Underwater detection system and method for hydraulic structure

Abstract

The invention discloses an underwater detection system and method for a hydraulic structure, belongs to the field of underwater detection systems for hydraulic structures, and adopts fractal tree-shaped path planning to realize double-optimal solutions of detection efficiency and precision, namely a fractal algorithm based on a self-similar growth rule of fern leaves. A tree-shaped path is recursively generated by taking an initial detection point as a root node, a main branch can quickly cover a detection surface area to form a macroscopic skeleton path, a sub-branch dynamically adjusts the extension direction according to coordinates of key parts, the coverage rate of a curvature mutation area is improved, a detection blind area of a traditional fixed route is effectively eliminated, and the detection efficiency is improved. Through innovative combination of a bionic fractal algorithm, multi-source data intelligent fusion and dynamic feedback control, the problem of a one-cut detection strategy in traditional hydraulic underwater detection is systematically solved, and an efficient, accurate and intelligent technical scheme is provided for structural safety monitoring in the fields of hydraulic engineering, ocean engineering and the like. And the method has remarkable engineering application value and economic benefit.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of underwater detection system of hydraulic structures, and more particularly, to an underwater detection system and method of hydraulic structures. BACKGROUND

[0002] The underwater structure of hydraulic structures (such as dams, bridge piers, port facilities, etc.) is easily affected by factors such as water flow erosion, corrosion, and foundation settlement, and is prone to cracks, concrete deterioration, and structural deformation, which seriously threatens the safety of the project. At present, although the traditional underwater detection system and method of hydraulic structures have many advantages, the inventors have found that the following shortcomings still exist:

[0003] 1. The imbalance between global coverage and local focus of the detection path under complex structure, the traditional underwater detection path planning mostly adopts grid traversal, random roaming or preset flight path mode, which is essentially a "one-size-fits-all" detection strategy. First, the global coverage ability is relatively insufficient. For hydraulic structures with irregular curved surfaces and concave-convex structures, fixed grid or straight flight path is prone to cause detection blind spots due to rigid path, especially for the regions with sudden changes in structural curvature, which are difficult to effectively cover;

[0004] Secondly, the local focusing precision is insufficient. For key parts such as building connections and crack-prone areas, the traditional method needs to manually preset the encrypted detection area, lacks an adaptive focusing mechanism based on structural characteristics, and often has resource mismatching problems such as "missing detection of key areas" or "redundant detection of non-key areas";

[0005] 2. The bionic application of natural growth law in path planning is blank, for example, the fractal growth pattern of fern leaves in nature shows the efficient space filling ability of "global coverage by main veins and local focus by fine veins", that is, the main branches extend at the optimal angle to form a skeleton coverage, and each level of sub-branches dynamically adjusts the growth direction and density according to the environmental signals. This feature is highly consistent with the demand of "overall coverage + local enhancement" of hydraulic detection, but the existing technology has not introduced such fractal growth rules into underwater detection path planning, resulting in:

[0006] Lack of multi-scale path generation algorithm that can simultaneously consider "global traversal of detection surface" and "adaptive encryption of key parts";

[0007] It is difficult to dynamically adjust the path branch parameters based on real-time data such as structural curvature and defect probability, so as to construct a hierarchical detection network in complex three-dimensional space;

[0008] Based on the above, we propose a kind of underwater detection system and method of hydraulic structure, through bionic fractal algorithm, the natural growth rule is converted into engineering detection logic, for the first time in the underwater detection of hydraulic engineering " the organic unity of global skeleton coverage and local intelligent focus", break through the technical limitations of traditional path planning in complex structure " lose both sides for one", provide a new solution for the efficiency and accuracy of underwater detection. SUMMARY

[0009] 1. Technical problems to be solved

[0010] In view of the problems existing in the prior art, the purpose of the present application is to provide an underwater detection system and method of hydraulic structure, which converts the natural growth rule into engineering detection logic through bionic fractal algorithm, realizes the "organic unity of global skeleton coverage and local intelligent focus" for the first time in the underwater detection of hydraulic engineering, breaks through the technical limitations of traditional path planning in complex structure "lose both sides for one", and provides a new solution for the efficiency and accuracy of underwater detection.

[0011] 2. Technical scheme

[0012] In order to solve the above problems, the present application adopts the following technical scheme.

