sensitivity coefficient (such as the brittleness coefficient of Q345 steel is 1.2) and the position weight (such as the weight of 1.5 at 0.1m from the weld) to finally generate the risk propagation factor.

[0046] S5. Dynamically adjust construction parameters based on the project structure risk score and generate a project supervision report.

[0047] Based on historical structural risk event data, define low risk threshold R1 and high risk threshold R2; It should be noted that low risk (<30 points): accident rate <5% (such as buildings in normal use); medium risk (30-70 points): accident rate 5%-20% (such as bridges with local cracks); high risk (≥70 points): accident rate >20% (such as severely corroded chemical pipelines).

[0048] Furthermore, the accident rate refers to the historical statistical probability of safety accidents occurring during the actual use of engineering structures. This data is derived from long-term enterprise operation and maintenance records monitored by the supervision industry. By analyzing actual damage cases of similar projects with similar defect characteristics, Bayesian statistical methods are used to calculate the conditional probability of each risk interval.

[0049] When the engineering structure risk score is less than R1, the engineering structure is considered to be in a low-risk state; when R1≤engineering structure risk score<R2, the engineering structure is considered to be in a medium-risk state; when the engineering structure risk score is greater than or equal to R2, the engineering structure is considered to be in a high-risk state.

[0050] It should be explained that, for example, based on historical structural risk event data, a low-risk threshold R1 = 30 points and a high-risk threshold R2 = 70 points are set; after calculation, the current risk score of a certain engineering structure is 55 points. Since 30 points ≤ score < 70 points, the structure is judged to be in a medium-risk state; if the current risk score = 25 points, the score < 30 points, the structure is judged to be in a low-risk state; if the current risk score = 80 points, 70 ≤ score, the structure is judged to be in a high-risk state.

[0051] Construction parameters include grouting pressure, curing cycle and equipment vibration frequency; It should be noted that the grouting pressure is monitored in real time by pressure sensors to ensure that the slurry is fully filled without damaging the structure; the curing cycle is determined based on material tests to ensure that the strength meets the standards and prevent construction delays; the equipment vibration frequency is measured by vibration sensors to optimize the density of the concrete and prevent segregation.

[0052] Based on the defect location coordinates, combined with the construction strategy library, dynamic adjustment instructions are generated and construction parameters are adjusted; It should be noted that based on the defect location coordinates and risk level (such as medium risk 55 points), the optimization plan in the construction strategy library is first matched (such as the crack area requires grouting pressure +10%), and dynamic adjustment instructions are generated (such as increasing the grouting pressure from 0.4MPa to 0.44MPa, and adjusting the vibration frequency from 70Hz to 85Hz), and the execution effect is monitored through real-time sensors.

[0053] Furthermore, the construction strategy library is a rule database built based on historical engineering data and expert experience. Its core function is to establish a standardized mapping relationship between the types of engineering structural defects, risk levels, and construction parameter adjustment plans. Specifically, it contains three types of key data: (1) parameter adjustment coefficients for different defect types (such as cracks, hollows, and slag inclusions) in different risk score ranges; (2) baseline construction parameters for various engineering structural parts; and (3) effective range rules for parameter adjustments (such as local adjustments require the specified grid number G-205). This is achieved by real-time matching of currently detected defect characteristics with risk status.

[0054] The engineering supervision report is generated through summary logic based on the defect location coordinates, engineering structure risk score, multimodal fusion features and adjusted construction parameters.

[0055] It should be noted that first, all input data are converted into a unified dimension based on the defect location coordinates, engineering structure risk score, and multimodal fusion features. Then, the multimodal features are matched with the defect types to compare the construction parameters before and after adjustment (such as grouting pressure from 0.4→0.44MPa), and the range of change is recorded. Finally, core indicators such as the proportion of high-risk areas and the parameter adjustment compliance rate are counted to generate a standard supervision report.

[0056] This embodiment also provides an artificial intelligence-based intelligent engineering supervision system, including: Multimodal generation module, used to collect multimodal fusion data and perform preprocessing to generate multimodal data packets and coordinate mapping tables; The cross-modal graph fusion module is used to extract three-dimensional field frequency features based on multimodal data packets and coordinate mapping tables, and fuse them with multimodal graph neural networks to generate multimodal fusion features; Topology extraction module, used to construct VR complex based on multimodal fusion features and coordinate mapping table, and generate three-dimensional topological feature matrix through continuous homology analysis; The dual-drive module is used to build a defect detection model, which generates a three-dimensional topological feature matrix and combines multimodal fusion features and coordinate mapping tables to obtain defect location coordinates and engineering structure risk scores; The report generation module is used to dynamically adjust construction parameters according to the project structure risk score and generate a project supervision report.

[0057] This embodiment also provides a computer device suitable for the case of an intelligent engineering supervision method based on artificial intelligence, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the intelligent engineering supervision method based on artificial intelligence proposed in the above embodiment.

[0058] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0059] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the artificial intelligence-based smart engineering supervision method proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0060] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, and all of these should be included in the scope of protection of the present invention.

