A Method and System for Intelligent Identification of Construction Risk Behaviors of Steel Trestle Bridges Based on Feature Mining

Abstract

This invention relates to an intelligent identification method and system for risk behaviors in steel trestle construction based on feature mining. The method dynamically represents the component installation status information from the steel trestle construction progress data, and calculates the bearing capacity reduction by combining the component-section bearing capacity contribution matrix pre-generated by the finite element model, generating a local ultimate bearing capacity distribution. Simultaneously, it identifies heavy machinery targets based on construction images collected by a multi-view vision system and performs three-dimensional spatial positioning, generating a dynamic load-position matrix for the current moment. The bearing capacity distribution and load matrix are input into a coupled risk digital twin model to quantify the structure-behavior coupled risk, calculating the first coupled risk value for the current moment. Then, based on the time series database of the mechanical target's motion parameters, its future position and load are predicted, and a second coupled risk value is calculated by combining the bearing capacity distribution for the future moment. By comparing the two risk values, the high-risk behavioral intent is determined and an identification result is generated.

Description

Technical Field

[0001] This invention relates to the field of construction safety technology, specifically to an intelligent identification method and system for risk behaviors in steel trestle construction based on feature mining. Background Technology

[0002] Steel trestle bridges, serving as temporary construction passages across rivers and canyons in water conservancy, hydropower, and transportation bridge projects, are widely used in engineering construction due to their simple structure, convenient assembly and disassembly, and strong load-bearing capacity. These temporary structures are typically assembled from components such as Bailey beams, transverse connecting systems, and bridge decks. Their construction process exhibits significant staged characteristics—structural integrity gradually develops as construction progresses, with substantial differences in load-bearing capacity at each stage, from the initial Bailey beam erection to the final bridge deck paving. During trestle bridge construction, heavy construction machinery such as crawler excavators and concrete mixer trucks frequently operate on the bridge. The matching relationship between their dynamic loads and the trestle bridge's staged load-bearing capacity directly affects construction safety; therefore, real-time safety monitoring of the construction process is crucial.

[0003] However, existing safety monitoring technologies for steel trestle bridge construction have significant limitations. On the one hand, traditional structural health monitoring mainly focuses on static or quasi-static indicators such as stress, deformation, and vibration of the trestle itself, using finite element analysis or sensor networks for periodic assessments. These methods struggle to establish a dynamic correlation with real-time changes in construction behavior. On the other hand, while vision-based construction behavior monitoring systems can identify the type and location of mechanical targets, their analysis logic assumes the trestle is always in its complete design state, failing to perceive the phased decrease in load-bearing capacity due to changes in construction progress. When the trestle is in an incomplete state, such as before the lateral connection system is installed or the bridge deck is not fully laid, its actual load-bearing capacity is far lower than the design value. If heavy machinery operates on the bridge at this time, existing monitoring systems, lacking the ability to perceive the dynamic load-bearing state of the structure, cannot couple the machinery's movement with the structure's safety margin for early warning, making it difficult to predict major safety risks such as overload collapse. Summary of the Invention

[0004] Based on this, the purpose of this invention is to provide a method and system for intelligent identification of construction risk behavior of steel trestle bridges based on feature mining, which can perform correlation analysis between the dynamic load-bearing state of steel trestle bridge structure and the construction behavior of heavy machinery.

[0005] The objective of this invention is achieved through the following solution:

[0006] In a first aspect, the present invention provides an intelligent identification method for construction risk behaviors of steel trestle bridges based on feature mining, comprising the following steps:

[0007] S1: Dynamically characterize the component installation status information in the acquired steel trestle bridge construction progress data, call the component-section bearing capacity contribution matrix pre-generated by the finite element model to perform bearing capacity reduction calculation on the component installation status information, and generate local ultimate bearing capacity distribution;

[0008] S2: The construction images collected in real time by the multi-view vision system deployed at key nodes of the steel trestle are processed for heavy machinery target identification and three-dimensional spatial positioning. The type and model of the machinery target are identified and the real-time projection position of the machinery target on the trestle bridge surface is determined. The curb weight and rated load parameters of the machinery target are matched from the preset machinery parameter database to generate the dynamic load-position matrix at the current moment.

[0009] S3: Input the local ultimate bearing capacity distribution and dynamic load-location matrix into the preset coupling risk digital twin model to quantify the structural-behavioral coupling risk, calculate the moment contribution coefficient of each load point to the current weakest section, and calculate the first coupling risk value at the current moment in combination with the local ultimate bearing capacity distribution;

[0010] S4: Based on the mechanical parameter database, predict the motion trajectory of the mechanical target to determine its behavioral intent. Based on the historical motion parameters, predict the predicted position and predicted load of each mechanical target at future time. Combine this with the local ultimate bearing capacity distribution to calculate the second coupling risk value at future time.

[0011] S5: Compare the first coupling risk value with the second coupling risk value. When the second coupling risk value reaches the preset danger threshold and the first coupling risk value does not reach the preset danger threshold, it is determined that the mechanical target has a high-risk behavioral intention, and a high-risk behavioral intention recognition result is generated.

[0012] Secondly, this invention provides an intelligent identification system for risk behaviors in steel trestle bridge construction based on feature mining. This system is configured with the following modules:

[0013] The component bearing capacity calculation module is used to dynamically represent the component installation status information in the acquired steel trestle bridge construction progress data. It calls the component-section bearing capacity contribution matrix pre-generated by the finite element model to perform bearing capacity reduction calculation on the component installation status information and generate local ultimate bearing capacity distribution.

[0014] The dynamic load positioning module is used to identify heavy machinery targets and perform three-dimensional spatial positioning processing on the construction images collected in real time by the multi-view vision system deployed at key nodes of the steel trestle. It identifies the type and model of the machinery target and determines the real-time projection position of the machinery target on the trestle bridge surface. It also matches the curb weight and rated load parameters of the machinery target from the preset machinery parameter database to generate the dynamic load-position matrix at the current moment.

[0015] The coupling risk quantification module is used to input the local ultimate bearing capacity distribution and dynamic load-location matrix into the preset coupling risk digital twin model to quantify the structure-behavior coupling risk, calculate the moment contribution coefficient of each load point to the current weakest section, and calculate the first coupling risk value at the current moment in combination with the local ultimate bearing capacity distribution.

[0016] The risk trend prediction module is used to predict the behavioral intent of the mechanical target's motion trajectory based on the mechanical parameter database, predict the predicted position and predicted load of each mechanical target at future time based on historical motion parameters, and calculate the second coupled risk value at future time by combining the local ultimate bearing capacity distribution.

[0017] The high-risk intent recognition module is used to compare and process the first coupled risk value and the second coupled risk value. When the second coupled risk value reaches the preset danger threshold and the first coupled risk value does not reach the preset danger threshold, it is determined that the mechanical target has a high-risk behavioral intent and a high-risk behavioral intent recognition result is generated.

[0018] Thirdly, this application provides a computer device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements any of the above-mentioned intelligent identification methods for construction risk behaviors of steel trestle bridges based on feature mining.

[0019] Fourthly, this application provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements any of the above-mentioned intelligent identification methods for construction risk behaviors of steel trestle bridges based on feature mining.

[0020] In summary, the intelligent identification method for construction risk behavior of steel trestle bridges based on feature mining provided in this application integrates construction progress data and visual perception information to achieve coupled analysis of the dynamic load-bearing state of the steel trestle bridge structure and the construction behavior of heavy machinery. This addresses the technical problem in existing monitoring systems where structural state monitoring and behavior recognition are disconnected, leading to an inability to predict the risk of overload collapse. By constructing a local ultimate bearing capacity distribution that evolves with the construction progress and simultaneously identifying the real-time load position of the machinery, a precise data foundation for risk quantification can be provided. By calculating and comparing the first coupled risk value at the current moment with the second coupled risk value at a future moment, risk prediction of the machinery's movement trend can be achieved, identifying potential dangers that may be caused by machinery operation when the structure is not yet completed. By identifying and outputting high-risk behavioral intentions, forward-looking early warning information can be provided to the construction site to assist safety management personnel in taking intervention measures before risks occur, thereby improving the level of intelligent safety management in the construction process of steel trestle bridges.

[0021] To better understand and implement this invention, the following detailed description is provided in conjunction with the accompanying drawings. Attached Figure Description

[0022] Figure 1 A flowchart illustrating an intelligent identification method for construction risk behaviors of steel trestle bridges based on feature mining, provided in an embodiment of this application;

[0023] Figure 2 This is a structural schematic diagram of a feature mining-based intelligent identification system for construction risk behavior of steel trestle bridges, provided as another embodiment of this application. Detailed Implementation

[0024] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Preferred embodiments of the invention are shown in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a thorough and complete understanding of the disclosure of the invention.