[0013] An underwater detection system of hydraulic structure, comprising a detection platform, a detection device, a fractal tree path planning module, a control module, a data processing module and a communication module;

[0014] The detection platform is provided with a propulsion device and a positioning module, which is used to carry the detection device and move in the underwater area of hydraulic structure according to the preset path, and the positioning module collects the position coordinates of the detection platform in real time and transmits them to the control module;

[0015] The detection device includes a camera array, a three-dimensional sonar module, a pressure sensor and a water quality sensor, each sensor synchronously collects data through a unified time stamp, and the digital signal converted by an analog-digital conversion module is transmitted to the data processing module through an internal bus;

[0016] The fractal tree path planning module is provided with a three-dimensional modeling interface and a fractal algorithm unit, the three-dimensional modeling interface receives the point cloud data of the underwater structure of hydraulic structure, generates a three-dimensional grid model containing the spatial coordinates of the key detection parts, and the key detection parts include but are not limited to the connection of hydraulic structure, the fractal algorithm unit is based on the self-similarity growth rule of fern leaf, takes the initial detection point as the root node, generates a multi-level tree-shaped detection path through recursive calculation, the extension direction of each branch is dynamically adjusted by the structure curvature in the three-dimensional model and the key detection part coordinates, forming the main path covering the detection surface and the sub-path focusing on the key area;

[0017] The control module comprises a path analysis unit and a device driving unit, the path analysis unit converts the fractal tree detection path into motion control instructions of the detection platform, the motion control instructions include but are not limited to moving speed, turning angle, hovering position instructions, the device driving unit dynamically adjusts the working parameters of the detection device according to different branch levels of the detection path, the working parameters of the detection device include but are not limited to camera array resolution and sonar scanning frequency;

[0018] The data processing module integrates a data fusion unit, an anomaly detection unit and a path optimization unit, the data fusion unit performs space-time registration on multi-sensor data to generate a detection data matrix containing position labels, the anomaly detection unit identifies cracks in image data through a convolutional neural network, constructs a structure contour model in combination with sonar point cloud data, detects structure deformation through a difference algorithm, and the path optimization unit generates an encrypted sub-path based on the spatial coordinates of abnormal points in the three-dimensional model when an abnormal area is detected, with the parent branch as the reference, the branch spacing of the sub-path is 1 / 2-1 / 3 of the parent branch;

[0019] The communication module uses a water acoustic communication protocol to realize data interaction between the detection system and the ground control center, and the data interaction includes three-dimensional model data, real-time detection data matrix and detection report.

[0020] Further, the specific working process of the fractal algorithm unit includes:

[0021] Based on the three-dimensional grid model, the boundary contour line of the detection surface is extracted, and 3-5 points are selected as initial detection points in the vertices of the contour line;

[0022] For each initial detection point, a first branch is generated at a main branch angle of 45°-60°, the branch length of the first branch is 1 / 5-1 / 3 of the maximum span of the detection surface, forming a main skeleton path covering the detection surface;

[0023] When the spatial distance between the branch end and the key detection part is less than a preset threshold, the preset threshold is 0.5-1.0 meters, a local focusing mechanism is triggered, a second branch is generated with the end as a new node, the branch angle of the second branch is adjusted to 20°-30°, and the branch length of the second branch is gradually shortened by a decay factor of 0.6-0.8 level by level until the maximum branch level is reached, and the maximum branch level is 8-10 levels.

[0024] Further, the data processing module is provided with a feedback interface with the fractal tree path planning module, when the structural defect confidence output by the anomaly detection unit is greater than or equal to 85%, a feedback signal containing the three-dimensional coordinates of the abnormal area is generated, and the fractal tree path planning module dynamically inserts encryption detection nodes within 1 meter around the abnormal area according to the feedback signal, and the distribution density of the encryption nodes is 3 times that of the conventional detection area.

[0025] Further, the camera array contains a multispectral camera for collecting structural surface texture images, temperature field distribution and concrete deterioration fluorescence signals, and each spectral image generates a comprehensive detection image through a feature fusion algorithm.

[0026] A method for underwater detection of hydraulic structures, comprising the following steps:

[0027] S1, three-dimensional model construction and key part marking, obtaining the point cloud data of the surface of the hydraulic structure by underwater laser scanning, generating a three-dimensional grid model with an accuracy of less than or equal to 5mm after noise reduction filtering, then marking the key detection parts of the building connection and crack prone area in the model, and generating a key part coordinate list containing geometric feature parameters;

[0028] S2, fractal detection path initialization, selecting the vertex with the best connectivity from the boundary vertices of the three-dimensional grid model as the initial detection point, determining the first branch direction of the initial detection point based on the Delaunay triangulation algorithm, then generating the initial detection path according to the fractal growth rule, wherein the first branch covers an area with a detection surface of greater than or equal to 80%, the second to third branches form a surrounding situation for the key detection parts, and the branch spacing is automatically adjusted according to the curvature of the detection surface;

[0029] S3, multi-sensor cooperative data acquisition, the detection platform moves along the initial detection path, the positioning module outputs the position coordinates at a frequency of 10Hz, and the control module dynamically adjusts the sensor working mode according to the path branch level:

[0030] Main path: the main path contains first to third branches, the sonar performs 360° scanning at a frequency of 2Hz, and the camera collects panoramic images at a frequency of 10fps;

