Multi-device linkage method at the edge of transmission tower based on self-organizing network and two-dimensional and three-dimensional fusion

Abstract

The present invention discloses a method for linking multiple devices at the edge of a transmission tower based on self-organizing networking and two- and three-dimensional fusion. This method relates to the field of intelligent transmission network technology and includes deploying multiple devices, embedding passive sensors, and generating a three-dimensional spatial reference model of the tower using a drone-mounted laser radar. The method also includes planning autonomous drone inspection routes based on the tower's three-dimensional spatial reference model, synchronously collecting laser point clouds, infrared thermal imaging, and backscatter data, extracting passive sensor deformation parameters by parsing channel state information, and constructing a multidimensional dataset. The multidimensional datasets are then fused to generate a dynamic risk heat map and annotate the risk level. When a high risk level is detected, the device deployment strategy is triggered and operational data is recorded. By constructing a dynamic monitoring system for transmission towers that couples mechanical, thermal, and geometric fields, the present invention provides a monitoring paradigm for smart grids that combines high-precision positioning, real-time response, and simplified operation and maintenance, thereby promoting the evolution of transmission facility management towards intelligent and unmanned operations.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent power transmission networks, and in particular to a multi-device linkage method at the edge of a power transmission tower based on self-organizing networks and two- and three-dimensional fusion. Background Art

[0002] In recent years, intelligent monitoring technology for transmission towers has gradually evolved into a fusion of multi-source sensing and edge computing. Existing technologies primarily rely on a combination of fixed sensor networks (such as strain gauges and fiber Bragg gratings) and periodic drone inspections, enabling anomaly identification through cloud-based big data platforms. At the data collection level, LiDAR and infrared thermal imaging technologies have been successfully applied to tower deformation and temperature field detection, while ad hoc networking technologies based on ZigBee and LoRa offer a viable solution for connecting devices in remote areas. However, existing technologies face significant bottlenecks: First, traditional sensors rely on wired power or battery maintenance, are susceptible to electromagnetic interference in complex metal tower structures, and are costly to deploy. Second, the lack of a unified benchmark for spatial and temporal alignment of multidimensional data (mechanical, thermal, and geometric) limits fault location accuracy and hinders the ability to trigger timely, precise, and coordinated responses. Summary of the Invention

[0003] In view of the above existing problems, the present invention is proposed.

[0004] Therefore, the present invention provides a multi-device linkage method at the edge of a transmission tower based on self-organizing networking and two- and three-dimensional fusion to solve the problems of temporal and spatial inaccuracy of multi-source heterogeneous data on transmission towers, insufficient fault location accuracy, and high delay in device linkage response.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0006] In a first aspect, the present invention provides a method for linking multiple devices at the edge of a transmission tower based on ad hoc networking and two- and three-dimensional fusion. The method comprises deploying multiple devices at the edge of a transmission tower, embedding passive sensors based on environmental backscatter, and generating a three-dimensional spatial reference model of the tower using a laser radar mounted on an unmanned aerial vehicle.

[0007] Based on the three-dimensional spatial benchmark model of the tower, the drone's autonomous inspection route is planned, point cloud data, infrared thermal imaging, and backscatter data are collected simultaneously, and the deformation parameters of the passive sensor are extracted by analyzing the channel state information to construct a multidimensional data set;

[0008] Fusing the multidimensional datasets at the edge node, generating a dynamic risk heat map and annotating the risk levels;

[0009] When a high risk level is detected, the equipment deployment strategy is triggered, the operation data is recorded, and the three-dimensional spatial benchmark model of the tower is updated.

[0010] As a preferred solution of the multi-device linkage method at the edge of a transmission tower based on self-organizing network and two- and three-dimensional fusion as described in the present invention, the multi-device deployment includes deploying edge computing nodes at the bottom of the tower, integrating ZigBee / Wi-Fi dual-mode self-organizing network communication tools, and physically connecting with the embedded airport management system.

[0011] As a preferred solution of the method for multi-device linkage at the edge of a transmission tower based on self-organizing networking and two- and three-dimensional fusion described in the present invention, the three-dimensional spatial reference model of the tower is obtained by flying a Velodyne VLP-16 laser radar on an unmanned aerial vehicle along a calibration path to collect tower point cloud data, aligning the point cloud data with the BIM model coordinate system through the ICP algorithm to generate a fusion coordinate system, and mapping the installation position coordinates of the passive sensor to the BIM model coordinate system to obtain the three-dimensional spatial reference model of the tower.

[0012] As an optimal solution for the multi-device linkage method at the edge of a transmission tower based on self-organizing networking and two- and three-dimensional fusion as described in the present invention, the planning of the autonomous inspection route of the drone is based on the three-dimensional spatial benchmark model of the tower. By setting the horizontal safety distance and the vertical layered scanning path, the drone's surrounding route is dynamically generated, and a real-time obstacle avoidance strategy is integrated.

