A dam metal structure gate detection method and system

Abstract

The present application relates to dam gate detection technical field, specifically to a kind of dam metal structure gate detection method and system, method includes: step S1: based on dam metal structure gate collection image data, each gate leaf beam lattice surface partition high-definition image is obtained by segmentation, and the information of corresponding gate leaf beam lattice is extracted;Step S2: determine multiple initial selection flexure risk gate leaf beam lattice information;Step S3: determine multiple push test gate leaf beam lattice and the detection push test of corresponding push test gate leaf beam lattice by initial selection flexure risk gate leaf beam lattice information;Step S4: using detection push test, the corresponding push test gate leaf beam lattice of dam metal structure gate is pushed one by one to push test detection, and the deformation point cloud image of dam metal structure gate under each detection push test is obtained;Step S5: generate the simulation deformation point cloud video of dam metal structure gate from low water level initial condition to design full water level limit condition process;Step S6: determine multiple high flexure risk gate leaf beam lattice.

Description

Technical Field

[0001] This invention relates to the field of dam gate inspection technology, specifically to a method and system for inspecting dam metal structure gates. Background Technology

[0002] Metal gate structures in dams are crucial for water retention and level regulation in hydraulic engineering projects, and their structural quality directly impacts the safety of the project's operation. After gate manufacturing or installation, structural inspection is typically required to assess the structural condition of critical load-bearing components such as the gate leaf beams and grids, and to determine if any flexural deformation or potential structural risks exist.

[0003] In existing technologies, gate inspection typically obtains data through manual inspections, local structural measurements, or jacking tests at a limited number of locations. However, these methods have a limited number of inspection points, making it difficult to comprehensively reflect the stress and deformation characteristics of the overall gate structure. Furthermore, the gate leaf beams are numerous and interconnected, making it difficult for the inspection results of a single beam to accurately reflect the overall structural condition, potentially leading to inaccurate identification of high-risk beams. In addition, existing inspection methods are mostly based on discrete inspection results, lacking the ability to simulate and analyze the overall deformation process of the gate under different water level conditions, thus affecting the completeness and accuracy of the inspection results. Summary of the Invention

[0004] To address the technical problem that existing gate inspection methods cannot efficiently and accurately assess the deformation characteristics of dam metal structure gates under all operating conditions, thus affecting the accuracy of identifying high-deflection-risk gate leaf beams, the present invention aims to provide a method for inspecting dam metal structure gates. The specific technical solution adopted is as follows: Step S1: Based on the image data collected from the metal structure gate of the dam, the surface partition high-definition image of each gate leaf beam grid is obtained from the image data by using the beam grid information determination model, and the corresponding gate leaf beam grid information is extracted. Step S2: Combine the high-resolution surface partition images of each door leaf beam grid with the corresponding door leaf beam grid information to determine multiple initial selections of door leaf beam grid information with deflection risk; Step S3: Determine multiple jacking test door leaf beams and the detection jacking force of the corresponding jacking test door leaf beams based on the information of the initial selected deflection risk door leaf beam grids; Step S4: Use the jacking force detection method to detect the jacking force of the corresponding jacking test gate leaf beam grid of the dam metal structure gate one by one, and obtain the deformation point cloud image of the dam metal structure gate under each detected jacking force. Step S5: Generate a simulated deformation point cloud video of the dam's metal structure gate from the initial low water level condition to the design full water level ultimate condition based on the deformation point cloud image; Step S6: Identify multiple high-deflection-risk door leaf beam grids based on simulated deformation point cloud video.

[0005] Preferably, the beam grid information determination model is a convolutional neural network.

[0006] Preferably, step S1 includes: Step S11: Use an industrial camera to take multiple high-resolution images of the surface of the dam's metal structure gate to be inspected before leaving the factory, and integrate them to form image data; Step S12: Input the image data into the convolutional neural network, extract the features of the image data through convolution operation, segment the high-definition image of the surface of each dam metal structure gate, output the surface partition high-definition image of each gate leaf beam grid, and extract the corresponding gate leaf beam grid information.

[0007] Preferably, the door leaf beam grid information includes the surface contour feature information of the corresponding door leaf beam grid, the weld appearance morphology information, and the coordinates of the local deformation texture distribution.

[0008] Preferably, in step S2, the specific steps are as follows: A convolutional neural network is used to filter high-resolution images of surface partitions and corresponding information of each door leaf beam grid, predicting door leaf beam grid information that is prone to flexural deformation during stress, and identifying multiple initial door leaf beam grid information with flexural risk.

[0009] Preferably, the preliminary flexural risk gate leaf beam grid information includes the surface contour feature information of the corresponding gate leaf beam grid, weld appearance morphology information, local deformation texture distribution coordinates, flexural risk level, and deformation texture anomaly sequence.

[0010] Preferably, step S3 includes: Step S31: Cluster the information of the initially selected flexural risk door leaf beam grids to obtain multiple clusters and multiple representative door leaf beam grids, and collect the ultrasonic scanning test data of the weld of each representative door leaf beam grid, wherein each representative door leaf beam grid is the door leaf beam grid closest to the center of each cluster. Step S32: Determine multiple cooperating door leaf beam grids corresponding to each representative door leaf beam grid by using multiple clusters and ultrasonic scanning inspection data of the weld seams of each representative door leaf beam grid, and collect ultrasonic scanning inspection data of the weld seams of each cooperating door leaf beam grid. Step S33: Determine the corresponding detection thrust based on the representative door leaf beam grid and the cooperative door leaf beam grid respectively.

[0011] Preferably, step S33 specifically includes: Multiple beam grid maps are constructed based on each representative gate leaf beam grid and its corresponding multiple cooperative gate leaf beam grids. A graph neural network model is used to process the beam grid maps to determine the detected thrust force of each representative gate leaf beam grid and the detected thrust force of each cooperative gate leaf beam grid. Each beam grid map includes multiple nodes and multiple edges. Each node includes a representative gate leaf beam grid node and its corresponding multiple cooperative gate leaf beam grids. The representative gate leaf beam grid node and the cooperative gate leaf beam grids are connected to form an edge.

[0012] Preferably, step S6 specifically includes: A preset safe yield deformation limit is established. The simulated deformation point cloud video is split into a continuous point cloud video frame sequence using a Transformer model. The Transformer model includes a self-attention mechanism and a feedforward neural network. The self-attention mechanism obtains the relative displacement change rate of the point cloud coordinates of different door leaf beam grid regions in each frame, tracking the deformation acceleration of the door leaf beam grid during the pressure increase process. The feedforward neural network captures the dynamic correlation features in the point cloud video frame sequence and compares them with the safe yield deformation limit. When the relative displacement change rate of the door leaf beam grid continuously approaches the safe yield deformation limit in multiple consecutive frames, and the deformation acceleration does not slow down, the corresponding door leaf beam grid is determined to be a high-deflection-risk door leaf beam grid.