[0031] Sub-path: the sub-path contains fourth and above branches, the sonar scanning frequency is increased to 5Hz, the camera is switched to a 10 million pixel high definition mode, and the key detection parts are focused;

[0032] S4, real-time data processing and path dynamic optimization, the data fusion unit synchronizes the sensor data in time and registers the sensor data in space, generates a detection data set with a position label, the anomaly detection unit detects defects in the image data through the trained FasterR-CNN model, when a suspected crack is detected at the building joint, the path optimization unit is triggered, and an encryption sub-path around the crack center coordinate is generated according to the fractal rule, and the control module drives the detection platform to move along the encryption sub-path to perform 360° detection, and the sonar synchronously generates a high-precision three-dimensional point cloud model of the region;

[0033] S5, detection result generation and transmission, the data processing module performs spatial interpolation and defect quantization analysis on the detection data set to generate a detection report containing the defect position, type and size, the communication module spatially correlates the detection report with the three-dimensional model, and transmits the detection report to the ground control center through the underwater acoustic communication link to support GIS map visualization display.

[0034] Further, the mathematical expression of the fractal growth rule is:

[0035] L n =L n-1 ×r,θ n =θ n-1 ±Δθ

[0036] wherein L n is the length of the nth branch, r is a length attenuation factor of 0.6-0.8, θ n is the angle of the nth branch, and Δθ is an angle adjustment amount of ±15°-±30°, and the adjustment direction is determined by the spatial coordinates of the key detection part.

[0037] Further, the path dynamic optimization process in step S4 comprises:

[0038] When the defect feature change rate in the continuous 3 detection data of the same key detection part is >10%, secondary encryption detection is triggered, and a primary grand branch is generated based on the original encryption sub-path, and the branch spacing is reduced to 1cm;

[0039] The detection platform adopts a spiral surrounding trajectory when moving on the encryption path, and the surrounding radius is 1.5 times the maximum size of the abnormal area, ensuring that the detection coverage rate is ≥99%.

[0040] Further, the step S4 further comprises: performing histogram equalization preprocessing on the RGB image collected by the camera, performing pixel-level registration on the depth image obtained by the sonar to generate an RGBD image containing color and depth information, and fusing the data of the pressure sensor and the positioning module by using an extended Kalman filter algorithm to correct the depth coordinate error of the detection platform in real time.

[0041] 3. Benefits

[0042] Compared with the prior art, the advantages of the present application are:

[0043] (1) The scheme adopts a fractal tree path planning to realize double-optimized solution of detection efficiency and accuracy, that is, a fractal algorithm based on the self-similarity growth rule of fern leaves recursively generates a tree path with an initial detection point as a root node, a main branch can quickly cover a detection surface area to form a macroscopic skeleton path, and a sub-branch dynamically adjusts an extension direction according to a key part coordinate to improve the coverage of a curvature mutation area and effectively eliminate the detection blind area of a traditional fixed route;

[0044] (2) The abnormality detection unit of the scheme jointly analyzes image data and sonar point clouds through a FasterR-CNN model and a difference algorithm, when a structural defect with a confidence degree ≥ 85% is detected, the path optimization unit automatically generates an encrypted sub-path based on a fractal rule, and the detection platform can complete path reconstruction and start secondary detection in a short time, compared with the traditional offline analysis-manual re-planning process, the degree of automation and the ability to capture sudden hazards are significantly improved;

[0045] (3) The control module of the scheme can dynamically adjust sensor parameters according to path branch levels, the main path (one to three branches) adopts sonar low-frequency scanning and camera panoramic mode, the energy consumption is reduced compared with full high-definition mode, the sub-path (fourth and above branches) switches to high-frequency scanning and high-definition acquisition, realizes the resource optimization configuration of “global rough measurement energy saving and local precise measurement efficiency improvement”, and improves the single detection endurance time;

[0046] (4) The scheme solves the problem of “one-size-fits-all” detection strategy in traditional hydraulic underwater detection through the innovative combination of bionic fractal algorithm, multi-source data intelligent fusion and dynamic feedback control, provides an efficient, accurate and intelligent technical solution for structural safety monitoring in the fields of water conservancy engineering and marine engineering, and has significant engineering application value and economic benefits. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1 is a schematic diagram of the system main body architecture of the present application;

[0048] Figure 2 is a schematic diagram of the system architecture of the present application Figure 1 ;

[0049] Figure 3 is a schematic diagram of the system architecture of the present application Figure 2 ;

[0050] Figure 4 is a mind map of the system key technologies and algorithms of the present application;

[0051] Figure 5The schematic diagram of the main steps of the method of the present application;

[0052] Figure 6 The schematic diagram of the main steps of the method of the present application Figure 1 ;

[0053] Figure 7 The schematic diagram of the main steps of the method of the present application Figure 2 ;

[0054] Figure 8 The schematic diagram of the main steps of the method of the present application Figure 3 .