[0013] As a preferred solution of the method for multi-device linkage at the edge of a transmission tower based on self-organizing network and two-dimensional and three-dimensional fusion described in the present invention, the channel state information includes signal strength, phase offset, propagation time difference and subcarrier index.

[0014] As a preferred solution of the multi-device linkage method at the edge of a transmission tower based on self-organizing networking and two- and three-dimensional fusion described in the present invention, the passive sensor deformation parameters include cantilever beam displacement and bolt loosening angle.

[0015] As an optimal solution for the multi-device linkage method at the edge of a transmission tower based on self-organizing networking and two- and three-dimensional fusion as described in the present invention, generating a dynamic risk heat map and marking the risk level is to use an AI model to associate the three-dimensional spatial benchmark model of the tower with the fused multidimensional data set, calculate the risk value, divide the risk level, and generate a dynamic risk heat map.

[0016] As a preferred solution of the multi-device linkage method at the edge of a transmission tower based on self-organizing networking and two- and three-dimensional fusion as described in the present invention, the equipment deployment strategy includes deploying drones to approach the target location to collect high-definition images, dispatching inspection robots to perform on-site maintenance, and correcting the three-dimensional spatial reference model of the tower at the same time.

[0017] In a second aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the method for linking multiple devices at the edge of a transmission tower based on self-organizing networks and two- and three-dimensional fusion as described in the first aspect of the present invention is implemented.

[0018] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the method for multi-device linkage at the edge of a transmission tower based on self-organizing network and two- and three-dimensional fusion as described in the first aspect of the present invention.

[0019] The beneficial effects of the present invention are as follows: by constructing a dynamic monitoring system for transmission towers that couples mechanical, thermal, and geometric fields, first, the passive sensor based on environmental backscatter analyzes the displacement of the cantilever beam through phase modulation, and combined with the coordinate mapping of the three-dimensional space reference model, it will reduce the error in detecting the loosening angle of the bolts, and accurately lock the microscopic defects through spatial positioning. Secondly, relying on edge computing and self-organizing network communication, the AI ​​model integrates multi-dimensional data to generate a dynamic risk heat map, and triggers drones to approach for shooting and robots for precise maintenance, forming a perception-decision-execution closed loop to improve response efficiency. In addition, passive sensors eliminate the need for battery maintenance through radio frequency power supply; the self-organizing network dynamic routing protocol maintains a high connectivity rate in the event of a single node failure, which is significantly better than a fixed topology solution. It provides a monitoring paradigm for smart grids that combines high-precision positioning, real-time response, and extremely simple operation and maintenance, and promotes the evolution of transmission facility management towards intelligence and unmanned operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0021] Figure 1 This is a flow chart of the multi-device linkage method at the edge of a transmission tower based on self-organizing networking and two- and three-dimensional fusion in Example 1.

[0022] Figure 2 This is a flow chart for generating a three-dimensional space reference model of an iron tower in Example 1.

[0023] Figure 3 Generate a logic diagram for the risk heat map in Example 1.

[0024] Figure 4 This is the data fusion processing diagram in Example 1. DETAILED DESCRIPTION

[0025] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0026] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0027] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0028] Example 1, with reference to Figures 1 to 4 This embodiment provides a method for linking multiple devices at the edge of a transmission tower based on a self-organizing network and two-dimensional and three-dimensional fusion, comprising the following steps:

[0029] S1. Deploy multiple devices at the edge of the transmission tower, embed passive sensors based on environmental backscatter, and use drones equipped with lidar to generate a three-dimensional spatial benchmark model of the tower.

[0030] S1.1. Multi-device deployment includes deploying an edge computing node at the base of the tower, integrating a ZigBee / Wi-Fi dual-mode ad hoc network communication tool (transmit power 20dBm, coverage radius 500m), and physically connecting to an embedded airport management system (with a built-in drone landing platform and wireless charging stations);

[0031] It should be noted that passive sensors capture energy and modulate backscattered data by coupling external RF signals. Passive sensors based on environmental backscattering are embedded at 3mm intervals at key locations such as the bolted joints and wire joints of the transmission tower. The surface of the passive sensor is coated with a metal layer of the same material as the tower to couple the Wi-Fi signal. Each passive sensor has a built-in miniature dipole antenna, and the antenna length is adjusted to the resonant wavelength of the 2.4GHz frequency band (31.25mm). The phase offset of the modulated data is adjusted by changing the displacement of the cantilever beam (accuracy ±0.1μm). The phase offset is linearly related to the bolt loosening angle.