[0025] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0026] In one embodiment, such as Figure 1 As shown, a method for intelligent identification of construction risk behaviors in steel trestle bridges based on feature mining is provided. This embodiment illustrates the application of this method to a terminal. It is understood that this method can also be applied to a server, or to a system including both a terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0027] S1: Dynamically characterize the component installation status information in the acquired steel trestle bridge construction progress data, call the component-section bearing capacity contribution matrix pre-generated by the finite element model to perform bearing capacity reduction calculation on the component installation status information, and generate local ultimate bearing capacity distribution.

[0028] Specifically, the system acquires construction progress data of the steel trestle bridge and dynamically represents the component installation status information in the data. The system calls a pre-generated component-section bearing capacity contribution matrix from the finite element model and uses this matrix to calculate the bearing capacity reduction of the component installation status information. The finite element model is constructed based on the steel trestle bridge design drawings. The system simulates the structural stress state under different component installation combinations using the finite element model, obtaining the contribution weight of each component to the ultimate bearing capacity of different sections of the trestle bridge, thus forming the component-section bearing capacity contribution matrix. This matrix is ​​a two-dimensional matrix, with row vectors corresponding to the core components of the steel trestle bridge, column vectors corresponding to the key longitudinal and transverse sections of the trestle bridge, and matrix elements representing the contribution coefficients of the corresponding components to the ultimate bearing capacity of the corresponding sections. The system analyzes the component installation status information, extracting key parameters such as the installation completion degree, connection tightness, and installation position deviation of the core components. Preferably, the bearing capacity reduction calculation can use a weighted reduction method. The system uses the extracted parameters as the basis for reduction and corrects the corresponding contribution coefficients in the component-section bearing capacity contribution matrix. The system calculates the ultimate bearing capacity of each section of the steel trestle by superimposing the corrected contribution coefficients and combining them with the actual installation quantity and distribution of each component. The system constructs a two-dimensional local ultimate bearing capacity distribution using the longitudinal and transverse coordinates of the trestle as indices. This distribution reflects the upper limit of the actual bearing capacity at each location of the steel trestle under the current construction progress.

[0029] The system continuously updates the construction progress data of the steel trestle bridge, dynamically adjusting the representation results of component installation status information to ensure consistency between component installation status information and actual construction progress. The system performs real-time verification of the corrected component-section bearing capacity contribution matrix, identifying deviations in the matrix calculation process and ensuring the accuracy of the matrix data. When calculating the ultimate bearing capacity of each section of the steel trestle bridge, the system integrates the installation distribution information of each core component to avoid calculation deviations caused by differences in component installation positions. After constructing the local ultimate bearing capacity distribution, the system performs integrity verification on the distribution data to ensure that the distribution data covers all key locations of the steel trestle bridge, providing reliable structural bearing capacity data support for subsequent coupling risk quantification. The system stores the local ultimate bearing capacity distribution data in a structured manner for easy subsequent retrieval and updates, and establishes a data association mechanism to achieve synchronous linkage between the local ultimate bearing capacity distribution and construction progress data, ensuring data real-time performance and consistency.

[0030] S2: The construction images collected in real time by the multi-view vision system deployed at key nodes of the steel trestle are processed for heavy machinery target identification and three-dimensional spatial positioning. The type and model of the machinery target are identified and the real-time projection position of the machinery target on the trestle surface is determined. The mechanical target's curb weight and rated load parameters are matched from the preset mechanical parameter database to generate the dynamic load-position matrix at the current moment.

[0031] Specifically, the system controls a multi-view vision system deployed at key nodes of the steel trestle bridge to acquire construction images in real time. The multi-view vision system consists of multiple industrial cameras. The system controls the camera installation positions according to a preset deployment plan, ensuring that the shooting range covers the entire trestle bridge surface, with overlapping shooting areas between adjacent cameras to ensure the continuity and accuracy of 3D positioning. After deployment, the system uses calibration methods to calibrate the camera's intrinsic and extrinsic parameters, obtaining the camera's intrinsic parameter matrix, distortion coefficients, and extrinsic parameter matrix for subsequent image correction and 3D coordinate transformation. The system preprocesses the acquired construction images, removing image noise through filtering algorithms, optimizing image quality through contrast enhancement algorithms, and correcting camera distortion through distortion correction algorithms to ensure that the image data meets the requirements for subsequent target recognition and positioning. The system uses a deep learning target detection algorithm to identify heavy machinery targets in the preprocessed images, extracting key features of the machinery targets to identify their categories, and outputting the bounding box coordinates of the machinery targets in the image.

[0032] Furthermore, based on the principle of multi-view visual triangulation and combined with camera calibration parameters, the system calculates the three-dimensional world coordinates of the mechanical target. Then, through a coordinate transformation algorithm, it converts the three-dimensional world coordinates into a two-dimensional projection position in the coordinate system of the steel trestle bridge deck. Based on the identified mechanical target category and model, the system matches it against a pre-set mechanical parameter database, retrieving core parameters such as the target's curb weight and rated load, and determines the actual load according to the machine's operating status. The system constructs a dynamic load-position matrix using the current time as a timestamp. This matrix records the category, model, bridge deck projection coordinates, and actual load value of each heavy machinery target and updates it in real time. The system verifies the mechanical target identification results, checks for identification errors, and ensures the accuracy of mechanical target category and model identification. The system updates the dynamic load-position matrix in real time, synchronizing changes in the mechanical target's position and load, ensuring that the matrix data accurately reflects the current construction load status.

[0033] S3: Input the local ultimate bearing capacity distribution and dynamic load-location matrix into the preset coupled risk digital twin model to quantify the structural-behavioral coupled risk, calculate the moment contribution coefficient of each load point to the current weakest section, and calculate the first coupled risk value at the current moment in combination with the local ultimate bearing capacity distribution.

[0034] Specifically, the system inputs the local ultimate bearing capacity distribution and dynamic load-location matrix into a pre-defined coupled risk digital twin model for structure-behavior coupled risk quantification. The coupled risk digital twin model uses the physical entity of the steel trestle bridge as a prototype, recreating its structural dimensions, component distribution, and construction progress, integrating functions related to structural bearing capacity, load analysis, and risk quantification. Through the load analysis functions of the coupled risk digital twin model, the system calculates the bending moment values ​​generated by each load point on each section of the steel trestle bridge. Combined with the local ultimate bearing capacity distribution, it determines the weakest section of the steel trestle bridge at the current moment. The system verifies the calculated bending moment values, checking for deviations in the calculation process to ensure the accuracy of the bending moment data. The system compares the ultimate bearing capacity of each section of the steel trestle bridge based on the local ultimate bearing capacity distribution, selecting the section with the lowest ultimate bearing capacity as the weakest section, providing a clear analysis object for subsequent risk quantification. The system stores the relevant data of the weakest section for subsequent calculation of the bending moment contribution coefficient and the first coupled risk value.

[0035] Preferably, the system can employ the moment contribution coefficient method to calculate the moment contribution coefficient of each load application point on the weakest section. This coefficient is the ratio of the moment value generated by a single load application point on the weakest section to the total moment value generated by all load application points on that section. The system summarizes the calculated moment contribution coefficients to ensure that the moment contribution coefficients of all load application points are calculated completely and without omission. The system combines the ultimate moment value of the weakest section in the local ultimate bearing capacity distribution to calculate the first coupling risk value at the current moment. The calculation method is the ratio of the sum of the moment contribution coefficients of each load application point and the moment values ​​generated by the corresponding loads to the ultimate moment value of the weakest section. This value reflects the real-time impact of the current construction activity on the structural safety of the steel trestle bridge. The system verifies the calculation process of the first coupling risk value, checks for calculation deviations, and ensures the accuracy of the first coupling risk value.

[0036] S4: Based on the mechanical parameter database, predict the motion trajectory of the mechanical target to determine its behavioral intent. Based on historical motion parameters, predict the predicted position and load of each mechanical target at future time. Combine this with the local ultimate bearing capacity distribution to calculate the second coupling risk value at future time.

[0037] Specifically, the system predicts the behavioral intent of mechanical targets based on a mechanical parameter database and historical motion parameters of the mechanical targets. The system extracts historical motion parameters of the mechanical targets collected by a multi-view vision system, including historical position coordinates, motion speed, acceleration, and motion direction, to construct a motion trajectory prediction model. Using this model, combined with relevant parameters of the mechanical targets in the mechanical parameter database and the construction procedures of the steel trestle bridge, the system corrects the future motion trajectory of the mechanical targets, improving the rationality and accuracy of the prediction. The system filters the extracted historical motion parameters, removing outliers to ensure the reliability of the historical motion parameters and provide high-quality data support for the construction of the motion trajectory prediction model. The system verifies the motion trajectory prediction model, checking for prediction deviations to ensure the rationality of the model's prediction results. The system adjusts the predicted motion trajectory in conjunction with the steel trestle bridge construction procedures to ensure that the prediction results are consistent with actual construction needs, providing a basis for determining the subsequent predicted location and predicted load.