[0013] To solve the above-mentioned technical problems, the present invention also provides: a dam metal structure gate inspection system, the system comprising: The information acquisition and determination module is used to: acquire image data based on the metal structure gate of the dam, use the beam grid information determination model to segment high-definition images of the surface partitions of each gate leaf beam grid from the image data, and extract the corresponding gate leaf beam grid information; The initial risk determination module is used to: combine the high-resolution surface partition images of each leaf beam grid with the corresponding leaf beam grid information to determine multiple initial flexural risk leaf beam grid information; The test scheme determination module is used to: determine multiple jacking test door leaf beams and the corresponding detection jacking force of the jacking test door leaf beams based on the information of the initially selected deflection risk door leaf beams; The jacking detection execution module is used to: detect the jacking force of the corresponding jacking test gate leaf beam grid of the dam metal structure gate one by one using the detection jacking force, and obtain the deformation point cloud image of the dam metal structure gate under each detection jacking force; The simulated video generation module is used to generate simulated deformation point cloud videos of the dam's metal structure gate from the initial low water level condition to the design full water level limit condition based on the deformation point cloud images. The beam risk determination module is used to determine multiple high-deflection-risk gate leaf beams based on simulated deformation point cloud video.

[0014] The present invention has the following beneficial effects: 1. Analysis of image data collected from dam metal structure gates, implemented through multiple steps, enables efficient and accurate assessment of the deformation characteristics of dam metal structure gates under all operating conditions and identification of high-deflection-risk gate leaf beams. Specifically, image data is segmented and high-resolution images of the surface partitions of each gate leaf beam are extracted, along with corresponding gate leaf beam information, to determine multiple initial deflection-risk gate leaf beam information. Then, multiple jacking test gate leaf beams and their corresponding detection jacking forces are determined, and jacking force detection is performed to obtain deformation point cloud images. These images are used to generate simulated deformation point cloud videos of the dam metal structure gates from the initial low-water-level condition to the design full-water-level ultimate condition, identifying multiple high-deflection-risk gate leaf beams. This comprehensive implementation method systematically, efficiently, and accurately assesses the overall deformation characteristics and behavioral patterns of dam metal structure gates under various water level conditions, successfully identifying high-deflection-risk gate leaf beams requiring special attention, providing a reliable technical basis for gate safety monitoring and maintenance decisions.

[0015] 2. The dam metal structure gate detection system provided by this invention has the same beneficial effects as the dam metal structure gate detection method provided by this invention, and will not be described in detail here. Attached Figure Description

[0016] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the structure of a dam metal structure gate according to an embodiment of the present invention; Figure 2 A flowchart illustrating the steps of a method for detecting a dam metal structure gate, as provided in one embodiment of the present invention; Figure 3 This is a schematic block diagram of a dam metal structure gate detection system provided in one embodiment of the present invention; In the picture: 1. Information collection and determination module; 2. Preliminary risk determination module; 3. Test plan determination module; 4. Jacking and testing execution module; 5. Simulation video generation module; 6. Beam grid risk determination module. Detailed Implementation

[0018] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a dam metal structure gate detection method and system proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0019] 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.

[0020] Please see Figure 1 The following describes in detail, with reference to the accompanying drawings, the specific scheme of the dam metal structure gate detection method and system provided by the present invention.

[0021] To better illustrate, in hydraulic engineering systems, a dam metal structure gate is a metal plate-type water-retaining structure used to effectively regulate water levels in reservoirs, rivers, or channels. It can precisely block or open water as needed to control the flow of water. The main load-bearing structure of the dam metal structure gate is composed of multiple gate leaf beams, which form the main load-bearing framework on the back of the gate panel. These beams are assembled from multiple sets of beam components. In actual engineering, the gate leaf beams are typically welded or riveted to the back of the gate panel. The dam metal structure gate relies on these beams to withstand enormous hydrostatic and hydrodynamic pressures, as well as external loads such as sediment pressure, ice pressure, and wave impact, ensuring that the gate maintains structural integrity and operational stability even under heavy loads, thus fulfilling its intended water control and impounding functions.

[0022] High flexural deformation can not only directly affect the sealing performance and smoothness of the gate's opening and closing operation, but also pose a serious challenge to the mechanical stability and long-term durability of the overall structure. Therefore, identifying the high flexural risk gate leaf beam grid of the dam's metal structure gate can ensure the safety and reliability of the dam's metal structure gate in long-term operation and effectively prevent potential structural risks and functional failures caused by high flexural deformation.

[0023] Therefore, a method for detecting metal gates of dams is proposed, and a corresponding electronic device is proposed, including: a processor, a communication interface, a memory, and a communication bus. The processor, communication interface, and memory communicate with each other through the communication bus. The processor calls logical instructions in the memory to execute a method for detecting metal gates of dams. Therefore, whether the device and program data are integrated or different hardware is configured to produce functions with similar effects to those achieved by this invention, they all fall within the protection scope of this invention.

[0024] Please see Figure 2 The diagram illustrates a flowchart of a method for detecting a dam metal structure gate according to an embodiment of the present invention, the method comprising: Step S1: Based on the image data collected from the metal structure gate of the dam, the surface partition high-definition image of each gate leaf beam grid is obtained from the image data by using the beam grid information determination model, and the corresponding gate leaf beam grid information is extracted. Step S2: Combine the high-resolution surface partition images of each door leaf beam grid with the corresponding door leaf beam grid information to determine multiple initial selections of door leaf beam grid information with deflection risk; Step S3: Determine multiple jacking test door leaf beams and the detection jacking force of the corresponding jacking test door leaf beams based on the information of the initial selected deflection risk door leaf beam grids; Step S4: Use the jacking force detection method to detect the jacking force of the corresponding jacking test gate leaf beam grid of the dam metal structure gate one by one, and obtain the deformation point cloud image of the dam metal structure gate under each detected jacking force. Step S5: Generate a simulated deformation point cloud video of the dam's metal structure gate from the initial low water level condition to the design full water level ultimate condition based on the deformation point cloud image; Step S6: Identify multiple high-deflection-risk door leaf beam grids based on simulated deformation point cloud video.

[0025] The explanation is as follows: Based on steps S1 to S6, high-deflection-risk gate leaf beam grids are identified from the dam's metal structure gates. High deflection risk refers to the fact that under the continuous or cyclic action of external loads such as long-term water pressure, self-weight, and siltation, some of the gate leaf beam grid structures may exhibit bending deformations that exceed the design allowable range or safety threshold, i.e., the possibility of deflection is significantly higher.

[0026] Furthermore, Liang Ge's information determined that the model was a convolutional neural network.

[0027] It can be explained that a convolutional neural network (CNN) is a neural network structure that extracts features from an image through multiple layers of convolutional operations. CNNs can extract local spatial features through convolutional layers and reduce data dimensionality through pooling layers. In turn, through the multi-layer structure, an understanding of the overall structure of the image can be gradually formed. Using a convolutional neural network can accurately identify the region division and structural boundaries in an image.

[0028] Further, step S1 includes: Step S11: Use an industrial camera to take multiple high-resolution images of the surface of the dam's metal structure gate before it leaves the factory for inspection, and integrate them to form image data.