[0032] S1.2. The tower's 3D spatial benchmark model is constructed by using a drone equipped with a Velodyne VLP-16 lidar to fly along a calibrated path, collecting tower point cloud data. The point cloud data is then aligned with the BIM (Building Information Model) coordinate system (existing technology) using the ICP algorithm to generate a fused coordinate system. The coordinates of the passive sensor's installation locations are then mapped to the BIM model coordinate system to obtain the tower's 3D spatial benchmark model.

[0033] It should be noted that the tower design drawings (format: IFC) were imported into the BIM model coordinate system using a DJI Matrice 300 RTK drone equipped with a Velodyne VLP-16 lidar (16 beams, vertical field of view ±15°, horizontal accuracy ±3cm), with a scanning frequency of 10Hz and a point cloud density of 300,000 points / second. The redundant components were cleaned up, the geometry and attribute data of the main materials (angle steel, bolts) were retained, and the coordinate system was converted to WGS84. UTM; extract the tower's outer contour in the BIM model coordinate system, extend it 20 meters outward to create a safe flight boundary, and plan the route using the contour-level cutting method. This method uses 5-meter intervals (e.g., 0m, 5m, 10m, etc.) around the tower. Each layer generates a closed polygonal path around the tower (radius = tower width + 10m), with a flight altitude error control of ±0.5m. The drone's flight speed is set to 3m / s, and the lidar tilt angle is adjusted to -10°. The drone flies along the calibrated path, with the lidar outputting point cloud data at a 10Hz frequency, recording timestamps in real time. Intensity correction is enabled, and noise is filtered based on material reflectivity.

[0034] It's important to note that the DJI Matrice 300 RTK drone integrates high-precision RTK (Real-Time Kinematic) technology, designed for complex mission scenarios such as inspection, surveying, and security. RTK technology utilizes real-time differential correction between a ground base station and the drone to eliminate satellite signal errors and significantly improve positioning accuracy. The RTK+GNSS dual-redundancy system supports multiple satellite systems, including GPS, GLONASS, Galileo, and BeiDou, achieving centimeter-level positioning accuracy (±1 cm horizontally, ±1.5 cm vertically). The Velodyne VLP-16 lidar utilizes 16 vertically spaced laser emitters with a vertical field of view of ±15° (30° total) and a 360° horizontal field of view. It outputs 300,000 point cloud data points per second, with an adjustable horizontal angular resolution of 0.1° to 0.4°.

[0035] The drone outputs positioning data, which is aligned with the lidar timestamp via the NMEA-0183 protocol. Five control points are set on the ground (the four corners and the center of the tower base), and reflective markers are attached. A total station (such as the Leica TS16) is used to measure the precise coordinates (with an error of ±2mm) to serve as the reference for point cloud registration.

[0036] Voxel filtering (voxel size 0.05m) was used to downsample the point cloud registration benchmark, retaining the main structural points of the tower (removing interference points such as vegetation and flying birds). A statistical outlier removal algorithm was applied to calculate the average distance of 50 points within the neighborhood (radius 0.1m) of each main structural point of the tower. Points exceeding the mean + 3σ (σ is the standard deviation of the distance) were removed. A seven-parameter transformation method (3 translations + 3 rotations + 1 scaling) was used to calculate the transformation matrix from the control point coordinates. This transformation matrix was then used to map all point cloud data to the BIM model coordinate system.

[0037] The Iterative Closest Point (ICP) algorithm is used for registration in two stages, including coarse registration and fine registration. In coarse registration, the endpoints of the BIM model angle steel are selected as feature points and RANSAC matching is performed with the point cloud data (maximum number of iterations is 1000). In fine registration, the coarse registration result is used as the initial value and the entire point cloud is iterated (maximum number of iterations is 50, and the convergence conditions are translation < 0.01m and rotation < 0.1°).

[0038] It should be noted that RANSAC matching refers to randomly selecting 3 pairs of BIM angle steel endpoints and suspected angle steel endpoints in the point cloud, calculating the rigid body transformation (rotation + translation), verifying the number of inliers (the inlier threshold is set to 5 cm), reaching 1000 iterations or finding a transformation matrix with an inlier rate ≥ 90%, and outputting the optimal coarse registration result; full point cloud iteration is based on the coarse registration result, constructing a point cloud KD-Tree index, quickly matching the corresponding points of the BIM model and the point cloud, assigning higher weights to high-confidence areas in the point cloud (such as the angle steel surface), reducing the influence of noise points, and terminating when the translation change is <0.01m and the rotation change is <0.1° in 10 consecutive iterations, or reaching the upper limit of 50 iterations.

[0039] Semantically associate the registered point cloud data with BIM model components (such as bolt holes and crossarms), assign attribute labels to the point clouds (such as tower leg angle L3 and insulator bracket), and generate a fused coordinate system.