[0038] The system uses a motion trajectory prediction model to output the bridge deck projection coordinates of each mechanical target at different future times, serving as the predicted position. Based on the current operating status of the mechanical targets, combined with the equipment weight, rated load parameters, and construction procedure requirements in the mechanical parameter database, the system determines the actual load value of each mechanical target at future times, serving as the predicted load. The system substitutes the predicted position and predicted load into the coupled risk digital twin model, and, combined with the local ultimate bearing capacity distribution generated by S1, calculates the coupled risk value at each future time. The system selects the maximum coupled risk value within a future period as the second coupled risk value, reflecting the maximum safety risk that construction activities may bring within that period. The system verifies the predicted position and predicted load, checking for prediction deviations to ensure data accuracy. The system summarizes the coupled risk values ​​at each future time, selects the maximum value as the second coupled risk value, and stores this value in a structured manner for subsequent comparison with the first coupled risk value.

[0039] S5: Compare the first coupling risk value with the second coupling risk value. When the second coupling risk value reaches the preset danger threshold and the first coupling risk value does not reach the preset danger threshold, it is determined that the mechanical target has a high-risk behavioral intention, and a high-risk behavioral intention recognition result is generated.

[0040] Specifically, the system compares the first coupled risk value with the second coupled risk value, and, in conjunction with a preset danger threshold, determines whether the mechanical target has a high-risk behavioral intent, generating a high-risk behavioral intent identification result. The preset danger threshold is determined based on the design safety margin of the steel trestle bridge, relevant industry construction safety specifications, and actual construction experience, and can be adjusted according to the structural type, construction conditions, load level, and other actual conditions of the steel trestle bridge. The system compares the values ​​according to preset logic, focusing on whether the second coupled risk value reaches the preset danger threshold, while simultaneously determining whether the first coupled risk value is within a safe range. The system verifies the preset danger threshold to ensure that the threshold setting meets actual construction safety requirements, providing a reliable basis for determining high-risk behavioral intent. The system records the comparison process between the first and second coupled risk values ​​to ensure the standardization and traceability of the comparison logic. Based on the comparison results, the system preliminarily determines the risk level of the mechanical target's behavioral intent.

[0041] When the second coupled risk value reaches a preset danger threshold while the first coupled risk value does not, the system determines that the mechanical target has a high-risk behavioral intent. This means that the current construction activity is safe, but future changes in the mechanical target's trajectory and load will cause the construction risk to exceed the threshold, posing a potential safety hazard. The high-risk behavioral intent identification result generated by the system includes core information such as the mechanical target's number, category, model, current location, current load, first coupled risk value, second coupled risk value, predicted location, predicted load, and judgment basis. The identification result is output in structured data form and transmitted to the construction safety monitoring platform, providing risk alerts to on-site management personnel and facilitating control measures to prevent safety accidents. The system verifies the generated identification results, checking for omissions and errors to ensure the completeness and accuracy of the identification results. The system establishes a mechanism for transmitting the identification results to ensure that the results are transmitted to the construction safety monitoring platform in real time and accurately, guaranteeing the timely implementation of control measures.

[0042] In summary, the intelligent identification method for construction risk behavior of steel trestle bridges based on feature mining provided in this application integrates construction progress data and visual perception information to achieve coupled analysis of the dynamic load-bearing state of the steel trestle bridge structure and the construction behavior of heavy machinery. This addresses the technical problem in existing monitoring systems where structural state monitoring and behavior recognition are disconnected, leading to an inability to predict the risk of overload collapse. By constructing a local ultimate bearing capacity distribution that evolves with the construction progress and simultaneously identifying the real-time load position of the machinery, a precise data foundation for risk quantification can be provided. By calculating and comparing the first coupled risk value at the current moment with the second coupled risk value at a future moment, risk prediction of the machinery's movement trend can be achieved, identifying potential dangers that may be caused by machinery operation when the structure is not yet completed. By identifying and outputting high-risk behavioral intentions, forward-looking early warning information can be provided to the construction site to assist safety management personnel in taking intervention measures before risks occur, thereby improving the level of intelligent safety management in the construction process of steel trestle bridges.

[0043] In one embodiment, S1 of the intelligent identification method for construction risk behavior of steel trestle bridge based on feature mining provided by the present invention specifically includes the following steps:

[0044] S11: Perform component classification and spatial positioning analysis on the component list and BIM model in the acquired steel trestle construction progress data. According to the structural stress characteristics, each component is divided into main beam components, transverse connection components and bridge deck system components. The three-dimensional spatial occupancy range of each component in the trestle coordinate system is extracted from the BIM model to generate a component classification list with spatial labels.

[0045] Specifically, the system acquires the construction progress data of the steel trestle bridge, extracts the component list and BIM model from it, and performs component classification and spatial positioning analysis on both. First, the system sorts all components in the component list, classifying them into main beam components, transverse connection components, and bridge deck components based on the structural stress characteristics of the steel trestle bridge. The system then calls the BIM model, analyzes it, and extracts the three-dimensional spatial occupancy range of each component in the trestle bridge coordinate system. This three-dimensional spatial occupancy range reflects the specific spatial location range of each component within the trestle bridge structure. The system associates the component classification results with the corresponding three-dimensional spatial occupancy ranges, adding a spatial label to each component. The spatial label contains the component's category and three-dimensional spatial occupancy range information. Finally, the system integrates all spatially labeled component information to generate a component classification list with spatial labels, which clearly presents the category and spatial distribution of each component.

[0046] S12: Perform cross-sectional discretization association processing on the component classification list with spatial labels, divide multiple key cross-sections along the length of the trestle at preset intervals, and establish a set of component indexes for each key cross-section based on the intersection relationship between the spatial occupancy interval of each component and the key cross-section, generating a cross-section-component association mapping table.

[0047] Specifically, the system performs cross-sectional discretization and association processing on the component classification list with spatial labels. The system divides the trestle into multiple key sections along its length at preset intervals. The division of key sections must cover the entire length of the trestle to ensure a comprehensive reflection of the structural stress state at each location. The system extracts the spatial occupancy range of each component, analyzes the intersection relationship between the spatial occupancy range of each component and each key section, and determines which components overlap with each key section. The system establishes a component index set for each key section, containing the identification information of all components that intersect with that key section. The system integrates all key sections and their corresponding component index sets to generate a section-component association mapping table, which clarifies the correspondence between each key section and its components.

[0048] S13: Perform sensitivity analysis on the component contribution coefficient of the section-component association mapping table and the preset finite element model. In the finite element model, each component is masked one by one and nonlinear buckling simulation is performed. Record the decrease in bearing capacity at each key section relative to the intact state and generate the initial contribution coefficient matrix of each component to each key section.

[0049] Specifically, the system performs sensitivity analysis on the component contribution coefficients of the section-component association mapping table and the preset finite element model. The system first calls the preset finite element model, which has been pre-built and stores relevant structural parameters. Following the component index in the section-component association mapping table, the system masks each component one by one in the finite element model. After masking a single component, the system performs a nonlinear buckling simulation to simulate the stress condition of the trestle structure under component-deficient conditions. The system records the bearing capacity data at each critical section during each simulation and compares this data with the bearing capacity data at the critical sections under the complete state of the trestle in the finite element model, calculating the decrease in bearing capacity at each critical section relative to the complete state. Based on the decrease, the system determines the initial contribution coefficient of each component to each critical section, integrates all initial contribution coefficients, and generates an initial contribution coefficient matrix for each component to each critical section.

[0050] S14: Perform continuous interpolation expansion on the initial contribution coefficient matrix, and use a piecewise interpolation function to fit the contribution coefficients at discrete key sections into a function form that is continuously distributed along the bridge length, thereby generating the component-section bearing capacity contribution matrix.

[0051] Specifically, the system performs continuous interpolation expansion on the initial contribution coefficient matrix. The contribution coefficients in the initial matrix correspond to discrete key sections and cannot reflect the continuous variation along the bridge length. The system uses a piecewise interpolation function to fit the contribution coefficients at the discrete key sections, transforming them into a function form that is continuously distributed along the bridge length. During the fitting process, the system reasonably divides the interpolation segments based on the distribution of each key section, ensuring that the fitted continuous function accurately reflects the variation of the contribution coefficients along the bridge length. Through piecewise interpolation fitting, the system obtains the contribution coefficients of each component to any section along the bridge length. Integrating these contribution coefficients generates a component-section bearing capacity contribution matrix, which provides continuously distributed component contribution coefficient data along the bridge length, meeting the needs of subsequent bearing capacity reduction calculations.

[0052] S15: Perform load-bearing capacity reduction and fusion calculation on the installation state coefficients of each component and the component-section bearing capacity contribution matrix read in real time from the construction progress data. Convert the installation state coefficients of each component into the bearing capacity retention ratio. Combined with the benchmark bearing capacity distribution in the complete state, calculate the local bearing capacity value of each section at the current moment and generate the local ultimate bearing capacity distribution.