[0055] Explanation of figure numbers:

[0056] 100, detection platform; 101, propulsion device; 102, positioning module; 200, detection equipment; 201, camera array; 202, three-dimensional sonar module; 203, pressure sensor; 204, water quality sensor; 300, fractal tree path planning module; 301, three-dimensional modeling interface; 302, fractal algorithm unit; 400, control module; 401, path analysis unit; 402, equipment driving unit; 500, data processing module; 501, data fusion unit; 502, anomaly detection unit; 503, path optimization unit; 600, communication module. DETAILED DESCRIPTION

[0057] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the specification of the present application; obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments, and all other embodiments obtained by a person of ordinary skill in the art without creative labor based on the embodiments in the present application are within the protection scope of the present application.

[0058] Embodiment 1:

[0059] Please refer to Figures 1-8 , the working principle of the underwater detection system and method for the water structure will be described in detail as follows:

[0060] I. System initialization and fractal path generation driven by three-dimensional model

[0061] 1. Technical details of three-dimensional model construction The underwater structure of the water structure is scanned in all directions by the underwater laser scanning equipment to obtain high-density point cloud data (point spacing ≤5mm). After the data is denoised by the bilateral filtering algorithm, the moving least squares method (MLS) is used to fit the surface to generate a three-dimensional grid model with an accuracy of ≤5mm. In the model, the key detection parts are marked by the following methods:

[0062] Geometric feature extraction: Calculate the curvature value of each area based on the grid model (using the PrincipalCurvature algorithm), and automatically mark areas with sudden changes in curvature (such as building connections and corners) as key areas;

[0063] Historical defect mapping: Combined with the operation and maintenance archives of hydraulic structures, the coordinates of historical cracks and leakage points are superimposed on the model to generate a weighted coordinate list of key locations (including X / Y / Z coordinates and defect type priority).

[0064] 2. Mathematical modeling and initialization of fractal tree paths The fractal algorithm unit (302) is inspired by the fern leaf growth model and generates a detection path through recursive iteration. The specific process is as follows:

[0065] Initial detection point selection: Calculate the connectivity index of the vertices on the boundary of the 3D mesh model based on the Delaunay triangulation algorithm (the higher the vertex degree, the better the connectivity). Select the first 3-5 vertices as the initial root nodes to ensure that the path covers the key positions on the boundary of the detection surface.

[0066] Main skeleton path generation: The primary branch extends at a main branch angle of 45°-60°, and its length is 1 / 5 to 1 / 3 of the maximum span (Lmax) of the detection surface (i.e. L1 = L max ×(0.2-0.33)), ensuring coverage of ≥80% of the detection surface; the secondary branch uses the end of the primary branch as a node, and dynamically adjusts the angle according to the curvature of the detection surface (the greater the curvature, the greater the angle adjustment Δθ, ranging from ±15° to ±30°), and the branch length is shortened by an attenuation factor r=0.6-0.8 (L2=L1×r), forming an encirclement around the key parts;

[0067] Local focus trigger mechanism: When the Euclidean distance between the branch end and the coordinates of the key part is less than the preset threshold (0.5-1.0 meters), the fractal dimension is triggered to increase, and three or more branches are generated at a small angle of 20°-30°. The length is gradually attenuated (maximum level 8-10 levels), realizing layered detection of "long-range coarse scanning → close-range fine scanning".

[0068] 2. Multi-sensor spatiotemporal collaboration and adaptive detection

[0069] 1. Collaborative working mechanism of sensor array The sensors on the detection platform (100) are synchronized through a unified hardware clock to ensure that the data collection time is consistent:

[0070] Camera array (201): a multispectral camera (visible light + infrared + fluorescence band) synchronously collects structural surface texture, temperature field distribution and concrete deterioration fluorescence signal;

[0071] Each spectral image generates a comprehensive detection image through a Laplacian pyramid fusion algorithm, enhancing the defect contrast (such as the highlighting effect of cracks in the fluorescence band);

[0072] Three-dimensional sonar module (202): The main path (one to three branches) uses low-frequency scanning to obtain 360° contour point cloud for global structure modeling; The sub-path (fourth and above branches) switches to high-frequency scanning, with reduced point spacing, and cooperates with the camera to realize "vision-sonar" stereo imaging;

[0073] Pressure sensor (203) and positioning module (102): The pressure sensor monitors the water depth pressure, and through the extended Kalman filtering algorithm and the data fusion of the positioning module, the platform depth coordinate error is corrected in real time.

[0074] 2. The dynamic parameter scheduling path analysis unit (401) of the control module converts the fractal path into platform motion instructions:

[0075] Mobile control: The main path uses constant speed cruise, the sub-path switches to low speed precision control mode, and realizes precise steering through the control of the propulsion device (101);

[0076] Device driving strategy: main path: higher camera frame rate, wider sonar scanning range (horizontal x vertical);

[0077] Sub-path: camera triggers "region of interest (ROI)" mode, only high-definition imaging of key parts (frame rate is reduced but pixel density is improved), sonar focus angle scanning range is reduced, and local scanning resolution is improved.