[0029] Specifically, the industrial camera is a high-resolution industrial camera, which can accurately capture the fine textures and potential defects on the gate surface, ensuring the reliability and accuracy of the image information. The high-resolution industrial camera is used to capture images of the entire surface of the gate leaf of the dam metal structure before it leaves the factory for inspection. This results in multiple high-definition images of the dam metal structure gate surface, which include appearance details such as welds, plate flatness, and component splicing on the surface of the gate leaf. All high-definition images of the dam metal structure gate surface are integrated into image data.

[0030] Step S12: Input the image data into the convolutional neural network, extract the features of the image data through convolution operation, segment the high-definition image of the surface of each dam metal structure gate, output the surface partition high-definition image of each gate leaf beam grid, and extract the corresponding gate leaf beam grid information.

[0031] Furthermore, the door leaf beam grid information includes the surface contour feature information of the corresponding door leaf beam grid, the weld appearance morphology information, and the coordinates of the local deformation texture distribution.

[0032] It can be explained that the surface profile feature information includes the reference profile dimensions of the leaf beam grid edge, the mean straightness deviation, and the profile fit degree. The reference profile dimensions of the leaf beam grid edge refer to the standard profile boundary dimensions based on engineering specifications and functional requirements. The mean straightness deviation refers to the statistical average of the fitting deviations between the edge of each straight segment of the leaf beam grid and the standard straight line; the smaller the value, the straighter the edge. The profile fit degree refers to the degree of matching between the actual overall profile of the beam grid and the design standard profile, with a corresponding value range of [range missing]. The closer the contour fit is to 1, the better the contour fits the design requirements.

[0033] The weld appearance morphology information includes the reference weld width, weld direction continuity, surface straightness, reinforcement uniformity, and appearance defect types in the gate leaf beam grid weld area. The reference weld width refers to the measured and confirmed standard width between the weld root and the surface, which assesses whether the basic dimensions of the weld cross-section meet design requirements. Weld direction continuity focuses on whether the weld maintains consistency during its spatial extension, without interruptions, jumps, or irregular abrupt changes, reflecting the stability of the welding process and the consistency of operation. Surface straightness refers to the flatness of the weld's outer surface relative to the base material surface, reflecting heat input control and forming quality during welding, avoiding irregularities such as unevenness or wavy deformation. Reinforcement uniformity refers to whether the height distribution of the weld reinforcement layer, i.e., the reinforcement portion, is balanced and uniform along its length; excessively high, low, or undulating reinforcement can affect the stress distribution and fatigue performance of the structure. Appearance defect types include absence of welds, undercut, surface microcracks, weld beads, and excessive reinforcement, which directly affect the integrity and safety of the weld.

[0034] Local deformation texture distribution coordinates refer to the spatial location of minute, localized plastic deformations or texture changes on the surface of the door leaf beam grid caused by stress, manufacturing process, or wear. This is a set of spatial coordinate data to accurately locate and quantify the specific distribution location of local deformation or texture features on the surface of the door leaf beam grid.

[0035] Specifically, a high-resolution image of each gate leaf beam grid is obtained by segmenting the high-resolution image of the dam's metal structure gate surface using a convolutional neural network. This involves multi-level feature extraction from the input high-resolution image of the dam's metal structure gate surface. The convolutional layer scans and identifies key structural features such as weld boundaries and component splicing contours of the gate leaf beam grids in the image, accurately distinguishing the regional range of different gate leaf beam grids. Then, the pooling layer performs dimensionality reduction processing on the extracted structural features to retain core boundary and position information, reducing the redundancy of subsequent calculations. The fully connected layer integrates the features processed by the pooling layer to complete the image segmentation of different gate leaf beam grid regions, obtaining high-resolution images of the independent regions of each individual gate leaf beam grid. Finally, information on the corresponding gate leaf beam grids is extracted from the high-resolution image of the dam's metal structure gate surface to characterize the surface structure and appearance of each individual gate leaf beam grid.

[0036] Furthermore, in step S2, specifically: A convolutional neural network is used to filter high-resolution images of surface partitions and corresponding information of each door leaf beam grid, predicting door leaf beam grid information that is prone to flexural deformation during stress, and identifying multiple initial door leaf beam grid information with flexural risk.

[0037] To clarify, the convolutional neural network, i.e., the initial risk determination model, takes as input a high-resolution image of the surface partition of each door leaf beam grid and the corresponding door leaf beam grid information, and outputs multiple initial deflection risk door leaf beam grid information. The high-resolution image of the surface partition of each door leaf beam grid provides an intuitive visual representation of the grid surface and clearly reflects visual features such as surface appearance, providing intuitive image evidence for the model to identify anomalies and determine structural status. The door leaf beam grid information provides quantified surface contour and weld information, as well as structured data such as local deformation texture distribution coordinates, which complement the visual information of the image and avoids the limitations of judging solely through images. This ensures that the model can accurately combine visual features and structured parameters to complete the initial screening.

[0038] Furthermore, the preliminary information on the deflection risk gate leaf beam grid includes the surface contour feature information of the corresponding gate leaf beam grid, the weld appearance morphology information, the coordinates of the local deformation texture distribution, the deflection risk level, and the deformation texture anomaly sequence.

[0039] It can be explained that the surface contour feature information, weld appearance morphology information, and local deformation texture distribution coordinates of the door leaf beam grid correspond to the door leaf beam grid information; while the deflection risk level refers to the level index of the potential deflection hazard of the door leaf beam grid. The higher the level, the higher the risk of the door leaf beam grid having a deflection hazard; the deformation texture anomaly sequence refers to the numerical sequence of the degree of difference between the surface texture of the beam grid and the standard texture recorded in positional order.

[0040] Specifically, the convolutional neural network reads the pixel matrix from the high-resolution image of the surface partition of each door leaf beam grid through convolutional layers and extracts the deep visual representation of the local deformation texture fed back from the door leaf beam grid information. At the same time, the convolutional neural network receives the surface contour feature information and weld appearance morphology information from each door leaf beam grid information, and aligns and stitches the extracted deep visual features with the structural parameter features fed back from the high-resolution image of the surface partition of the door leaf beam grid. By analyzing the defect type in the weld appearance morphology information and the mean deviation in the surface contour feature information, and comparing with the texture abnormal areas in the image, the degree of structural abnormality of the current door leaf beam grid can be determined, providing a basis for subsequent stress analysis. When the convolutional neural network identifies any abnormality in the appearance of the weld seam of the door leaf beam grid and the profile fitting degree is low, it judges the degree of potential deflection risk based on the abnormal features and derives the level index of the potential deflection risk of the door leaf beam grid currently being analyzed. Finally, the convolutional neural network integrates the calculated deflection risk level and the numerical sequence of the difference between the surface texture of the door leaf beam grid and the standard texture recorded in positional order with the original structural information, that is, with the door leaf beam grid information obtained in the previous steps, and outputs multiple preliminary deflection risk door leaf beam grid information.

[0041] Furthermore, step S3 includes: Step S31: Cluster the information of the initially selected flexural risk door leaf beam grids to obtain multiple clusters and multiple representative door leaf beam grids, and collect the ultrasonic scanning test data of the weld of each representative door leaf beam grid, wherein each representative door leaf beam grid is the door leaf beam grid closest to the center of each cluster.