[0053] Specifically, the system reads the installation status coefficients of each component in real time from the construction progress data. These coefficients reflect the current installation completion status of each component. The system performs a load-bearing capacity reduction and fusion calculation on the installation status coefficients of each component and the component-section bearing capacity contribution matrix. The system converts the installation status coefficients of each component into a bearing capacity retention ratio. This retention ratio is directly related to the installation status coefficients and reflects the impact of the component's installation status on its bearing capacity. The system calls upon the baseline bearing capacity distribution under complete conditions and, combined with the bearing capacity retention ratio and the component-section bearing capacity contribution matrix, calculates the local bearing capacity value for each section at the current moment. Based on the spatial coordinates of the trestle bridge, the system integrates the local bearing capacity values ​​of all sections to generate a local ultimate bearing capacity distribution. This distribution reflects the actual bearing capacity of each location on the steel trestle bridge under the current construction progress.

[0054] In one embodiment, S2 of the intelligent identification method for construction risk behavior of steel trestle bridge based on feature mining provided by the present invention specifically includes the following steps:

[0055] S21: Perform frame synchronization and distortion correction on the left and right eye video streams acquired by the multi-view vision system deployed at key nodes of the steel trestle bridge. Use the epipolar constraint algorithm to perform stereo matching on the corrected image pairs to generate 3D point cloud data of the construction scene with depth information.

[0056] Specifically, the system invokes a multi-view vision system deployed at key nodes of the steel trestle bridge to acquire the left and right view video streams collected by the system. The system performs frame synchronization processing on the left and right view video streams to ensure accurate correspondence between the left and right view video frames, eliminating the impact of frame deviations on subsequent processing. The system then performs distortion correction processing on the synchronized video frames to correct image distortions caused by the camera's inherent characteristics in the multi-view vision system, ensuring the images accurately reflect the actual conditions of the construction scene. The system uses an epipolar constraint algorithm to perform stereo matching on the corrected left and right view image pairs, establishing the correlation between corresponding pixels in the left and right view images through algorithmic calculations, and calculating the depth information of each pixel based on these correlations. The system integrates the pixels with depth information to construct 3D point cloud data of the construction scene, which can fully present the spatial location and morphological characteristics of various objects in the construction scene.

[0057] S22: Perform target detection and segmentation processing on the 3D point cloud data. Use a deep learning-based instance segmentation algorithm to identify heavy machinery targets in the point cloud, and extract the point cloud clusters of each machinery target and their bounding boxes in the image coordinate system to generate a preliminary list of machinery targets.

[0058] Specifically, the system performs target detection and segmentation on the generated 3D point cloud data. The system employs a deep learning-based instance segmentation algorithm to analyze the 3D point cloud data point by point, identifying point cloud data belonging to heavy machinery targets. The system separates the point cloud data of different heavy machinery targets using the instance segmentation algorithm, forming point cloud clusters corresponding to individual machinery targets, with each cluster representing an independent heavy machinery target. Simultaneously, the system extracts the bounding box of each machinery target's point cloud cluster in the image coordinate system. The bounding box can define the complete range of the machinery target in the image, clearly indicating its position. The system then organizes the point cloud clusters, bounding boxes in the image coordinate system, and related recognition information for each machinery target, integrating them into a preliminary list of detected machinery targets in a unified format.

[0059] S23: Perform model identification and parameter matching processing on the preliminary inspection list of mechanical targets, compare the point cloud shape features of the mechanical targets with the preset mechanical model database, determine the specific model of the mechanical target, and retrieve the corresponding curb weight and rated load parameters from the mechanical parameter database.

[0060] Specifically, the system performs model identification and parameter matching on the preliminary list of mechanical targets. The system extracts the point cloud shape features of each mechanical target in the preliminary list. These point cloud shape features reflect the structural morphology of the mechanical target and are the core basis for distinguishing different models of machinery. The system calls upon a pre-set machinery model database, which stores the point cloud shape features and corresponding model information of various types of heavy machinery. The system compares the extracted point cloud shape features of the mechanical targets with the feature information in the machinery model database one by one to determine the specific model of each mechanical target. Based on the determined specific model of the machinery, the system calls upon a pre-set machinery parameter database to retrieve the corresponding curb weight and rated load parameters for that model of machinery. The system then associates the parameter information with the corresponding mechanical target to complete the machinery model identification and parameter matching.

[0061] S24: Based on the spatial coordinate calculation of each mechanical target of a specific model, calculate the coordinates of the projection points of the mechanical targets on the bridge surface according to the coordinates of the center point of the point cloud cluster and the plane equation of the trestle bridge surface, generate the projection position of each mechanical target and the corresponding equivalent value of dynamic load, and collect them into a dynamic load-position matrix.

[0062] Specifically, the system performs spatial coordinate calculations based on the identified specific models of each mechanical target. The system extracts the center point coordinates of each mechanical target's point cloud cluster, reflecting the target's spatial center position. The system calls upon the planar equations of the trestle bridge deck, pre-generated from the steel trestle bridge's BIM model, which accurately reflects the bridge deck's spatial position. Based on the center point coordinates of the point cloud clusters and the trestle bridge deck's planar equations, the system calculates the projection coordinates of the mechanical targets on the bridge deck using spatial coordinate calculations. These projection coordinates accurately reflect the actual positions of the mechanical targets on the trestle bridge deck. The system combines the mechanical targets' curb weight and rated load parameters to calculate the equivalent dynamic load value for each mechanical target. The system then compiles the projected positions and corresponding equivalent dynamic load values ​​for each mechanical target, organizing them according to the trestle bridge coordinate system and the mechanical target's number to form a dynamic load-position matrix.

[0063] In one embodiment, S3 of the intelligent identification method for construction risk behavior of steel trestle bridge based on feature mining provided by the present invention specifically includes the following steps:

[0064] S31: Identify the weakest section in the local ultimate bearing capacity distribution, calculate and compare the local bearing capacity values ​​at each location along the bridge length, locate the section position corresponding to the minimum bearing capacity as the current weakest section, and record the bearing capacity benchmark value of the current weakest section.

[0065] Specifically, the system identifies the weakest section in the local ultimate bearing capacity distribution generated by S15. The system retrieves the local ultimate bearing capacity distribution data, traverses all sections along the length of the steel trestle bridge, and calculates the local bearing capacity value at each location. The system compares all calculated local bearing capacity values ​​one by one, selecting the section with the lowest bearing capacity value and designating it as the current weakest section. The system records the bearing capacity benchmark value of the current weakest section, which is the upper limit of the actual bearing capacity of the weakest section under the current construction conditions. When calculating the local bearing capacity value, the system uses a preset local bearing capacity calculation formula, the specific formula of which is:

[0066]

[0067] in, Let be the local bearing capacity value at position x along the bridge length at time t. This represents the baseline bearing capacity at the bridge length x in its intact state. Let be the reduction factor for the installation state of the components at position x along the bridge length at time t. This is the section degradation coefficient. The system uses this formula to calculate the local bearing capacity value section by section, ensuring the accuracy of the calculation results and providing reliable data support for locating the weakest section. Simultaneously, it records the value corresponding to the weakest section. As a benchmark value for bearing capacity.

[0068] S32: Calculate the moment contribution coefficient for each load point in the dynamic load-position matrix. Based on the influence line theory of structural mechanics, establish a moment influence line function for the weakest section. Substitute the coordinates of each load point along the bridge length into the influence line function to generate the moment contribution coefficient of each mechanical target for that section.

[0069] Specifically, the system calculates the bending moment contribution coefficient for each load application point in the dynamic load-location matrix. Based on the influence line theory of structural mechanics, the system establishes a bending moment influence line function for the currently weakest section located at S31. The bending moment influence line function describes the degree of bending moment influence on the weakest section as the load application point varies along the bridge length. The system extracts the coordinates along the bridge length of each load application point in the dynamic load-location matrix, substitutes these coordinates into the established bending moment influence line function, and calculates the bending moment contribution coefficient of each load application point to the currently weakest section. The specific form of the bending moment influence line function is as follows:

[0070]

[0071] in, Let the load point at position x along the bridge length at time t be the weakest section. The bending moment contribution coefficient, The weakest section The moment influence line function is determined by the structural characteristics of the steel trestle bridge and the location of the weakest section. It can reflect the correspondence between the location of the load application point and the moment contribution coefficient.

[0072] S33: Perform a weighted summation calculation on the bending moment contribution coefficient and the corresponding equivalent value of dynamic load. Multiply the equivalent value of dynamic load of each mechanical target by its bending moment contribution coefficient and sum them up to generate the total equivalent bending moment effect value generated by all machines at the weakest section at the current moment.