[0078] Three, data closed loop processing and intelligent feedback optimization

[0079] 1. The spatio-temporal registration and feature extraction data fusion unit (501) performs three-level processing:

[0080] Time synchronization: based on sensor timestamps, different rate data (such as 10Hz positioning data and 5Hz sonar data) are aligned through linear interpolation algorithm;

[0081] Spatial registration: Zhang's calibration method is used to establish the conversion relationship between sensor coordinate systems, map the RGB image pixel coordinates and sonar point cloud three-dimensional coordinates, and generate RGBD images with XYZ labels;

[0082] Feature fusion: histogram equalization preprocessing is performed on image data to enhance contrast, then FasterR-CNN defect detection model (training data contains 100,000+ water structure defect samples) is input, and crack length, width and other parameters are output, at the same time, sonar point cloud is fitted by moving least squares method to calculate structure deformation.

[0083] 2. Multi-stage path optimization and dynamic encryption mechanism

[0084] Primary encryption (abnormal trigger): when the anomaly detection unit outputs a defect confidence ≥ 85%, the path optimization unit (503) generates a ring-encircling encryption sub-path centered on the abnormal point according to the fractal rule:

[0085] The parent branch is the reference, and the sub-path branch spacing is set to 1 / 2-1 / 3 of the parent branch;

[0086] The platform moves along a spiral trajectory, with a ring radius of 1.5 times the maximum size of the abnormal area, ensuring that the detection coverage is ≥ 99%;

[0087] Secondary encryption (trend trigger): if the defect feature change rate of the same part is > 10% for 3 consecutive detections, the grandchild branch is triggered to generate, and the branch spacing is further reduced to 1 cm, combined with sonar scanning, to achieve sub-centimeter level defect monitoring;

[0088] Feedback loop: the data processing module transmits the three-dimensional coordinates of the abnormal area to the fractal path planning module through the feedback interface, and the latter inserts detection nodes with a density 3 times that of the regular area within 1 meter around the abnormal area, forming an intelligent closed loop of "detection-analysis-re-detection".

[0089] Four, underwater acoustic communication and detection result visualization communication module (600) adopts orthogonal frequency division multiplexing underwater acoustic communication protocol to solve the problem of underwater acoustic wave transmission delay and multipath effect:

[0090] Data compression: the detection data matrix (including RGBD image, point cloud, sensor parameter) is compressed to 1 / 10 of the original volume through wavelet compression algorithm, ensuring real-time transmission;

[0091] Spatial correlation: the defect coordinates in the detection report and the three-dimensional model are bound through a unified geographic coordinate system (such as WGS84), and after receiving by the ground control center, they can be visualized on the GIS map, facilitating engineers to make life prediction and maintenance decisions.

[0092] Example 2:

[0093] In view of the above example 1, further description is made, referring to Figures 1-8 Taking the actual application scenario of underwater detection of the dam foundation of a concrete gravity dam as an example, the working principle of the underwater detection system and method for such hydraulic structures is further explained:

[0094] I. Application scenario and detection object

[0095] Scenario: a concrete gravity dam with a height of 150m in a reservoir, the detection area is 0-40m deep underwater in the dam foundation, and the key detection parts include:

[0096] 1. Dam body and bedrock contact surface (300 m long, 15 transverse joints in total, joint width design value 20 mm);

[0097] 2. Dam foundation drainage hole group area (hole diameter 150 mm, hole spacing 3 m, a total of 200 holes);

[0098] 3. 2 crack suspected areas found in historical detection (No. A-01, B-02, located based on 2024 sonar data).

[0099] Detection target: Identify cracks ≥ 0.2 mm, structural deformation ≥ 5 mm, detect concrete spalling area ≥ 0.1 m2, and evaluate drainage hole blockage.

[0100] 0.1㎡, evaluate drainage hole blockage.

[0101] II. System hardware configuration and parameter setting

[0102]

[0103]

[0104] III. Detection details of the whole process

[0105] Step S1: Three-dimensional model construction and key part marking

[0106] 1. Point cloud data acquisition:

[0107] Use underwater laser scanner (scanning rate 800,000 points / s, accuracy ±3 mm) to scan along the dam foundation transversely to obtain original point cloud data (density 200 points / m2);

[0108] Remove water disturbance noise by bilateral filtering algorithm, and generate a three-dimensional grid model with an accuracy of 3 mm by moving least squares method (MLS).