[0040] It should be noted that the point cloud after segmentation and registration based on geometric features (such as curvature, normal vector, density) is extracted to extract local point cloud clusters corresponding to BIM components (such as cylindrical point cloud of bolt holes and plane point cloud of crossarms); the target component (such as the center point of the crossarm and the axis of the bolt hole) is located by coordinates in the BIM model to obtain the bounding box, geometric parameters and attribute information (such as component ID and material specifications); the segmented local point cloud cluster is spatially aligned with the BIM component (such as fine-tuning the position through the ICP algorithm) to verify the geometric similarity (such as diameter and angle); the attributes of the BIM component (such as "tower leg angle L3" and "insulator bracket-model X") are associated with the corresponding local point cloud cluster, and point cloud attribute labels (including component name, functional classification, and maintenance record) are generated; based on the BIM model coordinate system, the point cloud registration benchmark (rotation matrix, translation vector) is applied to convert the point cloud data to the same coordinate system to generate a fused coordinate system; an associated index between the point cloud data and the BIM model is established to support automatic synchronization of labels and spatial relationships when new point cloud data is added or the BIM model is modified.

[0041] When embedding passive sensors, use an RTK mobile station (such as Trimble R12, Trimble R12 real-time dynamic positioning measurement system) to measure the 3D coordinates of the installation location, record the passive sensor ID, and map the measured 3D coordinates of the installation location to the BIM model coordinate system through a transformation matrix. Based on the passive sensor coordinates, the fused coordinate system is updated through automatic matching and manual correction to generate a 3D spatial reference model of the tower.

[0042] It should be noted that automatic matching searches for the nearest BIM component (such as a bolt hole) within a 5cm radius of the BIM model coordinate system and binds the passive sensor to the component; manual correction is to manually mark the coordinates in the BIM model coordinate system and associate the component when automatic matching fails (such as the passive sensor is located on a non-standard part).

[0043] It should be noted that passive sensors based on environmental backscatter are embedded in key locations such as bolts and wire joints. They are powered by Wi-Fi signals and modulate phase offsets to achieve micron-level deformation monitoring (cantilever beam displacement accuracy of ±0.1μm). The drone, equipped with a lidar, flies along a calibrated path, aligning the point cloud with the BIM model using the ICP algorithm. Combined with RTK positioning, the sensor coordinates are mapped to form a fused coordinate system. The passive sensors are powered by radio frequency, eliminating battery dependence. Their surface is coated with a metal layer made from the same material as the tower to improve signal coupling efficiency, achieving maintenance-free operation for more than 10 years in harsh environments (traditional battery sensors need to be replaced every 1-2 years). Through semantic-level alignment of point cloud data and BIM models (error <5cm), a unified coordinate system is provided for subsequent data fusion, solving the spatial inaccuracy problem of traditional multi-source data (traditional methods have an error of >1m). Automatic matching (searching for BIM components within a 5cm radius) is combined with manual correction to ensure accurate mapping of passive sensor coordinates for non-standard parts (such as reinforcement brackets), increasing the adaptation rate from 70% to 98%.

[0044] S2. Based on the three-dimensional spatial reference model of the tower, the autonomous inspection route of the UAV is planned, and laser point cloud, infrared thermal imaging and backscattering data are collected simultaneously. The deformation parameters of the passive sensor are extracted by analyzing the channel state information to construct a multidimensional data set.

[0045] S2.1. Planning the autonomous inspection route of the UAV is based on the three-dimensional spatial benchmark model of the tower. By setting the horizontal safety distance and vertical layered scanning path, the UAV's circular route is dynamically generated and a real-time obstacle avoidance strategy is integrated.

[0046] It should be noted that the horizontal safety distance is set at 1.2H based on the tower height H (unit: meters). For example, when the tower is 50 meters high, the drone's flight radius remains at 60 meters, and the vertical scanning layer is divided into 5-meter height layers. The drone flies at a constant speed of 3m / s, and the flight path of each layer is a closed spiral (pitch 5 meters) around the tower. Based on known obstacles in the 3D spatial reference model of the tower (such as adjacent towers and trees), an AI algorithm is used to plan local obstacle avoidance paths in real time, with a minimum obstacle avoidance distance of 2 meters.

[0047] S2.2,CSI is obtained by analyzing the physical layer characteristics of wireless communication signals, including signal strength, phase offset, propagation time difference and subcarrier index;

[0048] The passive sensor deformation parameters include cantilever beam displacement and bolt loosening angle;

[0049] Multi-sensor collaborative collection of point cloud data, infrared thermal imaging and backscatter data;

[0050] Specifically, the Velodyne VLP-16 radar operates in 10-line scan mode, with a vertical angular resolution of 2°, a horizontal angular resolution of 0.1°, a point cloud density ≥ 300 points / square meter, and a data output frequency of 10Hz. The FLIR T1020 camera is set to a temperature range of -20°C to 150°C, a thermal sensitivity of 0.05°C, and a frame rate of 30fps. It is triggered synchronously with the lidar hardware. Night mode is enabled with an 850nm near-infrared LED array (power 10W, illumination distance 30 meters), a camera exposure time of 1 / 60s, and an ISO of 1600.