[0073] Specifically, the system performs a weighted summation calculation on the calculated bending moment contribution coefficient and the corresponding equivalent dynamic load value. The system extracts the equivalent dynamic load value of each mechanical target from the dynamic load-position matrix, extracts the bending moment contribution coefficient corresponding to each mechanical target from the calculation results, and introduces a load eccentricity correction coefficient to correct the additional torque effect generated when the load application point deviates from the bridge deck centerline. The system multiplies the equivalent dynamic load value, bending moment contribution coefficient, and load eccentricity correction coefficient of each mechanical target to obtain the equivalent bending moment effect value of a single mechanical target at the weakest section. The system accumulates the equivalent bending moment effect values ​​of all mechanical targets to generate the total equivalent bending moment effect value generated by all machines at the weakest section at the current moment. Preferably, the formula for calculating the total equivalent bending moment effect value is:

[0074]

[0075] in, The total equivalent bending moment effect value at time t. Let be the equivalent dynamic load value of the k-th mechanical target at time t. Point of application of the load The moment contribution coefficient of the current weakest section. This is the load eccentricity correction factor. This represents the total number of heavy machinery targets on the trestle at the current moment.

[0076] S34: The ratio of the total equivalent bending moment effect value to the bearing capacity benchmark value of the weakest section is calculated to generate a dimensionless first coupling risk value.

[0077] Specifically, the system calculates the ratio between the total equivalent bending moment effect value and the bearing capacity benchmark value of the weakest section. This ratio generates a dimensionless first coupling risk value, which directly reflects the real-time impact of current construction activities on the structural safety of the steel trestle bridge. The system extracts the generated total equivalent bending moment effect value and the recorded bearing capacity benchmark value of the weakest section, while introducing a section degradation coefficient to characterize the reduction in bearing capacity due to cumulative fatigue or localized damage. The system uses the total equivalent bending moment effect value as the numerator and the product of the bearing capacity benchmark value of the weakest section and the section degradation coefficient as the denominator to calculate the first coupling risk value. The formula for calculating the first coupling risk value is:

[0078]

[0079] in, Let be the equivalent dynamic load value of the k-th mechanical target at time t. Point of application of the load For the current weakest section The bending moment contribution coefficient, This is the load eccentricity correction factor, used to correct for the additional torque effect caused when the load application point deviates from the bridge deck centerline. It is determined by looking up a table using the ratio of eccentricity to bridge deck width. The weakest section Local bearing capacity at the location The section degradation coefficient characterizes the reduction in load-bearing capacity of the section due to cumulative fatigue or local damage, and is estimated in real time through visually detected crack length or corrosion area.

[0080] In one embodiment, S4 of the intelligent identification method for construction risk behavior of steel trestle bridge based on feature mining provided by the present invention specifically includes the following steps:

[0081] S41: Extract motion pattern features from historical trajectory data in the mechanical parameter database, smooth and filter the position sequence of each mechanical target for multiple consecutive frames, and calculate the rate of change of velocity and the rate of change of acceleration to generate a feature vector sequence containing motion trend features.

[0082] Specifically, the system accesses a mechanical parameter database to extract historical trajectory data for each mechanical target, and then performs motion pattern feature extraction processing on the extracted historical trajectory data. For each mechanical target, the system extracts a multi-frame position sequence, which contains the target's spatial position information at different times. The system then performs smoothing filtering on the multi-frame position sequence to eliminate abnormal data interference and ensure the stability of the position data. Preferably, the smoothing filtering can employ a moving average filtering method, with the following formula:

[0083]

[0084] in, Let be the position coordinates of the k-th mechanical target at time t after smoothing and filtering. The length of the sliding window. Let be the original position coordinates of the k-th mechanical target at time ti. Based on the smoothed position sequence, the system calculates the rate of change of velocity and the rate of change of acceleration at each time step. The rate of change of velocity reflects the trend of the mechanical target's velocity, and the rate of change of acceleration reflects the trend of the mechanical target's acceleration. The system integrates the position sequence, rate of change of velocity, and rate of change of acceleration of each mechanical target to construct a feature vector containing motion trend characteristics. Multiple consecutive sets of feature vectors are arranged in chronological order to generate a feature vector sequence.

[0085] S42: Perform time-series prediction model input construction processing on the feature vector sequence, organize multiple consecutive sets of feature vectors within a fixed time window into a three-dimensional tensor format, and use it as input data for the long short-term memory network. The dimensions of the three-dimensional tensor correspond to the number of samples, time step, and feature dimension.

[0086] Specifically, the system processes the generated feature vector sequence into a time-series prediction model. The system sets a fixed time window length, determined based on the changing patterns of the mechanical target's trajectory, ensuring complete capture of the target's motion patterns. From the feature vector sequence, the system extracts multiple consecutive sets of feature vectors according to the fixed time window length, organizing these extracted feature vectors into a three-dimensional tensor format that conforms to the input requirements of a Long Short-Term Memory (LSTM) network. The dimensions of the three-dimensional tensor correspond to the number of samples, the time step, and the feature dimension, respectively. The number of samples is the number of extracted feature vector sets, the time step is the fixed time window length, and the feature dimension is the number of features contained in each feature vector. Preferably, the formula for constructing the three-dimensional tensor is:

[0087]

[0088] in, To construct the complete 3D tensor, For the sample size, For time step, The system performs data validation on the constructed 3D tensor to ensure that the tensor dimension conforms to the input standard of the Long Short-Term Memory (LSTM) network, eliminating the problem of data dimension mismatch. The validated 3D tensor is then used as the input data for the LSM network, providing input support for subsequent forward propagation calculations.

[0089] S43: Perform forward propagation computation on the Long Short-Term Memory network, control the information flow through the input gate, forget gate and output gate, extract the temporal dependency features of the motion trajectory, and output the predicted position coordinates and predicted velocity values ​​of each mechanical target at a preset time after mapping through a fully connected layer.

[0090] Specifically, the system performs forward propagation computation on the Long Short-Term Memory (LSTM) network. The LTM network includes an input gate, a forget gate, and an output gate. These three gates control the flow and filtering of information, retaining valuable temporal information for predicting the mechanical target's trajectory while discarding irrelevant information. The input gate controls whether current-time feature information is input into the network, the forget gate controls whether historical temporal information is retained, and the output gate controls whether the processed information is output. The system inputs the three-dimensional tensor constructed by S42 into the LTM network and extracts temporal dependency features of the trajectory through the network's hidden layers. These temporal dependency features reflect the correlation between the mechanical target's motion states at different times. The extracted temporal dependency features are mapped through a fully connected layer, converting the feature dimension to the dimension required for prediction, and outputting the predicted position coordinates and predicted velocity values ​​of each mechanical target at a future preset time. The mapping formula for the LTM network output layer is:

[0091]

[0092] in, The predicted output for the future time τ includes the predicted position coordinates and the predicted velocity value. This is the weight matrix of the fully connected layer. These are the temporal dependency features output by the hidden layers of the network. This is the bias vector.

[0093] S44: Perform load trend calculation processing on the predicted location coordinates and predicted velocity values. Based on the current load value, predicted velocity value, and preset load-velocity correlation function, estimate the predicted load for future times. Combine the predicted values ​​for future times in the local ultimate bearing capacity distribution to calculate the second coupling risk value.

[0094] Specifically, the system performs load trend calculation processing on the predicted position coordinates and predicted velocity values ​​output by S43. The system extracts the load values ​​of each mechanical target at the current moment, combines them with the predicted velocity values, and calls a preset load-velocity correlation function to estimate the predicted loads of each mechanical target at future moments. The specific form of the load-velocity correlation function is as follows:

[0095]

[0096] in, Let τ be the predicted load on the k-th mechanical target at a future time. Let be the load value of the k-th mechanical target at time t. Let be the predicted velocity value of the k-th mechanical target at a future time τ. This is a load-velocity correlation function, used to characterize the correspondence between load and velocity. The system calls the predicted value corresponding to the future time in the local ultimate bearing capacity distribution. This predicted value is calculated based on the construction schedule and component installation trend. The system uses the same calculation logic as the first coupling risk value, substituting the predicted load and predicted location coordinates of the future time into the calculation to obtain the coupling risk value of the future time. The maximum coupling risk value within a preset future time period is selected as the second coupling risk value. Preferably, the calculation formula for the second coupling risk value is:

[0097]

[0098] in, This represents the second coupling risk value at a future time τ. Let τ be the predicted position of the kth mechanical target at a future time τ. The meanings of the remaining parameters are consistent with the formula for calculating the first coupling risk value. The system uses this formula to calculate the second coupling risk value.

[0099] In one embodiment, S5 of the intelligent identification method for construction risk behavior of steel trestle bridge based on feature mining provided by the present invention specifically includes the following steps:

[0100] S51: Perform time-series comparison and trend deviation calculation on the first coupling risk value and the second coupling risk value. Use the first coupling risk value at the current moment as the benchmark safety baseline, calculate the growth rate and magnitude of the second coupling risk value relative to the first coupling risk value, and generate a risk trend deviation index.