Claims

1. A smart engineering supervision method based on artificial intelligence, characterized in that: The following steps are involved: S1. Collect multimodal fusion data and perform preprocessing to generate multimodal data packets and coordinate mapping tables; S2. Based on the multimodal data packet and coordinate mapping table, extract the three-dimensional field frequency features and fuse them with the multimodal graph neural network to generate multimodal fusion features; S3, construct VR complex based on multimodal fusion features and coordinate mapping table, and generate three-dimensional topological feature matrix through continuous homology analysis; S4. Construct a defect detection model, and obtain defect location coordinates and engineering structure risk scores based on the generated three-dimensional topological feature matrix combined with multimodal fusion features and coordinate mapping table; S5. Dynamically adjust construction parameters based on the project structure risk score and generate a project supervision report.

2. The method for intelligent engineering supervision based on artificial intelligence according to claim 1, characterized in that: In S1, the multimodal fusion data includes downsampled point cloud data, infrared thermal field distribution characteristics and acoustic emission frequency domain characteristics; Based on the multimodal fusion data, multimodal data packets and coordinate mapping tables are generated through preprocessing through multimodal spatiotemporal alignment and feature fusion.

3. The method for intelligent engineering supervision based on artificial intelligence according to claim 2, characterized in that: In S2, based on the multimodal data packet and coordinate mapping table, the three-dimensional field frequency features are extracted and fused with the multimodal graph neural network to generate multimodal fusion features, including: Based on multimodal data packets and coordinate mapping tables, spatial node feature fusion is used to construct a cross-modal graph structure, and three-dimensional field frequency features are extracted through graph convolution aggregation; Based on three-dimensional field frequency features, multimodal graph neural network is combined to generate multimodal fusion features through neighborhood feature propagation.

4. The method for intelligent engineering supervision based on artificial intelligence according to claim 3, characterized in that: In S3, a VR complex is constructed based on multimodal fusion features and coordinate mapping tables, and a three-dimensional topological feature matrix is ​​generated through continuous homology analysis, including: Define the connection radius parameters based on the spatial constraint parameters in the coordinate mapping table; Based on the downsampled point cloud data and the connection radius parameter, the simplex structure is determined by the neighborhood point pairs, and the VR complex is generated by dimensionally expanding it. Multi-scale filtering is performed on the VR complex to calculate the number of annular holes, and a persistence histogram is generated using a persistent homology algorithm. The persistence histogram is mapped through spatial gridding to generate a three-dimensional topological feature matrix.

5. The method for intelligent engineering supervision based on artificial intelligence according to claim 4, characterized in that: In S4, a defect detection model is constructed. The defect location coordinates and engineering structure risk scores are obtained based on the generated three-dimensional topological feature matrix combined with multimodal fusion features and coordinate mapping tables, including: The three-dimensional topological feature matrix and the multimodal fusion feature are aligned in a spatial grid through a coordinate mapping table to obtain the defect type and defect location coordinates of the engineering structure; Based on the defect type of the engineering structure, the defect location coordinates are converted into the engineering coordinate system through the least squares coordinate transformation matrix, and the defect density is obtained through grid space division; Based on the density of engineering structure defects and the frequency of engineering structure defect types appearing in each construction area, the defect type weight is obtained through linear regression; A defect risk scoring model is constructed using an exponential function based on the engineering structure defect type, defect density and defect type weight. The engineering structure risk score is calculated using a neighborhood risk propagation algorithm based on the defect detection model and combined with multimodal fusion features and defect location coordinates.

6. The method for intelligent engineering supervision based on artificial intelligence according to claim 5, characterized in that: In S5, construction parameters are dynamically adjusted based on the engineering structure risk score, including: Based on historical structural risk event data, define low risk threshold R1 and high risk threshold R2; When the engineering structure risk score is less than R1, the engineering structure is considered to be in a low-risk state; when R1≤engineering structure risk score<R2, the engineering structure is considered to be in a medium-risk state; when the engineering structure risk score is greater than or equal to R2, the engineering structure is considered to be in a high-risk state; Based on the defect location coordinates, dynamic adjustment instructions are generated in combination with the construction strategy library, and construction parameters are adjusted.

7. The method for intelligent engineering supervision based on artificial intelligence according to claim 6, characterized in that: In S5, a project supervision report is generated, including: Construction parameters include grouting pressure, curing cycle and equipment vibration frequency; The engineering supervision report is generated through summary logic based on the defect location coordinates, engineering structure risk score, multimodal fusion features and adjusted construction parameters.

8. An artificial intelligence-based smart engineering supervision system, based on the artificial intelligence-based smart engineering supervision method according to any one of claims 1 to 7, characterized in that: include, Multimodal generation module, used to collect multimodal fusion data and perform preprocessing to generate multimodal data packets and coordinate mapping tables; The cross-modal graph fusion module is used to extract three-dimensional field frequency features based on multimodal data packets and coordinate mapping tables, and fuse them with multimodal graph neural networks to generate multimodal fusion features; Topology extraction module, used to construct VR complex based on multimodal fusion features and coordinate mapping table, and generate three-dimensional topological feature matrix through continuous homology analysis; The dual-drive module is used to build a defect detection model, which generates a three-dimensional topological feature matrix and combines multimodal fusion features and coordinate mapping tables to obtain defect location coordinates and engineering structure risk scores; The report generation module is used to dynamically adjust construction parameters according to the project structure risk score and generate a project supervision report.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the intelligent engineering supervision method based on artificial intelligence described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the intelligent engineering supervision method based on artificial intelligence described in any one of claims 1 to 7 are implemented.

Images