[0042] It is explained that the clustering algorithm used is the K-means clustering algorithm, which is an unsupervised clustering algorithm. The core of this algorithm is to divide the input dataset into K preset number of clusters so that the feature similarity of data points within each cluster is the highest and the feature similarity of data points between different clusters is the lowest. In this embodiment, the value of K can be preset in advance according to the actual screening requirements.

[0043] It can be explained that the feature grouping set obtained by classifying the features of multiple initially selected flexural risk gate leaf beam grids using the K-means clustering algorithm results in K clusters and multiple representative gate leaf beam grids. In practice, a pre-defined relationship table is constructed in advance between the value of K and the number of initially selected flexural risk gate leaf beam grids and the degree of dispersion of flexural risk levels. The value of K is determined according to this table. The more initially selected flexural risk gate leaf beam grids there are and the higher the degree of dispersion of flexural risk levels, the larger the value of K. Within each cluster, the gate leaf beam grids have highly similar flexural risk characteristics, and there are significant differences in the flexural risk characteristics of gate leaf beam grids between different clusters.

[0044] Multiple representative phylum leaf beam grids are selected from each cluster and are able to represent the overall flexural risk characteristics of the phylum leaf beam grids in the corresponding cluster; that is, each representative phylum leaf beam grid corresponds to an independent cluster, and each representative phylum leaf beam grid is the phylum leaf beam grid closest to the center of each cluster.

[0045] Specifically, a K-value clustering algorithm is used to cluster multiple initially selected deflection risk gate leaf beam grids to obtain K deflection risk feature clusters. Initially selected deflection risk gate leaf beam grids with similar features are grouped into the same cluster, achieving structured classification management of dispersed initially selected risk beam grids. In the corresponding implementation process, the quantified features in the input multiple initially selected deflection risk gate leaf beam grids are transformed into data points in a multi-dimensional feature space, and the initial cluster centers are set according to a preset K value. Then, by iteratively calculating the feature distance from each data point to each cluster center, each gate leaf beam grid is divided into... In the nearest cluster, the cluster center position of each cluster is recalculated, and this iterative process is repeated until the cluster center position no longer changes. This completes the feature clustering of all gate leaf beam lattices, and the algorithm finally outputs K stable clusters. That is, through the clustering operation, the core feature differences of potential flexural hazards of gate leaf beam lattices in different groups can be highlighted, while the local feature fluctuations of gate leaf beam lattices within a cluster can be weakened. In this way, the flexural risk distribution pattern of dam gate leaf beam lattices can be presented in units of clusters, which facilitates targeted feature analysis and risk assessment of gate leaf beam lattice sets with similar hazard attributes.

[0046] Next, representative leaf beam grids are selected from each flexural risk feature cluster. That is, the leaf beam grid closest to the cluster center is selected from each cluster as the representative leaf beam grid, which is used to centrally represent the common hidden danger characteristics and typical structural states of all leaf beam grids in the corresponding cluster. The representative leaf beam grid carries the overall characteristic attributes of this type of risk beam grid in the corresponding cluster as a single typical object, and can avoid the redundancy and complexity caused by analyzing all individuals in the cluster one by one.

[0047] Finally, the ultrasonic flaw detector was used to perform a full-coverage scan of all welds representing the door leaf beam grid to collect the detection data, which is the ultrasonic scanning detection data of the welds corresponding to the door leaf beam grid. This data includes the location, size, type, and weld penetration depth of the internal defects in the corresponding door leaf beam grid welds. Among them, the location of the internal defect refers to the relative position of the defect, which is used to locate the defect; the size of the defect refers to the spatial quantitative measurement of the determined internal defect; the defect type includes porosity, slag inclusion, lack of fusion, incomplete penetration, cracks, and poor shape, etc.; the weld penetration depth refers to the depth to which the base metal is melted during the fusion welding process, that is, the vertical distance from the weld surface to the bottom of the molten area.

[0048] Step S32: Determine the multiple cooperating door leaf beam grids corresponding to each representative door leaf beam grid by using the ultrasonic scanning inspection data of the welds of multiple clusters and each representative door leaf beam grid, and collect the ultrasonic scanning inspection data of the welds of each cooperating door leaf beam grid.

[0049] It can be explained that by using a collaborative determination model to input multiple clusters and ultrasonic scanning inspection data of weld seams from each representative door leaf beam grid, multiple collaborative door leaf beam grids corresponding to each representative door leaf beam grid are determined. The collaborative determination model is a deep neural network model, which includes a deep neural network (DNN). A deep neural network is an artificial neural network with a multi-layer nonlinear transformation structure, consisting of an input layer, multiple hidden layers, and an output layer. The hidden layers can perform multi-level feature learning and nonlinear mapping on the input data, and can uncover the implicit correlations and distribution patterns in the data. It can complete complex classification, matching, and prediction tasks.

[0050] To explain, based on the aforementioned steps, K clusters are obtained, which define the range of door leaf beam grid groups with similar risk characteristics; and the ultrasonic scanning inspection data of multiple representative door leaf beam grids can reflect the actual defect distribution and structural weak points inside the welds of the representative door leaf beam grids; the combination of K clusters and ultrasonic scanning inspection data of multiple representative door leaf beam grids enables the deep neural network to accurately locate door leaf beam grids with strong correlation in stress state within a local group with similar physical properties, thereby obtaining collaborative door leaf beam grids.

[0051] In this embodiment, the multiple cooperating door leaf beam grids corresponding to each representative door leaf beam grid are other door leaf beam grids within the same cluster as the corresponding representative door leaf beam grid, excluding the representative door leaf beam grid, and possessing highly similar flexural risk characteristics. In other words, cooperating door leaf beam grids are door leaf beam grids within the same cluster that have similar structural morphology, similar stress characteristics, and flexural risk levels to the representative door leaf beam grid. By determining the cooperating door leaf beam grids, the common structural characteristics and defect distribution patterns of this type of door leaf beam grid can be more comprehensively reflected, thereby achieving a comprehensive summary and description of the overall characteristics of door leaf beam grid components of the same type.

[0052] Specifically, after receiving the member distribution information of K clusters and the ultrasonic scanning inspection data of the welds of multiple representative door leaf beam grids, the deep neural network parses the ultrasonic scanning inspection data of the welds of the representative door leaf beam grids in the hidden layer to extract the coordinates of the internal defects and the hidden physical defect features of the representative door leaf beam grids. By cross-comparing these internal defect features with the surface deflection risk level of other door leaf beam grids in the K clusters, the deep neural network can simulate and calculate the stress concentration diffusion path caused by the internal defects of the representative door leaf beam grids when subjected to jacking pressure. Then, the model can determine whether other door leaf beam grids in the same cluster are within the range of this stress diffusion path, so as to screen out multiple door leaf beam grids that are most significantly affected by the deformation of the representative door leaf beam grids, and output these door leaf beam grids that are in the same cluster and have a close stress cooperative relationship with the representative door leaf beam grids as multiple cooperative door leaf beam grids corresponding to each representative door leaf beam grid.