[0101] Specifically, the system performs time-series comparison and trend deviation calculation on the first coupled risk value generated by S34 and the second coupled risk value generated by S44. The system extracts the first coupled risk value at the current moment and sets it as the baseline safety measure. This baseline measures the safety status of the current construction activity and serves as a reference for subsequent risk trend assessment. The system calculates the growth rate of the second coupled risk value relative to the first coupled risk value. This growth rate reflects the degree of change in future risk relative to the current risk. Simultaneously, the system calculates the growth rate, which reflects how fast the risk is increasing. The formula for calculating the growth rate is:

[0102]

[0103] in, This represents the increase in the second coupling risk value relative to the first coupling risk value. This is the second coupling risk value. This represents the first coupling risk value. The growth rate is calculated using the following formula:

[0104]

[0105] in, For the rate of risk growth, The system integrates the growth rate and magnitude of growth to construct a risk trend deviation index, which comprehensively reflects the degree of deviation of future risks from the current safety baseline. The expression for the risk trend deviation index is as follows:

[0106]

[0107] in, This is an indicator of risk trend deviation. and This is a weighting coefficient used to balance the effects of growth magnitude and growth rate.

[0108] S52: Perform hazard warning state identification processing on the risk trend deviation index and the second coupled risk value. When the second coupled risk value is detected to exceed the preset orange warning threshold and the first coupled risk value is lower than the orange warning threshold, mark the current state as a critical hazard warning state, and extract the corresponding mechanical target identifier and prediction time window under the critical hazard warning state.

[0109] Specifically, the system performs hazard warning state identification processing on the risk trend deviation index and the second coupled risk value generated by S51. The system calls a preset orange warning threshold, which is determined based on the safety margin of the steel trestle structure, industry construction safety specifications, and actual engineering conditions, and is used to delineate the boundary between a safe state and a hazard warning state. The system compares the second coupled risk value with the preset orange warning threshold, and simultaneously compares the first coupled risk value with the preset orange warning threshold. When the second coupled risk value exceeds the preset orange warning threshold and the first coupled risk value is lower than the preset orange warning threshold, the system marks the current state as a critical hazard warning state. This state indicates that the current construction behavior is within a safe range, but future construction behavior will approach a dangerous state, posing a safety hazard. After marking the critical hazard warning state, the system extracts the corresponding mechanical target identifier for this state. The mechanical target identifier is used to distinguish different heavy machinery. At the same time, it extracts the prediction time window, which is the future time period corresponding to the second coupled risk value.

[0110] S53: Perform continuous sliding window verification on mechanical targets marked as critical danger warning states. Track the change trajectory of the second coupled risk value in multiple consecutive prediction periods. If the second coupled risk value exceeds the preset orange warning threshold and the first coupled risk value is lower than the orange warning threshold and the risk trend deviation index is continuously positive in N consecutive prediction periods, then the behavior of the mechanical target is confirmed as high-risk behavior, and a high-risk behavior confirmation signal is generated.

[0111] Specifically, the system performs continuous sliding window verification on mechanical targets marked as being in a critical danger warning state. The system sets multiple consecutive prediction periods, each consistent with the prediction period for the second coupled risk value. Within each prediction period, the system tracks the real-time trajectory of the second coupled risk value of the corresponding mechanical target, while simultaneously monitoring changes in the first coupled risk value and the trend of the risk trend deviation index. Within these multiple prediction periods, the system verifies whether preset judgment conditions are met. The preset judgment conditions are: the second coupled risk value exceeds a preset orange warning threshold, the first coupled risk value is below the preset orange warning threshold, and the risk trend deviation index remains positive. A continuously positive risk trend deviation index indicates that the risk is continuously increasing without any signs of decline. The judgment formula for continuous verification is:

[0112]

[0113] in, For the number of consecutive prediction periods, The second coupling risk value for the i-th prediction period. To preset the orange alert threshold, This is the risk trend deviation index for the i-th prediction period. If this condition is met for multiple consecutive prediction periods, the system confirms that the mechanical target's behavior is high-risk and generates a high-risk behavior confirmation signal, which is used to trigger subsequent risk information encapsulation and early warning processes.

[0114] S54: Based on the high-risk behavior confirmation signal, the mechanical target is subjected to multi-dimensional risk information structured encapsulation processing. The risk type, risk level, mechanical target identification, real-time projection position, predicted motion trajectory, prediction time window, second coupled risk value and risk trend deviation index are integrated into a standardized data format to generate high-risk behavior intent recognition results.

[0115] Specifically, based on the high-risk behavior confirmation signal generated by S53, the system performs multi-dimensional risk information structured encapsulation processing on the corresponding mechanical target. The system extracts multi-dimensional risk-related information, including risk type, risk level, mechanical target identifier, real-time projection position, predicted trajectory, prediction time window, second coupled risk value, and risk trend deviation index. The risk type is determined based on the mechanical target's behavioral characteristics; the risk level is classified according to the second coupled risk value and risk trend deviation index; the real-time projection position is extracted from the dynamic load-position matrix; and the predicted trajectory is extracted from the prediction results of S43. The system integrates all extracted multi-dimensional risk information into a standardized data format. This standardized data format ensures the uniformity and standardization of risk information, facilitating subsequent transmission, storage, and parsing. The expression for the standardized data format is:

[0116]

[0117] in, To standardize risk information, As a risk type, Risk level, For mechanical target identification, For real-time projection position, To predict the trajectory of motion, To predict the time window, This is the second coupling risk value. This serves as a risk trend deviation indicator. Through structured encapsulation, the system generates high-risk behavior intent identification results. These results comprehensively and clearly present relevant information about high-risk behaviors, providing accurate and reliable data for on-site construction safety management.

[0118] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0119] Based on the same inventive concept, this application also provides a feature-mining-based intelligent identification system for construction risk behaviors of steel trestle bridges, used to implement the aforementioned intelligent identification method for construction risk behaviors of steel trestle bridges based on feature mining. The solution provided by this system is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more embodiments of the intelligent identification system for construction risk behaviors of steel trestle bridges based on feature mining provided below can be found in the limitations of the intelligent identification method for construction risk behaviors of steel trestle bridges based on feature mining described above, and will not be repeated here.

[0120] Preferably, such as Figure 2 As shown, this invention provides a feature mining-based intelligent identification system 600 for construction risk behaviors of steel trestle bridges. This system is configured with the following modules:

[0121] The component bearing capacity calculation module 610 is used to dynamically characterize the component installation status information in the acquired steel trestle bridge construction progress data, call the component-section bearing capacity contribution matrix pre-generated by the finite element model to perform bearing capacity reduction calculation on the component installation status information, and generate local ultimate bearing capacity distribution;

[0122] The dynamic load positioning module 620 is used to perform heavy machinery target identification and three-dimensional spatial positioning processing on the construction images collected in real time by the multi-view vision system deployed at key nodes of the steel trestle. It identifies the type and model of the machinery target and determines the real-time projection position of the machinery target on the trestle bridge surface. It also matches the curb weight and rated load parameters of the machinery target from the preset machinery parameter database to generate the dynamic load-position matrix at the current moment.

[0123] The coupling risk quantification module 630 is used to input the local ultimate bearing capacity distribution and dynamic load-location matrix into the preset coupling risk digital twin model to perform structural-behavioral coupling risk quantification, calculate the moment contribution coefficient of each load point to the current weakest section, and calculate the first coupling risk value at the current moment in combination with the local ultimate bearing capacity distribution.

[0124] The risk trend prediction module 640 is used to predict the behavioral intention of the mechanical target's motion trajectory based on the mechanical parameter database, predict the predicted position and predicted load of each mechanical target at future time based on historical motion parameters, and calculate the second coupled risk value at future time by combining the local ultimate bearing capacity distribution.

[0125] The high-risk intent recognition module 650 is used to compare and process the first coupling risk value and the second coupling risk value. When the second coupling risk value reaches the preset danger threshold and the first coupling risk value does not reach the preset danger threshold, it is determined that the mechanical target has a high-risk behavioral intent and a high-risk behavioral intent recognition result is generated.

[0126] Preferably, the component bearing capacity calculation module 610 provided in this application is configured with the following units:

[0127] The component classification and positioning analysis unit is used to classify and spatially locate components in the component list and BIM model in the acquired steel trestle construction progress data. According to the structural stress characteristics, each component is divided into main beam components, transverse connection components and bridge deck system components. The three-dimensional spatial occupancy range of each component in the trestle coordinate system is extracted from the BIM model to generate a component classification list with spatial labels.

[0128] The cross-section component association mapping unit is used to perform cross-section discretization association processing on the component classification list with spatial labels. It divides multiple key cross-sections along the length of the trestle at preset intervals. Based on the intersection relationship between the spatial occupancy interval of each component and the key cross-section, it establishes a set of component indexes that each key cross-section depends on, and generates a cross-section-component association mapping table.