[0109] 2. Key part marking:

[0110] Joint area: Automatically identify joint surface by curvature threshold method (curvature > 0.02 mm-1), import design drawing coordinates, and generate a line segment set from the starting point (0, -5, -40) to the endpoint (300, -5, -40) of the transverse joint;

[0111] Drainage hole group: Mark the center coordinates (15n, -8, -40) (n = 0, 1, …, 199) in the model based on the design drawing, with a radius of 0.075 m;

[0112] Historical suspected area: A-01 (120, -6, -35), B-02 (240, -7, -38), marked as a spherical area with a radius of 1 m.

[0113] Step S2: Fractal detection path initialization

[0114] 1. Initial detection point selection:

[0115] Based on the Delaunay triangulation algorithm, the connectivity of the model boundary vertices is calculated, and the end points of the transverse joint (0, -5, -40), (300, -5, -40) and the drainage hole area vertices (0, -8, -40), (300, -8, -40), (150, -8, -40) are selected as the initial root nodes.

[0116] 2. Main skeleton path generation:

[0117] First-level branch: From (0, -5, -40) to (34.4, -5, -16.2) with a main branch angle of 55° and a length of 60m (the maximum span of the detection surface is 300m x 1 / 5), covering 1 / 3 of the left side of the dam foundation;

[0118] Second-level branch: At a distance of about 8m (threshold value 1m) from the center of the drainage hole group (150, -8, -40), the angle is adjusted by Δθ = -25°, and the branch length is attenuated to 42m (60 x 0.7) according to r = 0.7, extending towards the hole group;

[0119] Path coverage: The spacing between the third-level branches is automatically adjusted to 20m (the curvature of the detection surface is relatively low), covering 82% of the detection surface, and the key parts are surrounded by the second-level branches.

[0120] Step S3: Multi-sensor cooperative data acquisition

[0121]

[0122]

[0123] Typical operation:

[0124] When the ROV approaches the transverse joint (150, -5, -40) to a distance of 0.9m along the second-level branch, local focusing is triggered:

[0125] The sonar switches to 5Hz and performs high-density scanning (point spacing 5cm) on the transverse joint area to generate a seam surface point cloud;

[0126] The multi-spectral camera focuses on the seam surface, and the visible light image resolution is improved to 4096 x 3072 pixels. The fluorescence camera detects a fluorescence intensity of 650cps (threshold value 500cps) in the A-01 area, which is marked as a suspected area of concrete deterioration.

[0127] Step S4: Real-time data processing and path dynamic optimization

[0128] 1. Data fusion and defect preliminary judgment:

[0129] Spatiotemporal registration: The Zhang calibration method is used to establish the conversion relationship between the camera and sonar coordinate systems, mapping the visible light image pixels (2000, 1500) to the sonar point cloud coordinates (120.5, -6.2, -35.1);

[0130] Defect Identification:

[0131] The Faster R-CNN model detected a 0.3mm wide crack in area A-01 (92% confidence) in the visible light image.

[0132] The sonar point cloud was used to construct a seam surface model using the Delaunay triangulation algorithm, and differential calculation revealed a local depression of 4.8 mm (close to the threshold of 5 mm).

[0133] 2. Generation of first-level encryption path:

[0134] The path optimization unit uses the A-01 coordinate (120, -6, -35) as the base point to generate a circular encrypted sub-path:

[0135] The parent branch spacing is 30m, and the child branch spacing is 10m (30×1 / 3), distributed radially;

[0136] The ROV moves along a spiral trajectory, circling a radius of 1.5m (maximum defect size 1m×1.5), completing 360° inspection.

[0137] 3. Secondary encryption trigger:

[0138] Three consecutive tests (10-minute intervals) revealed that the crack width increased from 0.3 mm to 0.35 mm (a change rate of 16.7% > 10%), generating a grandchild branch:

[0139] The branch spacing is 1 cm, the sonar point spacing is 0.5 cm, and the crack length is 1.8 m and the depth is 1.2 m.

[0140] The feedback interface transmits the abnormal coordinates to the path planning module, and detection nodes are inserted within a 1-meter range around A-01 (original spacing 5 meters → current spacing 5 / 3 ≈ 1.67 meters).

[0141] Step S5: Generating and transmitting test results

[0142] 1. Data analysis report:

[0143] Defect list:

[0144] Transverse crack area: 8 cracks were identified, with the largest width of 0.5 mm (located at A-01) and an average spacing of 25 m;

[0145] Drain hole group: 5 blockages were found (sonar reflection intensity > 80dB), with a blockage rate of 2.5%;

[0146] Quantitative analysis:

[0147] The concrete deterioration area is 0.8 square meters, and the average fluorescence intensity is 620 cps, which is determined as carbonation damage.

[0148] The structural deformation is less than 5mm, which meets the safety standard.

[0149] 2. Data visualization:

[0150] The communication module transmits the detection report and the three-dimensional model to the ground control center through the underwater acoustic communication protocol and displays it on the GIS map.