[0053] Next, the ultrasonic flaw detector is used to scan all the welds of the collaborative door leaf beam grid to obtain the detection data, which includes the location, size, type and penetration depth of the internal defects of the corresponding door leaf beam grid welds.

[0054] To better illustrate this, step S32 is analyzed in detail to determine the multiple cooperative door-leaf beam grids corresponding to each representative door-leaf beam grid. The method includes: Step S321: Determine the estimated structural strength level of each cluster, the similarity of defect locations between each beam grid in each cluster and the representative door leaf beam grid, and the stress influence radiation range curve of the representative door leaf beam grid in each cluster by using ultrasonic scanning inspection data of welds of multiple clusters and each representative door leaf beam grid.

[0055] It can be explained that a deep neural network is used to sequentially determine the estimated structural strength level of each cluster, the similarity of defect locations between each beam grid within each cluster and the representative door leaf beam grid, and the stress influence radiation range curve of the representative door leaf beam grid within each cluster. The estimated structural strength level of each cluster is determined by the deep neural network and is a quantitative grading index characterizing the overall structural resistance to deformation and failure of all door leaf beam grids within a single cluster; a higher level indicates higher overall structural strength of the corresponding cluster. The similarity of defect locations between each beam grid within each cluster and the representative door leaf beam grid is determined by the deep neural network. This value represents the degree of overlap and matching between the internal weld defect locations of other beam grids within the cluster (excluding the representative door leaf beam grid) and the internal weld defect locations of the corresponding representative door leaf beam grid. This value range is [value range missing]. The closer the value is to 1, the higher the matching degree of the defect locations of the two; the curve of the force influence radiation range of the phylum leaf beam grid in each cluster is generated by a deep neural network. The curve is a curve with the force value on the horizontal axis and the distance of the deformation influence of the phylum leaf beam grid on other beam grids in the cluster on the vertical axis. This curve can accurately reflect the dynamic change law of the spatial range of beam grids in the cluster that can be covered by the deformation influence of the phylum leaf beam grid under different force values. The larger the force value, the farther the influence distance, and the wider the range of beam grids covered in the cluster.

[0056] Specifically, the deep neural network integrates and analyzes the feature distribution information of K clusters and the ultrasonic scanning inspection data of welds from multiple representative door leaf beam grids. By combining the structural parameters of all beam grids within a cluster, weld defect data, and core weld parameters of the representative door leaf beam grids, the overall structural strength grade of each cluster is comprehensively derived. Then, the weld defect location coordinates of each beam grid within a cluster and the representative door leaf beam grids are compared and calculated one by one to quantify the similarity value of the corresponding defect locations. Finally, based on the overall structural strength of the clusters and the defect distribution characteristics of the representative door leaf beam grids, the deformation transmission law of the representative door leaf beam grids under different stress values ​​is simulated, and the deformation influence spatial range corresponding to different stress values ​​is derived, generating the stress influence radiation range curve of the representative door leaf beam grids within each cluster.

[0057] Step S322: Based on the estimated structural strength level of each cluster, the similarity of the defect location of each beam grid in each cluster and the representative door leaf beam grid, and the stress influence radiation range curve of the representative door leaf beam grid in each cluster, determine the stress transfer correlation score between the representative door leaf beam grid and each other beam grid in each cluster, and the deformation coordination probability of two adjacent door leaf beam grids in each cluster.

[0058] It can be explained that deep neural networks are still used to determine the stress transfer correlation score between the representative phylum leaf beam grid and each other beam grid in each cluster, and the deformation coordination probability of two adjacent phylum leaf beam grids in each cluster. Among them, the stress transfer correlation score between the representative phylum leaf beam grid and each other beam grid in each cluster is determined by deep neural networks. It is a numerical score of the degree of correlation between the stress of the representative phylum leaf beam grid in the cluster and the stress generated by it under stress. The higher the score, the stronger the stress transfer correlation between the two. The deformation coordination probability of two adjacent phylum leaf beam grids in each cluster refers to the probability value of two adjacent phylum leaf beam grids in the cluster undergoing synchronous deformation when subjected to external loads. The higher the value, the greater the possibility of deformation coordination between the two.

[0059] Specifically, the deep neural network uses the estimated structural strength level of each cluster as a basic threshold. Combining the similarity of the defect positions of each beam grid in each cluster with the representative door leaf beam grid, it analyzes the stress transfer path between the representative door leaf beam grid and other beam grids in the cluster. The model calculates the stress transfer correlation score between the two based on the smoothness of the path and the correlation of the defect positions. Then, it extracts the influence degree data of different stress values ​​from the stress influence radiation range curve of the representative door leaf beam grid in each cluster. At the same time, it combines the structural position relationship of adjacent beam grids with the similarity of defect positions to simulate the deformation response law of adjacent beam grids under different stress states. By statistically analyzing the proportion of the number of times the two deform synchronously under different stress scenarios, it derives the deformation coordination probability of two adjacent door leaf beam grids in each cluster.

[0060] Step S323: Combine the stress transfer correlation score between the representative door leaf beam grid and each other beam grid in each cluster with the deformation cooperation probability of two adjacent door leaf beam grids in each cluster to determine the multiple cooperative door leaf beam grids corresponding to each representative door leaf beam grid.

[0061] Specifically, deep neural networks are used to analyze the data, and a threshold for the stress transfer correlation score is set. Door leaf beams with stress transfer correlation scores higher than the threshold are selected from within the cluster, forming a preliminary candidate set of cooperative beam lattices. Then, the deep neural network further selects door leaf beams from this candidate set, considering the deformation cooperation probability of two adjacent door leaf beams within each cluster, and selects those with deformation cooperation probabilities that meet a preset standard. Simultaneously, door leaf beams with low stress transfer correlation and low deformation cooperation probabilities with the representative door leaf beam are removed. Finally, the deep neural network summarizes the selected door leaf beams to determine multiple cooperative door leaf beams corresponding to each representative door leaf beam.

[0062] Step S33: Determine the corresponding detection thrust based on the representative door leaf beam grid and the cooperative door leaf beam grid respectively.

[0063] To better illustrate, a graph is a data structure used to represent entities and their relationships, where entities are nodes and relationships are edges.

[0064] Furthermore, in step S33, specifically: Multiple beam grid maps are constructed based on each representative gate leaf beam grid and its corresponding multiple cooperative gate leaf beam grids. A graph neural network model is used to process the beam grid maps to determine the detected thrust force of each representative gate leaf beam grid and the detected thrust force of each cooperative gate leaf beam grid. Each beam grid map includes multiple nodes and multiple edges. Each node includes a representative gate leaf beam grid node and its corresponding multiple cooperative gate leaf beam grids. The representative gate leaf beam grid node and the cooperative gate leaf beam grids are connected to form an edge.