[0051] The edge node sends 802.11n beacon frames (center frequency 2.432 GHz, bandwidth 20 MHz) with a 10 ms period, a transmit power of 20 dBm, and a coverage radius of 50 meters. The passive sensor changes the antenna impedance by cantilever beam displacement (range ±0.5 mm, resolution 0.01 mm), encoding the displacement data as a phase offset in the reflected signal. The modulation depth is linearly related to the displacement.

[0052] The edge node uses an Intel 5300 network card to capture channel state information, extract the phase values ​​of 30 subcarriers, and calculate the phase difference between adjacent subcarriers;

[0053] According to the phase difference of adjacent subcarriers, the displacement of the cantilever beam is calculated and expressed as,

[0054] ;

[0055] in, represents the cantilever beam displacement, represents the phase difference between adjacent subcarriers, Indicates the Wi-Fi wavelength, =12.5cm;

[0056] Using the cantilever beam displacement, the bolt loosening angle is calculated and expressed as,

[0057] ;

[0058] in, Indicates the bolt loosening angle, Indicates the effective length of the bolt (e.g. =10cm, =0.2mm ≈1.15°);

[0059] The point cloud data is converted into a voxel grid (voxel size 0.1m×0.1m×0.1m). Each voxel stores the point cloud density, mean reflection intensity, and maximum height difference. The infrared thermal image is cropped into a 512×512 pixel area, the temperature value is normalized to [-1,1], and the local temperature gradient is extracted (such as calculating the temperature difference using a 3×3 sliding window). The backscattered data is grouped by the passive sensor ID and the bolt loosening angle is calculated. The sliding window mean of (window length 1 second, step length 0.1 second);

[0060] Combine laser point cloud voxels, infrared pixels and passive sensors The values ​​are aligned by timestamp interpolation with a time window of 10ms to ensure spatial consistency of data at the same moment (for example, the temperature at the bolt coordinate (10.2, 5.3, 15.0) corresponds to the deformation data), and are mapped to the three-dimensional spatial benchmark model of the tower to form a multidimensional dataset.

[0061] It should be noted that the horizontal safety distance (1.2H) and vertical layered scanning (5m interval) are set based on the three-dimensional spatial benchmark model of the tower, and AI real-time obstacle avoidance is used; the lidar (10Hz point cloud), infrared camera (0.05℃ thermal sensitivity) and backscatter data (10ms periodic analysis) are triggered synchronously at the hardware level; the 360° surface of the tower is covered by a closed spiral route (pitch 5m), and the point cloud density is ≥300 points / ㎡, which is 80% lower than the omission rate of traditional single-view scanning; based on the phase difference calculation of the channel state information (CSI) (Δφ=2πd / λ), the cantilever beam displacement analysis delay is compressed to 10ms; voxel filtering (0.05m³) and statistical outlier removal (3σ threshold) filter out vegetation and bird interference, and the proportion of valid data is increased from 60% to 95%.

[0062] S3. Fusion of multidimensional datasets at the edge node, and generation of dynamic risk heat maps with risk level annotations.

[0063] Use AI models to correlate the tower's 3D spatial benchmark model with the fused multidimensional dataset, calculate risk values, categorize risk levels, and generate dynamic risk heat maps.

[0064] Use a 3D convolutional layer (kernel size 3×3×3) with a voxel grid as input to extract spatial structural features;

[0065] It should be noted that the first layer of 3D convolution kernel (3×3×3) scans the voxel grid and identifies local geometric mutations through weight learning, including hole response and connector recognition. Hole response is a sudden drop in voxel density corresponding to an area with a high activation value (such as missing bolt holes and structural damage); connector recognition is to capture the voxel pattern of small-scale regular geometry such as bolt heads and angle steel end faces, superimposing the ReLU activation function and batch normalization (BatchNorm) to suppress noise voxel interference. By stacking multiple layers of 3D convolution (such as 4-6 layers), the physical range associated with a single voxel is gradually expanded. Level 1: identifies the direction of a single angle steel (such as horizontal / vertical), Level 2: locates the intersection of multiple angle steels (such as the connection area of ​​the tower leg node plate), and Level 3: maps the topological structure of the overall frame (such as the quadrilateral base of the transmission tower and the symmetry of the crossarm distribution). In the deep layer, strided convolution (StridedConvolution, such as stride = 2) is used to compress the spatial dimension and retain high-order semantics (such as uniform bolt distribution and key stress nodes). The high-resolution detail features of the shallow layer (such as the position of the bolt hole) and the deep global structural features (such as the inclination angle of the tower body) are spliced ​​along the channel dimension to form spatial structural features.