[0151] Four, technical parameters and effect verification

[0152] Index Implementation data Compared with traditional detection Detection coverage 99.2％ 70-85% (grid scanning missed corner) Crack recognition rate 95% (> 0.2mm) 60-75% (artificial interpretation easy to miss Positioning accuracy ±10cm ±50cm (traditional sonar positioning) Detection efficiency 6 hours to complete 40m water depth detection 10-12 hours (same range) Data utilization rate Multi-source fusion defect correlation 100% Single sensor false positive rate > 20%

[0153] Self-similarity of fractal path: from the first branch (covering the whole) to the tenth branch (focusing on millimeter-level defects), strictly following the growth rule of L n = L n-1 × 0.7, θ n = θ n-1 ± 20°;

[0154] Sensor dynamic scheduling: the camera ROI mode in the sub-path is completely matched with the sonar focusing area, the data transmission volume is reduced by 60%, but the resolution of the key area is improved by 4 times;

[0155] Feedback loop mechanism: from abnormal detection (confidence ≥ 85%) to encrypted detection execution, the delay is less than 30 seconds, ensuring real-time response.

[0156] The above is only the preferred specific embodiment of the present application; however, the protection scope of the present application is not limited thereto. Any skilled person in the art, according to the technical solution and the improvement concept of the present application, can make equivalent replacement or change within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. An underwater detection system for hydraulic structures, characterized in that: It comprises a detection platform (100), a detection device (200), a fractal tree path planning module (300), a control module (400), a data processing module (500) and a communication module (600); The detection platform (100) is equipped with a propulsion device (101) and a positioning module (102), and is used to carry the detection equipment (200) and move along a preset path in the underwater area of ​​the hydraulic structure. The positioning module (102) collects the position coordinates of the detection platform (100) in real time and transmits them to the control module (400); The detection device (200) includes a camera array (201), a three-dimensional sonar module (202), a pressure sensor (203), and a water quality sensor (204), wherein each sensor synchronously collects data through a unified timestamp, converts the data into a digital signal through an analog-to-digital conversion module, and transmits the digital signal to a data processing module (500) through an internal bus; The fractal tree-like path planning module (300) is configured with a three-dimensional modeling interface (301) and a fractal algorithm unit (302). The three-dimensional modeling interface (301) receives point cloud data of the underwater structure of a hydraulic structure and generates a three-dimensional grid model containing spatial coordinates of key detection parts, wherein the key detection parts include but are not limited to the connection parts of the hydraulic structure. The fractal algorithm unit (302) generates a multi-level tree-like detection path through recursive calculation based on the self-similar growth rule of fern leaves and an initial detection point as a root node. The extension direction of each level branch is dynamically adjusted according to the structural curvature and the coordinates of the key detection parts in the three-dimensional model, thereby forming a main path covering the detection surface and a sub-path focusing on the key area. The control module (400) comprises a path parsing unit (401) and a device driving unit (402), wherein the path parsing unit (401) converts a fractal tree-like detection path into motion control instructions for the detection platform (100), wherein the motion control instructions include but are not limited to instructions for movement speed, steering angle, and hovering position, and the device driving unit (402) dynamically adjusts the working parameters of the detection device (200) according to different branch levels of the detection path, wherein the working parameters of the detection device (200) include but are not limited to the resolution of the camera array (201) and the sonar scanning frequency; The data processing module (500) integrates a data fusion unit (501), an anomaly detection unit (502) and a path optimization unit (503); the data fusion unit (501) performs spatiotemporal registration on multi-sensor data to generate a detection data matrix containing position tags; the anomaly detection unit (502) performs crack recognition on image data through a convolutional neural network, builds a structural contour model in combination with sonar point cloud data, and detects structural deformation through a differential algorithm; and the path optimization unit (503) generates an encrypted sub-path based on the spatial coordinates of the abnormal point in the three-dimensional model and the parent branch when an abnormal area is detected, with the branch spacing of the sub-path being 1 / 2 to 1 / 3 of the parent branch; The communication module (600) uses an underwater acoustic communication protocol to implement data interaction between the detection system and a ground control center, wherein the content of the data interaction includes three-dimensional model data, a real-time detection data matrix, and a detection report.

2. The underwater detection system for hydraulic structures according to claim 1, characterized in that: The specific working process of the fractal algorithm unit (302) includes: Extract the boundary contour of the detection surface based on the 3D mesh model, and select 3-5 vertices of the contour line as the initial detection points; For each initial detection point, a primary branch is generated with a main branch angle of 45°-60°. The branch length of the primary branch is 1 / 5 to 1 / 3 of the maximum span of the detection surface, forming a main skeleton path covering the detection surface. When the spatial distance between the branch end and the key detection part is less than a preset threshold, the preset threshold is 0.5-1.0 meters, the local focusing mechanism is triggered, and a secondary branch is generated with the end as a new node. The branch angle of the secondary branch is adjusted to 20°-30°, and the branch length of the secondary branch is gradually shortened according to an attenuation factor of 0.6-0.8 until the maximum branch level is reached, and the maximum branch level is 8-10 levels.