[0065] It can be explained that the beam-and-leaf grid diagram includes multiple nodes and multiple edges. The multiple nodes include representative door-and-leaf beam-and-leaf grid nodes and multiple cooperative door-and-leaf beam-and-leaf grid nodes. In this embodiment, the representative door-and-leaf beam-and-leaf grid node is the central node, and each cooperative door-and-leaf beam-and-leaf grid node establishes an edge with the representative door-and-leaf beam-and-leaf grid node. Each node and edge has corresponding features. The representative door-and-leaf beam-and-leaf grid node represents the representative door-and-leaf beam-and-leaf grid within the corresponding cluster, and its node features include the ultrasonic scanning inspection data of the weld seam of the representative door-and-leaf beam-and-leaf grid and the information of the representative door-and-leaf beam-and-leaf grid. The cooperative door-and-leaf beam-and-leaf grid node represents the cooperative door-and-leaf beam-and-leaf grid corresponding to the representative door-and-leaf beam-and-leaf grid node, and its node features include the ultrasonic scanning inspection data of the weld seam of the cooperative door-and-leaf beam-and-leaf grid and the information of the cooperative door-and-leaf beam-and-leaf grid. An edge represents the association relationship between the representative door-and-leaf beam-and-leaf grid node and the cooperative door-and-leaf beam-and-leaf grid node; that is, an edge represents the positional relationship information between two nodes, including the relative orientation and distance between the representative door-and-leaf beam-and-leaf grid node and the cooperative door-and-leaf beam-and-leaf grid node.

[0066] Internal defects and stress deformation of the welds representing the door leaf beam grid directly affect the structural state of the cooperating door leaf beam grids in the same group; therefore, by constructing a beam grid diagram, the correlation between the representative door leaf beam grid and each cooperating door leaf beam grid can be clearly reflected.

[0067] To better illustrate, the graph neural network model includes a graph neural network (GNN) and fully connected layers. A graph neural network is a neural network model specifically designed for graph data processing. Its core is the transmission and aggregation of feature information between nodes and edges through a message-passing mechanism. It can simultaneously learn the feature information of a node itself and the relationships between nodes, and uncover the implicit global and local feature patterns in the graph to complete the classification, regression, and prediction of graph data. The detection push force for each representative door-leaf beam lattice and the detection push force for each cooperative door-leaf beam lattice are the quantified values ​​of the push force required to perform push tests on each representative door-leaf beam lattice and each cooperative door-leaf beam lattice, respectively.

[0068] The explanation is that by using a graph neural network to determine the detection thrust corresponding to the gate leaf beam grid, it is possible to accurately match the structural strength, weld quality, and deflection risk level of the corresponding gate leaf beam grid. Furthermore, it can accurately induce the deformation response of the gate leaf beam grid during testing without causing structural damage. The beam grid map naturally represents the relationship between the representative gate leaf beam grid node and the cooperative gate leaf beam grid node, which helps the graph neural network to capture the influence of the defects of the representative gate leaf beam grid itself on the structural state of the cooperative gate leaf beam grid in the same group; for example, the interaction between the distribution of internal weld defects, the positional relationship of the beam grid, and the deflection risk level of each beam grid surface.

[0069] Both the representative portal leaf beam grid nodes and the collaborative portal leaf beam grid nodes contain multiple node features, which provide rich contextual information for the graph neural network model. Specifically, the ultrasonic testing data of the welds of the representative portal leaf beam grid, along with its own structural information, helps to understand the distribution of internal weak points, while the ultrasonic testing data of the welds of the collaborative portal leaf beam grid, along with its structural information, provides information on the risk status of the same group of components. The ultrasonic testing data of the welds is used by the graph neural network model to accurately grasp the distribution of internal defects in the welds of the portal leaf beam grid, serving as the basis for determining the location of structural weaknesses and stress transmission paths. During the analysis, the beam grid map can comprehensively consider multiple factors such as the characteristics of internal weld defects, the deflection risk level, and the relative positional correlation of the beam grids, to achieve an accurate representation of the overall stress transmission relationship. For example, by analyzing the information in the beam grid map, the graph neural network can identify the extent of the defect's impact and determine the appropriate jacking force for the jacking test. In addition, the beam grid pattern helps to assess the potential impact of defects in representative leaf beam grids on the same group of cooperating leaf beam grids; for example, by analyzing the relative orientation and distance between leaf beam grids, the graph neural network can predict the spread of defects under stress, so as to select a reasonable test pushing force to ensure that the test results are accurate and reliable.

[0070] Specifically, the graph neural network performs dual learning of node features and structural features on multiple input beam grid maps. Through a message passing mechanism, feature information is transmitted and aggregated between representative door leaf beam grid nodes and cooperative door leaf beam grid nodes, while simultaneously learning the positional relationship information between door leaf beam grids carried by edges. Then, the graph neural network uses multi-layer graph convolution operations to mine the implicit mapping rules between the weld quality, surface deformation features, spatial positional relationships, and required jacking force of the door leaf beam grid. Combining the learned jacking force adaptation standards of different feature door leaf beam grids, the required detection jacking force of each door leaf beam grid is quantitatively calculated. At the same time, based on the feature differences between representative door leaf beam grids and cooperative door leaf beam grids within the same cluster, the jacking force value is finely adjusted, and finally, the detection jacking force of each representative door leaf beam grid and each cooperative door leaf beam grid is output.

[0071] As explained, in step S4, the jacking force is detected one by one on the corresponding jacking test gate leaf beam grid of the dam metal structure gate, and the deformation point cloud image of the dam metal structure gate under each detected jacking force is obtained.

[0072] It can be explained that the deformation point cloud image is a three-dimensional point cloud image used to present the deformation state of the gate surface under each corresponding detection thrust of the dam's metal structure gate. It includes the three-dimensional coordinates, deformation magnitude, and deformation direction of each point on the surface of the dam's metal structure gate.

[0073] As an optional implementation, in this embodiment, the deformation point cloud image is obtained by a three-dimensional laser scanning device to fully scan and collect the surface of the dam's metal structure gate during the process of each jacking force being applied to the corresponding jacking test gate leaf beam grid. The three-dimensional laser scanning device refers to a high-precision optical measuring instrument that can accurately measure the three-dimensional coordinates of an object's surface by emitting a laser beam and receiving reflected signals.

[0074] As explained, in step S5, a simulated deformation point cloud video of the dam's metal structure gate is generated based on the deformation point cloud image, depicting the process from the initial low water level condition to the design full water level limit condition.

[0075] Preferably, a deformation point cloud simulation model is used to generate a simulated deformation point cloud video of the dam's metal structure gate from the initial low water level condition to the design full water level limit condition. This video can completely simulate the deformation details of the dam's metal structure gate throughout the entire working process. In this embodiment, the deformation point cloud simulation model is a generative adversarial network (GAN), which is a deep learning model composed of a generator and a discriminator. The generator and discriminator can complete the data generation and simulation through adversarial game. The generator is responsible for learning the distribution pattern of real data and generating simulation data, while the discriminator is responsible for distinguishing whether the input data is real data or simulation data generated by the generator. The two are continuously optimized through iterative training, and finally the generator can output simulation data that closely matches the distribution pattern of real data.