[0066] Use a 2D convolutional layer (kernel size 5×5) to input the temperature gradient map and extract thermodynamic features;

[0067] It should be noted that the 5×5 convolution kernel learns the local temperature gradient, strengthens the temperature mutation boundary (such as the transition zone between the loose bolt area and the normal area), filters the environmental thermal noise through batch normalization (BatchNorm) and activation function (such as ReLU), outputs the hot spots on the equipment surface corresponding to the high activation area (such as the locally high-temperature bolt connection), marks the suspected abnormal positions, and uses the bolt position, weld morphology and other features extracted by the spatial structure branch (3D convolution) as the attention mask (AttentionMask) to guide the thermal convolution to focus on the relevant area. Through deep convolution stacking (such as ResNet blocks), long-distance thermal correlation is established (for example, the temperature symmetry anomaly at both ends of the crossarm indicates structural stress imbalance), forming thermodynamic characteristics.

[0068] Using the LSTM layer, input Value sequence, extracting temporal deformation features;

[0069] It should be noted that the LSTM unit receives displacement / strain values ​​step by step, regulates the information flow through the forget gate, input gate and output gate, retains the long-term trend (such as the seasonal change of bridge deflection), and inputs multi-channel data such as displacement, strain, temperature (environmental interference term) at the same time, learns cross-influences (such as temperature increase causing steel beam expansion deformation), and LSTM captures the pattern of continuous increase in strain value and increasing rate, combined with historical The critical fracture time is predicted by the value sequence, and the periodic oscillation of bridge deflection caused by the temperature difference between day and night (expansion during the day and contraction at night) is identified, forming a time series deformation feature.

[0070] Concatenate the output feature vectors of the three branches into fused features (dimension: 1024);

[0071] It should be noted that the spatial-attribute attention mechanism is used to assign weights, for example, the bolt area is assigned to the passive sensor The value weight is 0.7, the infrared temperature weight is 0.3, and the infrared temperature weight is 0.6 and the point cloud density weight is 0.4 for the wire area;

[0072] Collect historical laser point clouds (500GB), historical infrared images (200,000 images) and historical passive sensor time series data of transmission towers ( and infrared temperature ΔT records), covering three types of scenarios: normal, low risk, and high risk, inputting historical laser point cloud voxels (spatial density, reflection intensity), historical infrared temperature gradient maps, and historical Value sequence, using weighted cross entropy loss to train the AI ​​model;

[0073] The fused features are input into the fully connected layer (512 nodes) of the trained AI model. The output layer outputs the risk value and classifies the level into three categories: normal, low risk, and high risk based on the mechanical parameter threshold and thermodynamic parameter threshold.

[0074] Specific, high risk: >1.2° or infrared temperature ΔT>10°C;

[0075] Low risk: >0.8° or infrared temperature ΔT>5°C;

[0076] normal: ≤0.8° and infrared temperature ΔT≤5℃;

[0077] It should be noted that according to the transmission tower bolt preload design specification (such as DL / T 646-2012), when the loosening angle exceeds 30% of the material yield strength, the structural stability decreases significantly, indicating that the bolt preload loss reaches 40% when θ=1.2°, reaching the critical failure threshold. Therefore, the high-risk threshold of the mechanical parameter is set at 1.2°. According to historical operation and maintenance records, The failure rate increases exponentially in the range of 0.8°~1.2°, which requires attention but is not a high risk. Therefore, the low risk threshold of mechanical parameters is set to 0.8°; based on the expansion coefficient of the wire connector material (such as aluminum), A temperature rise of 10°C will cause the contact resistance to increase by 15%, triggering the risk of local overheating and meltdown. Therefore, the high-risk threshold of the thermodynamic parameter threshold is set to 10°C. Based on the national standard GB / T2314, a temperature rise of 5°C in the conductor is the starting point of abnormal temperature rise, which requires continuous monitoring. Therefore, the low-risk threshold of the thermodynamic parameter threshold is set to 5°C.