[0129] The bearing capacity contribution coefficient analysis unit is used to perform sensitivity analysis of the component contribution coefficient on the section-component association mapping table and the preset finite element model. In the finite element model, each component is masked one by one and nonlinear buckling simulation is performed. The decrease of bearing capacity at each key section relative to the intact state is recorded, and the initial contribution coefficient matrix of each component to each key section is generated.

[0130] The continuous contribution matrix extension unit is used to perform continuous interpolation extension processing on the initial contribution coefficient matrix. It uses a piecewise interpolation function to fit the contribution coefficients at discrete key sections into a function form that is continuously distributed along the bridge length, thereby generating the component-section bearing capacity contribution matrix.

[0131] The real-time bearing capacity reduction calculation unit is used to perform bearing capacity reduction and fusion calculation on the installation state coefficient of each component and the component-section bearing capacity contribution matrix read in real time from the construction progress data. It converts the installation state coefficient of each component into the bearing capacity retention ratio, and calculates the local bearing capacity value of each section at the current moment by combining the benchmark bearing capacity distribution in the complete state, thereby generating the local ultimate bearing capacity distribution.

[0132] Preferably, the dynamic load positioning module 620 provided in this application is configured with the following units:

[0133] The stereo vision point cloud generation unit is used to perform frame synchronization and distortion correction on the left and right eye video streams acquired by the multi-view vision system deployed at key nodes of the steel trestle bridge. The corrected image pairs are stereo matched using the epipolar constraint algorithm to generate three-dimensional point cloud data of the construction scene with depth information.

[0134] The mechanical target segmentation and detection unit is used to perform target detection and segmentation processing on 3D point cloud data. It uses a deep learning-based instance segmentation algorithm to identify heavy mechanical targets in the point cloud, and extracts the point cloud clusters of each mechanical target and its bounding box in the image coordinate system to generate a preliminary list of mechanical targets.

[0135] The mechanical model parameter matching unit is used to perform model identification and parameter matching processing on the preliminary inspection list of mechanical targets. It compares the point cloud shape features of the mechanical targets with the preset mechanical model database to determine the specific model of the mechanical target and retrieves the corresponding curb weight and rated load parameters from the mechanical parameter database.

[0136] The load position matrix calculation unit is used to perform spatial coordinate calculations based on the identified specific mechanical targets. According to the coordinates of the center point of the point cloud cluster and the plane equation of the trestle bridge surface, it calculates the coordinates of the projection points of the mechanical targets on the bridge surface, generates the projection position of each mechanical target and the corresponding dynamic load equivalent value, and collects them into a dynamic load-position matrix.

[0137] Preferably, the coupled risk quantification module 630 provided in this application is configured with the following units:

[0138] The weakest section location unit is used to identify the weakest section in the local ultimate bearing capacity distribution, calculate and compare the local bearing capacity values ​​at various locations along the bridge length, locate the section position corresponding to the minimum bearing capacity as the current weakest section, and record the bearing capacity benchmark value of the current weakest section.

[0139] The moment contribution coefficient calculation unit is used to calculate the moment contribution coefficient of each load point in the dynamic load-position matrix. Based on the influence line theory of structural mechanics, a moment influence line function is established for the weakest section. The coordinates of each load point along the bridge length are substituted into the influence line function to generate the moment contribution coefficient of each mechanical target for that section.

[0140] The total equivalent bending moment synthesis unit is used to perform weighted summation calculation on the bending moment contribution coefficient and the corresponding dynamic load equivalent value. The dynamic load equivalent value of each mechanical target is multiplied by its bending moment contribution coefficient and then accumulated to generate the total equivalent bending moment effect value generated by all machines at the weakest section at the current moment.

[0141] The coupled risk value solving unit is used to calculate the ratio between the total equivalent bending moment effect value and the bearing capacity benchmark value of the weakest section, and generate a dimensionless first coupled risk value.

[0142] Preferably, the risk trend prediction module 640 provided in this application is configured with the following units:

[0143] The motion trend feature extraction unit is used to extract motion pattern features from historical trajectory data in the mechanical parameter database, smooth and filter the position sequence of each mechanical target for multiple consecutive frames, calculate the rate of change of velocity and the rate of change of acceleration, and generate a feature vector sequence containing motion trend features.

[0144] The temporal prediction input building unit is used to process the feature vector sequence into the input of the temporal prediction model. It organizes multiple consecutive sets of feature vectors within a fixed time window into a three-dimensional tensor format, which serves as the input data for the long short-term memory network. The dimensions of the three-dimensional tensor correspond to the number of samples, the time step, and the feature dimension.

[0145] The LSTM trajectory prediction unit is used to perform forward propagation computation on the Long Short-Term Memory network. It controls the flow of information through input gates, forget gates, and output gates, extracts the temporal dependency features of the motion trajectory, and outputs the predicted position coordinates and predicted velocity values ​​of each mechanical target at a preset time after mapping through a fully connected layer.

[0146] The future coupling risk calculation unit is used to perform load trend estimation processing on the predicted location coordinates and predicted velocity values. Based on the current load value, predicted velocity value and preset load-velocity correlation function, it estimates the predicted load at future times and calculates the second coupling risk value by combining the predicted value at the corresponding future time in the local ultimate bearing capacity distribution.

[0147] Preferably, the high-risk intent recognition module 650 provided in this application is configured with the following units:

[0148] The risk trend deviation calculation unit is used to perform time-series comparison and trend deviation calculation on the first coupled risk value and the second coupled risk value. The first coupled risk value at the current moment is used as the benchmark safety baseline. The growth rate and magnitude of the second coupled risk value relative to the first coupled risk value are calculated to generate a risk trend deviation index.

[0149] The critical hazard warning identification unit is used to identify the hazard warning state of the risk trend deviation index and the second coupled risk value. When the second coupled risk value is detected to exceed the preset orange warning threshold and the first coupled risk value is lower than the orange warning threshold, the current state is marked as a critical hazard warning state, and the mechanical target identifier and prediction time window corresponding to the critical hazard warning state are extracted.

[0150] The continuous sliding window verification unit is used to continuously verify the mechanical target marked as a critical danger warning state. It tracks the change trajectory of the second coupled risk value in multiple consecutive prediction periods. If the second coupled risk value exceeds the preset orange warning threshold and the first coupled risk value is lower than the orange warning threshold and the risk trend deviation index is continuously positive in N consecutive prediction periods, the behavior of the mechanical target is confirmed as high-risk behavior, and a high-risk behavior confirmation signal is generated.

[0151] The high-risk intent encapsulation unit is used to perform multi-dimensional risk information structured encapsulation processing on mechanical targets based on high-risk behavior confirmation signals. It integrates risk type, risk level, mechanical target identification, real-time projection position, predicted motion trajectory, prediction time window, second coupled risk value, and risk trend deviation index into a standardized data format to generate high-risk behavior intent recognition results.

[0152] In one embodiment, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the above-described intelligent identification method for construction risk behavior of steel trestle bridge based on feature mining.

[0153] In one embodiment, this application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described intelligent identification method for construction risk behavior of steel trestle bridges based on feature mining.

[0154] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0155] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0156] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various variations or substitutions within the technical scope disclosed in this application, and these should all be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A steel trestle construction risk behavior intelligent identification method based on feature mining, characterized in that, Includes the following steps: S1: Dynamically characterize the component installation status information in the acquired steel trestle bridge construction progress data, call the component-section bearing capacity contribution matrix pre-generated by the finite element model to perform bearing capacity reduction calculation on the component installation status information, and generate local ultimate bearing capacity distribution; S2: Perform heavy machinery target identification and three-dimensional spatial positioning processing on the construction images collected in real time by the multi-view vision system deployed at key nodes of the steel trestle bridge. Identify the type and model of the mechanical target and determine the real-time projection position of the mechanical target on the trestle bridge surface. Match the curb weight and rated load parameters of the mechanical target from the preset mechanical parameter database to generate the dynamic load-position matrix at the current moment. S3: Input the local ultimate bearing capacity distribution and the dynamic load-position matrix into the preset coupling risk digital twin model to quantify the structural-behavioral coupling risk, calculate the moment contribution coefficient of each load point to the current weakest section, and calculate the first coupling risk value at the current moment in combination with the local ultimate bearing capacity distribution; S4: Based on the mechanical parameter database, predict the motion trajectory of the mechanical target to determine its behavioral intent; predict the predicted position and load of each mechanical target at future time based on historical motion parameters; and calculate the second coupling risk value at future time by combining the local ultimate bearing capacity distribution. S5: Compare the first coupling risk value with the second coupling risk value. When the second coupling risk value reaches a preset danger threshold and the first coupling risk value does not reach the preset danger threshold, determine that the mechanical target has a high-risk behavioral intention and generate a high-risk behavioral intention identification result.