[0076] It can be explained that the deformation point cloud image of the dam's metal structure gate under each tested jacking force records the local spatial three-dimensional coordinate change law of the gate under the real physical test of applying the corresponding tested jacking force to each gate leaf beam grid. This is used as the input of the deformation point cloud simulation model, providing accurate physical boundary conditions and basic reference form for the model to simulate the deformation evolution process when the water level and pressure change continuously. That is, it generates a three-dimensional point cloud video of the overall deformation evolution process of the dam's metal structure gate during the continuous process of the water level rising from the initial low water level to the design full water level limit state. Each frame in this video is a deformation point cloud image of the dam's metal structure gate under the corresponding water level load.

[0077] Specifically, a generator based on a generative adversarial network (GAN) receives deformation point cloud images of the dam's metal structure gate under each detected jacking force. It analyzes the relationship between deformation depth coordinates and force, and, based on the correlation between water level pressure and deformation, transforms the water level pressure from low to designed full water level into a continuous pressure field that gradually increases over time. The generator can extract the deformation evolution trend from the deformation point cloud images and use upsampling and interpolation techniques to deduce the three-dimensional displacement path of each spatial coordinate on the gate surface under the continuous pressure field, thus generating continuous simulated deformation point cloud frames in a time sequence. A discriminator based on the GAN receives the simulated deformation point cloud frames output by the generator and compares them to determine whether the physical deformation logic of the simulated frames conforms to real physical laws. If the discriminator identifies that the generated three-dimensional deformation depth is too large or the deformation edges are unnatural, it feeds the error back to the generator. The generator then adjusts the deformation evolution deduction process of the simulated deformation corona frames based on the discriminator's feedback and corrects unreasonable displacement paths. Based on the entire adversarial game process, the generator and discriminator repeatedly operate until the discriminator can no longer distinguish between the simulated frame and the real physical deformation. Finally, the generative adversarial network splices the continuous point cloud frames that conform to the physical laws in time sequence and outputs a simulated deformation point cloud video of the dam's metal structure gate from the initial working condition of low water level to the ultimate working condition of the design full water level.

[0078] As explained, in step S6, multiple high-deflection-risk door leaf beam grids are determined based on simulated deformation point cloud video.

[0079] To better illustrate this, a high-risk beam grid determination model was used to analyze and simulate deformation point cloud videos to identify multiple high-deflection-risk gate leaf beam grids. In this embodiment, the high-risk beam grid determination model is a Transformer model, which includes a self-attention mechanism and a feedforward neural network. During actual operation, the Transformer model can simultaneously focus on different time segments and spatial location information of the input sequence, and can also efficiently capture long-distance dependencies in the data and accurately capture dynamic correlation features in sequence changes. Furthermore, high-deflection-risk gate leaf beam grids refer to gate leaf beam grids whose deflection deformation exceeds the design safety threshold and poses a structural failure risk throughout the entire process from the initial low-water level condition to the design full-water level limit condition of the dam's metal structure gate.

[0080] Furthermore, in step S6, specifically: A preset safe yield deformation limit is used to decompose the simulated deformation point cloud video into a continuous sequence of point cloud video frames using a Transformer model. The Transformer model includes a self-attention mechanism and a feedforward neural network. The self-attention mechanism is used to obtain the relative displacement change rate of the point cloud coordinates of different door leaf beam grid regions in each frame, and to track the deformation acceleration of the door leaf beam grid during the pressure increase process. The feedforward neural network is used to capture the dynamic correlation features in the point cloud video frame sequence and compare them with the safe yield deformation limit. When the relative displacement change rate of the door leaf beam grid continuously approaches the safe yield deformation limit in multiple consecutive frames and the deformation acceleration does not slow down, the corresponding door leaf beam grid is determined to be a high-deflection-risk door leaf beam grid.

[0081] It can be noted that the simulated deformation point cloud video of the dam's metal structure gate from the initial working condition of low water level to the ultimate working condition of full water level can completely simulate the continuous dynamic process of the deformation of the dam's metal structure gate, and provide detailed evolution data in the time and space dimensions for the model to capture the critical point of sudden deformation acceleration under extreme pressure.

[0082] Specifically, the Transformer model reads the simulated deformation point cloud video of the dam's metal structure gate from the initial low water level condition to the design full water level limit condition, and then splits it into a continuous point cloud video frame sequence. The relative displacement change rate of the point cloud coordinates of different gate leaf beam grid regions in each frame of the point cloud video frame sequence is calculated through a self-attention mechanism. That is, the time series information is processed in parallel through a multi-head attention mechanism, and the deformation acceleration of a specific gate leaf beam grid during the pressure increase process is tracked. Then, the Transformer model is used to capture beam grid point cloud groups that deform slowly at low water levels but suddenly exhibit exponential growth in deformation displacement when approaching full water levels. These dynamic change characteristics are compared with the preset safe yield deformation limit. When it is found that the deformation displacement of the currently analyzed gate leaf beam grid continuously approaches the safe yield deformation limit in multiple consecutive video frames and the deformation acceleration does not slow down, it is determined that the gate leaf beam grid cannot withstand the ultimate stress of the design full water level. At the same time, the Transformer model is used to locate all gate leaf beam grid regions in the simulated deformation point cloud video that exhibit such nonlinear malignant deformation trends. These gate leaf beam grids that are identified as having an extremely high probability of structural instability are summarized and extracted to output multiple high-deflection-risk gate leaf beam grids.

[0083] Understandably, by analyzing image data collected from dam metal structure gates and implementing multiple steps, a highly efficient and accurate assessment of the deformation characteristics of dam metal structure gates under all operating conditions and the identification of high-deflection-risk gate leaf beams can be achieved. Specifically, based on image data, high-resolution images of the surface partitions of each gate leaf beam and corresponding gate leaf beam information are extracted, thereby determining multiple initial gate leaf beam information with deflection risk. Then, multiple gate leaf beams for jacking tests and the corresponding detection jacking forces are determined, and jacking force detection is performed to obtain deformation point cloud images. This generates simulated deformation point cloud videos of the dam metal structure gates from the initial low-water-level condition to the design full-water-level ultimate condition, identifying multiple high-deflection-risk gate leaf beams. In other words, through this comprehensive implementation method, the overall deformation characteristics and behavioral patterns of dam metal structure gates under various water level conditions can be systematically, efficiently, and accurately assessed, and high-deflection-risk gate leaf beams requiring special attention can be successfully identified, providing a reliable technical basis for gate safety monitoring and maintenance decisions.

[0084] Please see Figure 3 The second embodiment of the present invention provides a dam metal structure gate inspection system, the system comprising: Information acquisition and determination module 1 is used to: acquire image data based on the metal structure gate of the dam, use the beam grid information determination model to segment high-definition images of the surface partitions of each gate leaf beam grid from the image data, and extract the corresponding gate leaf beam grid information; Preliminary Risk Determination Module 2 is used to: combine the high-resolution surface partition images of each leaf beam grid with the corresponding leaf beam grid information to determine multiple preliminary flexural risk leaf beam grid information; Test scheme determination module 3 is used to: determine multiple jacking test door leaf beams and the detection jacking force of the corresponding jacking test door leaf beams based on the information of the initially selected deflection risk door leaf beams; The jacking detection execution module 4 is used to: use the detection jacking force to detect the jacking force of the corresponding jacking test gate leaf beam grid of the dam metal structure gate one by one, and obtain the deformation point cloud image of the dam metal structure gate under each detection jacking force. The simulated video generation module 5 is used to generate a simulated deformation point cloud video of the dam's metal structure gate from the initial low water level condition to the design full water level limit condition based on the deformation point cloud image. Module 6 for determining beam grid risk is used to: determine multiple high-deflection-risk gate leaf beam grids based on simulated deformation point cloud video.