[0078] The classification results are mapped to the 3D spatial benchmark model of the tower according to the 3D coordinates, and each voxel is labeled with the risk level (RGB color coding: red - high risk, yellow - low risk, green - normal);

[0079] Based on the finite element stress distribution model (existing technology), the impact range of high-risk areas is expanded outward (for example, areas where loose bolts cause increased stress in adjacent components). The stress attenuation radius is calculated based on material properties (such as a steel elastic modulus of 210 GPa) with the high-risk voxel as the center (for example, the stress drops by 50% within a radius of 2 meters). Within the attenuation radius, the risk level is linearly reduced with distance (for example, a distance of 0.5 meters is marked as high risk, and a distance of 1 meter is reduced to low risk). For voxels affected by stress diffusion, the original risk level and the diffusion weight are superimposed (for example, the risk level within 1 meter of the high-risk voxel is increased by one level). If the diffusion causes the new area to reach the mechanical parameter threshold and thermodynamic parameter threshold (for example, the original normal area becomes low risk due to stress propagation), recalculation is triggered.

[0080] It should be noted that the three branches of 3D convolution (spatial structure), 2D convolution (thermodynamics) and LSTM (temporal deformation) are integrated, and the spatial-attribute attention mechanism dynamically assigns weights; the risk area is expanded based on the finite element stress propagation model, and the stress attenuation radius is calculated according to the material properties (such as the elastic modulus of steel is 210GPa); the bolt area is assigned The weight is 0.7, the temperature weight is 0.3, and the conductor area temperature weight is 0.6, which reduces the false alarm rate by 40% compared with the traditional single threshold judgment; with the high-risk voxel as the center, the associated area is dynamically marked according to the stress attenuation radius (such as linear attenuation of risk level within a radius of 2m), and the early warning coverage rate is increased by 50%.

[0081] S4. When a high risk level is detected, the equipment deployment strategy is triggered and the operation data is recorded, and the three-dimensional space benchmark model of the tower is updated.

[0082] Edge nodes receive high-risk alerts output by AI models (such as bolt loosening angle =1.3°, temperature ΔT=12℃), automatically retrieve historical data of coordinates (past 24 hours) Change trends, temperature records), verify whether it is a persistent anomaly (such as Sustained growth > 0.1° / min);

[0083] If confirmed to be valid, start the priority sorting and assign task weights (e.g. main material > auxiliary material) based on the criticality of the tower structure (e.g. main material > auxiliary material) and risk level (high risk > low risk > normal) (e.g. main material bolt loosening weight = 10, auxiliary material = 3);

[0084] The event coordinates (e.g., (8.7, 3.5, 12.0)) and task instructions are broadcast over the ad hoc network. After receiving the instructions, the drone and inspection robot provide feedback on their current status (battery level, location, and task queue). The edge node selects the optimal device to perform the task, based on both drone and robot selection criteria. The drone selection criteria are closest to the target (<200m), battery level >40%, and no other high-priority tasks. The robot selection criteria are an obstacle-free path to the target and matching robotic arm tools (e.g., a torque wrench compatible with an M20 bolt).

[0085] The drone plans its approach path based on the tower's three-dimensional spatial benchmark model, including a global path and a terminal path. The global path is a B-spline curve from the current hovering point to the target point, with a horizontal obstacle avoidance distance of ≥ 2m. The terminal path switches to ultra-low altitude mode (1.5m high) when it is 5m away from the target point, and activates the lidar obstacle avoidance (obstacle detection distance accuracy of ±5cm).

[0086] A 30x optical zoom lens (resolution 4096×2160) captures bolt details from multiple angles (front, side, and top). A ring light (5000 lux) is simultaneously activated to ensure nighttime clarity. Images are then transmitted back to edge nodes in real time. A lightweight YOLOv5s model is used to detect detailed defects such as bolt corrosion and cracks, generating an image diagnostic report (e.g., "15% of the bolt surface is rusted").

[0087] If the image diagnosis report confirms that the risk is valid, the robot will be triggered to be dispatched;

[0088] If it is a false alarm (such as bird obstruction), terminate the mission and record the reason for the false alarm;

[0089] The robot plans its path based on the three-dimensional spatial benchmark model of the tower and uses multi-sensor fusion positioning (laser SLAM + UWB base station) to avoid ground obstacles (such as gravel and water) in real time. After arriving at the target point, the robot arm uses visual servoing to align the bolts (with the assistance of AR markers). The robot arm is equipped with a digital torque wrench (range 0-200Nm) and Calculate the required torque (e.g. =1.3° corresponds to torque T=120Nm), perform clockwise tightening, monitor the torque value in real time until it reaches the standard (such as stop when T=125Nm), and after the operation is completed, trigger the passive sensor to immediately feedback the updated Value (such as =0.2°), if the standard is not met, secondary tightening will be initiated. The robot controls the micro endoscope to extend into the gap of the bolt connection, take pictures of the internal status (such as whether the gasket is deformed), send the image back to the edge node for secondary confirmation, and record the torque value, Correction values ​​and endoscopic images form operation data;

[0090] According to the recorded operation data (such as =0.2°), update the mechanical parameters (such as preload and stress distribution) of the corresponding bolts in the three-dimensional spatial benchmark model of the tower, and recalculate the stress influence range of adjacent components based on the lightweight model of finite element analysis (for example, after bolt A is repaired, the stress of adjacent bolt B decreases by 20%). The corrected parameters are input into the AI ​​model to regenerate the dynamic risk heat map, downgrading the original high-risk area to normal (green), and synchronously updating the risk level of the associated area (for example, the adjacent bolt changes from yellow low risk to green normal).