2. The method of claim 1, wherein, S1 includes: S11: Perform component classification and spatial positioning analysis on the component list and BIM model in the acquired steel trestle construction progress data. According to the structural stress characteristics, each component is divided into main beam components, transverse connection components and bridge deck system components. The three-dimensional spatial occupancy range of each component in the trestle coordinate system is extracted from the BIM model to generate a component classification list with spatial labels. S12: Perform cross-sectional discretization association processing on the component classification list with spatial labels, divide multiple key cross-sections along the length of the trestle at preset intervals, and establish a set of component indexes for each key cross-section based on the intersection relationship between the spatial occupancy interval of each component and the key cross-sections, and generate a cross-section-component association mapping table. S13: Perform sensitivity analysis on the component contribution coefficient of the section-component association mapping table and the preset finite element model. In the finite element model, each component is shielded one by one and nonlinear buckling simulation is performed. Record the decrease in bearing capacity at each key section relative to the intact state and generate the initial contribution coefficient matrix of each component to each key section. S14: Perform continuous interpolation expansion processing on the initial contribution coefficient matrix, and use a piecewise interpolation function to fit the contribution coefficients at discrete key sections into a function form that is continuously distributed along the bridge length, thereby generating the component-section bearing capacity contribution matrix. S15: Perform load-bearing capacity reduction and fusion calculation on the installation state coefficient of each component at the current moment, which is read in real time from the construction progress data, and the component-section load-bearing capacity contribution matrix. Convert the installation state coefficient of each component into the load-bearing capacity retention ratio. Combined with the benchmark load-bearing capacity distribution in the complete state, calculate the local load-bearing capacity value at the current moment for each section and generate the local ultimate load-bearing capacity distribution.

3. The method of claim 1, wherein, S2 includes: S21: Perform frame synchronization and distortion correction on the left and right eye video streams collected by the multi-view vision system deployed at key nodes of the steel trestle bridge, and perform stereo matching on the corrected image pairs through the epipolar constraint algorithm to generate 3D point cloud data of the construction scene with depth information. S22: Perform target detection and segmentation processing on the three-dimensional point cloud data, use a deep learning-based instance segmentation algorithm to identify heavy machinery targets in the point cloud, and extract the point cloud clusters of each machinery target and their bounding boxes in the image coordinate system to generate a preliminary list of machinery target detections. S23: Perform model identification and parameter matching processing on the preliminary detection list of mechanical targets, compare the point cloud shape features of the mechanical targets with the preset mechanical model database, determine the specific model of the mechanical target, and retrieve the corresponding curb weight and rated load parameters from the mechanical parameter database; S24: Based on the identified mechanical targets of the specific model, perform spatial coordinate calculation processing. According to the coordinates of the center point of the point cloud cluster and the plane equation of the trestle bridge surface, calculate the coordinates of the projection point of the mechanical target on the bridge surface, generate the projection position of each mechanical target and the corresponding dynamic load equivalent value, and collect them into a dynamic load-position matrix.

4. The method of claim 1, wherein, S3 includes: S31: Perform weakest section identification processing on the local ultimate bearing capacity distribution, calculate and compare the local bearing capacity values ​​at each position along the bridge length, locate the section position corresponding to the minimum bearing capacity as the current weakest section, and record the bearing capacity benchmark value of the current weakest section; S32: Calculate the moment contribution coefficient for each load point in the dynamic load-position matrix. Based on the influence line theory of structural mechanics, establish a moment influence line function for the current weakest section. Substitute the coordinates of each load point along the bridge length into the influence line function to generate the moment contribution coefficient of each mechanical target for that section. S33: Perform a weighted summation calculation on the bending moment contribution coefficient and the corresponding equivalent value of dynamic load, multiply the equivalent value of dynamic load of each mechanical target by its bending moment contribution coefficient and then sum them up to generate the total equivalent bending moment effect value generated by all machines at the weakest section at the current moment. S34: Calculate the ratio between the total equivalent bending moment effect value and the bearing capacity benchmark value of the weakest section to generate a dimensionless first coupling risk value.

5. The method according to claim 4, characterized in that, The formula for calculating the first coupling risk value is: in, Let be the equivalent dynamic load value of the k-th mechanical target at time t. Point of application of the load For the current weakest section The bending moment contribution coefficient, This is the load eccentricity correction factor, used to correct for the additional torque effect caused when the load application point deviates from the bridge deck centerline. It is determined by looking up a table using the ratio of eccentricity to bridge deck width. The weakest section Local bearing capacity at the location The section degradation coefficient characterizes the reduction in load-bearing capacity of the section due to cumulative fatigue or local damage, and is estimated in real time through visually detected crack length or corrosion area.

6. The method according to claim 1, characterized in that, S4 includes: S41: Extract motion pattern features from the historical trajectory data in the mechanical parameter database, smooth and filter the position sequence of each mechanical target for multiple consecutive frames, and calculate the rate of change of velocity and the rate of change of acceleration to generate a feature vector sequence containing motion trend features. S42: Perform time-series prediction model input construction processing on the feature vector sequence, organize multiple consecutive sets of feature vectors within a fixed time window length into a three-dimensional tensor format, and use it as input data for the long short-term memory network. The dimensions of the three-dimensional tensor correspond to the number of samples, time step, and feature dimension. S43: Perform forward propagation computation on the long short-term memory network, control the information flow through the input gate, forget gate and output gate, extract the temporal dependency features of the motion trajectory, and output the predicted position coordinates and predicted velocity values ​​of each mechanical target at a future preset time after mapping through a fully connected layer. S44: Perform load trend estimation processing on the predicted position coordinates and predicted velocity values. Based on the current load value, predicted velocity value, and preset load-velocity correlation function, estimate the predicted load for future times. Combine this with the predicted values ​​for future times in the local ultimate bearing capacity distribution to calculate the second coupling risk value.

7. The method according to any one of claims 1-6, characterized in that, S5 includes: S51: Perform time-series comparison and trend deviation calculation on the first coupling risk value and the second coupling risk value, take the first coupling risk value at the current moment as the benchmark safety baseline, calculate the growth rate and growth magnitude of the second coupling risk value relative to the first coupling risk value, and generate a risk trend deviation index. S52: Perform hazard warning state identification processing on the risk trend deviation index and the second coupled risk value. When the second coupled risk value is detected to exceed the preset orange warning threshold and the first coupled risk value is lower than the orange warning threshold, mark the current state as a critical hazard warning state, and extract the mechanical target identifier and prediction time window corresponding to the critical hazard warning state. S53: Perform continuous sliding window verification on the mechanical target marked as a critical danger warning state, track the change trajectory of the second coupled risk value in multiple consecutive prediction cycles, and if the second coupled risk value exceeds the preset orange warning threshold and the first coupled risk value is lower than the orange warning threshold and the risk trend deviation index is continuously positive in N consecutive prediction cycles, then confirm that the behavior of the mechanical target is a high-risk behavior and generate a high-risk behavior confirmation signal. S54: Based on the high-risk behavior confirmation signal, the mechanical target is subjected to multi-dimensional risk information structured encapsulation processing, and the risk type, risk level, mechanical target identification, real-time projection position, predicted motion trajectory, prediction time window, second coupled risk value and risk trend deviation index are integrated into a standardized data format to generate high-risk behavior intent recognition results.

8. A smart identification system for construction risk behaviors of steel trestle bridges based on feature mining, characterized in that, The system includes: The component bearing capacity calculation module is used to dynamically characterize the component installation status information in the acquired steel trestle bridge construction progress data, call the component-section bearing capacity contribution matrix pre-generated by the finite element model to perform bearing capacity reduction calculation on the component installation status information, and generate local ultimate bearing capacity distribution; The dynamic load positioning module is used to perform heavy machinery target identification and three-dimensional spatial positioning processing on the construction images collected in real time by the multi-view vision system deployed at key nodes of the steel trestle. It identifies the type and model of the mechanical target and determines the real-time projection position of the mechanical target on the trestle surface. It also matches the curb weight and rated load parameters of the mechanical target from a preset mechanical parameter database to generate the dynamic load-position matrix at the current moment. The coupling risk quantification module is used to input the local ultimate bearing capacity distribution and the dynamic load-position matrix into a preset coupling risk digital twin model to perform structural-behavioral coupling risk quantification, calculate the moment contribution coefficient of each load point to the current weakest section, and calculate the first coupling risk value at the current moment in combination with the local ultimate bearing capacity distribution. The risk trend prediction module is used to predict the behavioral intent of the mechanical target's motion trajectory based on the mechanical parameter database, predict the predicted position and predicted load of each mechanical target at future time based on historical motion parameters, and calculate the second coupled risk value at future time in combination with the local ultimate bearing capacity distribution. The high-risk intent recognition module is used to compare the first coupling risk value with the second coupling risk value. When the second coupling risk value reaches a preset danger threshold and the first coupling risk value does not reach the preset danger threshold, it is determined that the mechanical target has a high-risk behavioral intent, and a high-risk behavioral intent recognition result is generated.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

Images