[0085] It can be explained that the dam metal structure gate detection system is integrated based on the information acquisition and determination module 1, the preliminary risk determination module 2, the test plan determination module 3, the jacking detection execution module 4, the simulation video generation module 5, and the beam grid risk determination module 6. That is, by combining the various modules, the dam metal structure gate detection method described in any embodiment of the present invention is executed. It has the same beneficial effects as the dam metal structure gate detection method provided above, and will not be elaborated here.

[0086] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0087] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method for detecting metal structure gates of dams, characterized in that, The method includes: Step S1: Based on the image data collected from the metal structure gate of the dam, the surface partition high-definition image of each gate leaf beam grid is obtained from the image data by using the beam grid information determination model, and the corresponding gate leaf beam grid information is extracted. The information of the door leaf beam grid includes the surface contour feature information of the corresponding door leaf beam grid, the weld appearance morphology information, and the local deformation texture distribution coordinates; Step S2: Combine the high-resolution surface partition images of each door leaf beam grid with the corresponding door leaf beam grid information to determine multiple initial selections of door leaf beam grid information with deflection risk; The preliminary flexure risk gate leaf beam grid information includes the surface contour feature information of the corresponding gate leaf beam grid, weld appearance morphology information, local deformation texture distribution coordinates, flexure risk level and deformation texture anomaly sequence; Step S3: Determine multiple jacking test door leaf beam grids and the corresponding detection jacking force of the jacking test door leaf beam grids based on the initial selection of deflection risk door leaf beam grid information, including: Step S31: Cluster the information of the initially selected flexural risk door leaf beam grids to obtain multiple clusters and multiple representative door leaf beam grids, and collect the ultrasonic scanning test data of the weld of each representative door leaf beam grid, wherein each representative door leaf beam grid is the door leaf beam grid closest to the center of each cluster. Step S32: Determine multiple cooperating door leaf beam grids corresponding to each representative door leaf beam grid by using multiple clusters and ultrasonic scanning inspection data of the weld seams of each representative door leaf beam grid, and collect ultrasonic scanning inspection data of the weld seams of each cooperating door leaf beam grid. Step S33: Determine the corresponding detection thrust based on the representative door leaf beam grid and the cooperative door leaf beam grid respectively; Specifically, step S33 is as follows: Multiple beam grid maps are constructed based on each representative door leaf beam grid and its corresponding multiple cooperative door leaf beam grids. A graph neural network model is used to process the beam grid maps to determine the detected thrust force of each representative door leaf beam grid and the detected thrust force of each cooperative door leaf beam grid. Each beam grid map includes multiple nodes and multiple edges. Each node includes one representative door leaf beam grid node and multiple corresponding cooperative door leaf beam grid nodes. The representative door leaf beam grid node and the cooperative door leaf beam grid nodes are connected to form an edge. Among them, the detection thrust corresponding to the door leaf beam grid is determined by the graph neural network in the graph neural network model. It can be adapted to the structural strength, weld quality and deflection risk level of the corresponding door leaf beam grid, and can accurately induce the deformation response of the door leaf beam grid in the test without causing structural damage to the door leaf beam grid. Step S4: Use the jacking force detection method to detect the jacking force of the corresponding jacking test gate leaf beam grid of the dam metal structure gate one by one, and obtain the deformation point cloud image of the dam metal structure gate under each detected jacking force. Step S5: Generate a simulated deformation point cloud video of the dam's metal structure gate from the initial low water level condition to the design full water level limit condition based on the deformation point cloud image; Step S6: Identify multiple high-deflection-risk door leaf beam grids based on simulated deformation point cloud video.

2. The method for detecting a dam metal structure gate according to claim 1, characterized in that, The model for determining beam grid information is a convolutional neural network.

3. The method for detecting a dam metal structure gate according to claim 2, characterized in that, Step S1 includes: Step S11: Use an industrial camera to take multiple high-resolution images of the surface of the dam's metal structure gate to be inspected before leaving the factory, and integrate them to form image data; Step S12: Input the image data into the convolutional neural network, extract the features of the image data through convolution operation, segment the high-definition image of the surface of each dam metal structure gate, output the surface partition high-definition image of each gate leaf beam grid, and extract the corresponding gate leaf beam grid information.

4. The method for detecting a dam metal structure gate according to claim 1, characterized in that, Step S2, specifically: A convolutional neural network is used to filter high-resolution images of surface partitions and corresponding information of each door leaf beam grid, predicting door leaf beam grid information that is prone to flexural deformation during stress, and identifying multiple initial door leaf beam grid information with flexural risk.

5. The method for detecting a dam metal structure gate according to claim 1, characterized in that, Step S6 is as follows: A preset safe yield deformation limit is established. The simulated deformation point cloud video is split into a continuous point cloud video frame sequence using a Transformer model. The Transformer model includes a self-attention mechanism and a feedforward neural network. The self-attention mechanism obtains the relative displacement change rate of the point cloud coordinates of different door leaf beam grid regions in each frame, tracking the deformation acceleration of the door leaf beam grid during the pressure increase process. The feedforward neural network captures the dynamic correlation features in the point cloud video frame sequence and compares them with the safe yield deformation limit. When the relative displacement change rate of the door leaf beam grid continuously approaches the safe yield deformation limit in multiple consecutive frames, and the deformation acceleration does not slow down, the corresponding door leaf beam grid is determined to be a high-deflection-risk door leaf beam grid.

6. A dam metal structure gate inspection system, used to implement the dam metal structure gate inspection method as described in any one of claims 1 to 5, characterized in that, The system includes: The information acquisition and determination module is used to: acquire image data based on the metal structure gate of the dam, use the beam grid information determination model to segment high-definition images of the surface partitions of each gate leaf beam grid from the image data, and extract the corresponding gate leaf beam grid information; The initial risk determination module is used to: combine the high-resolution surface partition images of each leaf beam grid with the corresponding leaf beam grid information to determine multiple initial flexural risk leaf beam grid information; The test scheme determination module is used to: determine multiple jacking test door leaf beams and the corresponding detection jacking force of the jacking test door leaf beams based on the information of the initially selected deflection risk door leaf beams; The jacking detection execution module is used to: detect the jacking force of the corresponding jacking test gate leaf beam grid of the dam metal structure gate one by one using the detection jacking force, and obtain the deformation point cloud image of the dam metal structure gate under each detection jacking force; The simulated video generation module is used to generate simulated deformation point cloud videos of the dam's metal structure gate from the initial low water level condition to the design full water level limit condition based on the deformation point cloud images. The beam risk determination module is used to determine multiple high-deflection-risk gate leaf beams based on simulated deformation point cloud video.

Images