3. The underwater detection system for hydraulic structures according to claim 1, characterized in that: A feedback interface is provided between the data processing module (500) and the fractal tree path planning module (300). When the confidence level of the structural defect output by the abnormality detection unit (502) is ≥85%, a feedback signal including the three-dimensional coordinates of the abnormal area is generated. The fractal tree path planning module (300) dynamically inserts encrypted detection nodes within a range of 1 meter around the abnormal area based on the feedback signal. The distribution density of the encrypted nodes is three times that of a conventional detection area.

4. The underwater detection system for hydraulic structures according to claim 1, characterized in that: The camera array (201) includes a multispectral camera for collecting structural surface texture images, temperature field distribution and concrete deterioration fluorescence signals, and each spectral image is used to generate a comprehensive detection image through a feature fusion algorithm.

5. A method for underwater detection of hydraulic structures, comprising the underwater detection system for hydraulic structures according to any one of claims 1 to 4, characterized in that: The following steps are involved: S1. 3D model construction and key area marking: underwater laser scanning is used to obtain surface point cloud data of hydraulic structures. After noise reduction and filtering, a 3D mesh model with an accuracy of ≤5mm is generated. Then, key inspection areas such as building joints and crack-prone areas are marked in the model, and a coordinate list of key areas containing geometric feature parameters is generated. S2. Initialize the fractal detection path. Select the vertex with the best connectivity from the boundary vertices of the 3D mesh model as the initial detection point. Delaunay triangulation algorithm is used to determine the first-level branch direction of the initial detection point. Then, the initial detection path is generated according to the fractal growth rule. The first-level branch covers ≥80% of the detection surface, and the second to third-level branches surround the key detection parts. The branch spacing is automatically adjusted according to the curvature of the detection surface. S3, multi-sensor collaborative data acquisition, the detection platform (100) moves along the initial detection path, the positioning module (102) outputs the position coordinates at a frequency of 10 Hz, and the control module (400) dynamically adjusts the sensor working mode according to the path branch level: Main path: The main path consists of one to three branches. The sonar scans 360° at a frequency of 2Hz, and the camera captures panoramic images at 10fps. Sub-path: The sub-path contains branches at level 4 or above. The sonar scanning frequency is increased to 5Hz, and the camera is switched to 10-megapixel high-definition mode to focus on key inspection areas. S4, real-time data processing and dynamic path optimization: the data fusion unit (501) performs time synchronization and spatial registration on the sensor data to generate a detection data set with location tags; the anomaly detection unit (502) uses the trained FasterR-CNN model to identify defects in the image data; when a suspected crack is detected at a building connection, the path optimization unit (503) is triggered to generate an encrypted sub-path around the base point based on the coordinates of the crack center and fractal rules; the control module (400) drives the detection platform (100) to move along the encrypted sub-path to perform 360° surround detection; and the sonar synchronously generates a high-precision three-dimensional point cloud model of the area; S5, generation and transmission of test results: the data processing module (500) performs spatial interpolation and defect quantification analysis on the test data set to generate a test report containing the defect location, type, and size; the communication module (600) spatially associates the test report with the three-dimensional model and transmits it to the ground control center via an underwater acoustic communication link, supporting GIS map visualization.

6. The underwater detection method for hydraulic structures according to claim 5, characterized in that: The mathematical expression of the fractal growth rule in step S2 is: L n =L n-1 ×r,θn=θn-1±Δθ Among them, Ln is the length of the n-th branch, r is the length attenuation factor of 0.6-0.8, θn is the n-th branch angle, Δθ is the angle adjustment amount of ±15°-±30°, and the adjustment direction is determined by the spatial coordinates of the key detection part.

7. The underwater detection method for hydraulic structures according to claim 5, characterized in that: The path dynamic optimization process described in step S4 includes: When the defect feature change rate in three consecutive inspection data of the same key inspection area is greater than 10%, the second-level encrypted inspection is triggered, and a first-level grandchild branch is generated based on the original encrypted sub-path, and the branch spacing is reduced to 1 cm; The detection platform (100) adopts a spiral orbit when moving in the encrypted path, and the orbit radius is 1.5 times the maximum size of the abnormal area, ensuring that the detection coverage rate is ≥99%.

8. The underwater detection method for hydraulic structures according to claim 5, characterized in that: The step S4 further includes performing histogram equalization preprocessing on the RGB image captured by the camera, performing pixel-level registration with the depth image acquired by the sonar, generating an RGBD image containing color and depth information, and using an extended Kalman filter algorithm to fuse the data of the pressure sensor (203) and the positioning module (102), thereby correcting the depth coordinate error of the detection platform (100) in real time.

Images