[0091] It should be noted that the ad hoc network broadcasts the event coordinates, the drone approaches and shoots according to the B-spline curve (30x zoom), and the robot performs torque control based on visual servoing; according to the maintenance The mechanical parameters of the three-dimensional space benchmark model of the tower are updated, and the stress influence range is recalculated by finite element analysis; after the robot is tightened, the gasket status is photographed through the endoscope, and the passive sensor is combined with real-time feedback The repair verification time is shortened from 2 hours to 5 minutes. After the bolts are repaired, the risk level of the area where the stress of the adjacent components drops by 20% is automatically downgraded, reducing redundant inspections by 30%.

[0092] This embodiment also provides a computer device, which is suitable for the method of linking multiple devices at the edge of a transmission tower based on a self-organizing network and two-dimensional and three-dimensional fusion, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the method of linking multiple devices at the edge of a transmission tower based on a self-organizing network and two-dimensional and three-dimensional fusion as proposed in the above embodiment.

[0093] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0094] This embodiment also provides a storage medium having a computer program stored thereon. When the program is executed by a processor, it implements the method for realizing multi-device linkage at the edge of a transmission tower based on self-organizing networking and two-dimensional and three-dimensional fusion as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0095] In summary, the present invention: by constructing a dynamic monitoring system for transmission towers that couples mechanical, thermal, and geometric fields, first, a passive sensor based on environmental backscatter analyzes the displacement of the cantilever beam through phase modulation, and combined with the coordinate mapping of the three-dimensional spatial reference model, it will reduce the error in detecting the loosening angle of the bolts, and accurately locate microscopic defects through spatial positioning. Secondly, relying on edge computing and self-organizing network communication, the AI ​​model integrates multi-dimensional data to generate a dynamic risk heat map, and triggers drones to approach for photography and robots for precise maintenance, forming a perception-decision-execution closed loop to improve response efficiency. In addition, passive sensors eliminate the need for battery maintenance through radio frequency power supply; the self-organizing network dynamic routing protocol maintains a high connectivity rate in the event of a single node failure, which is significantly better than a fixed topology solution. It provides a monitoring paradigm for smart grids that combines high-precision positioning, real-time response, and extremely simple operation and maintenance, promoting the evolution of transmission facility management towards intelligence and unmanned operation.

[0096] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A multi-device linkage method at the edge of a transmission tower based on ad hoc networking and two- and three-dimensional fusion, characterized by: include, Deploy multiple devices at the edge of transmission towers, embed passive sensors based on environmental backscatter, and use drone-mounted lidar to generate a three-dimensional spatial benchmark model of the towers. Based on the three-dimensional spatial benchmark model of the tower, the drone's autonomous inspection route is planned, point cloud data, infrared thermal imaging, and backscatter data are collected simultaneously, and the deformation parameters of the passive sensor are extracted by analyzing the channel state information to construct a multidimensional data set; Fusing the multidimensional datasets at the edge node, generating a dynamic risk heat map and annotating the risk levels; When a high risk level is detected, the equipment deployment strategy is triggered, the operation data is recorded, and the three-dimensional spatial reference model of the tower is updated; The multi-device deployment includes deploying edge computing nodes at the base of the tower, integrating ZigBee / Wi-Fi dual-mode ad hoc network communication tools, and physically connecting to the embedded airport management system; The tower's 3D spatial benchmark model is obtained by flying a drone equipped with a Velodyne VLP-16 lidar along a calibrated path to collect tower point cloud data. The point cloud data is then aligned with the BIM model coordinate system using the ICP algorithm to generate a fused coordinate system. The coordinates of the passive sensor installation locations are then mapped to the BIM model coordinate system to obtain the tower's 3D spatial benchmark model. The generation of a dynamic risk heat map and the marking of risk levels utilize an AI model to associate the three-dimensional spatial benchmark model of the tower with the fused multidimensional dataset, calculate the risk value, classify the risk levels, and generate a dynamic risk heat map; The equipment deployment strategy includes deploying drones to approach the target location to collect high-definition images, dispatching inspection robots to perform on-site maintenance, and correcting the three-dimensional spatial reference model of the tower.

2. The method for multi-device linkage at the edge of a transmission tower based on ad hoc networking and two- and three-dimensional fusion according to claim 1, characterized in that: The planned autonomous inspection route of the UAV is based on the three-dimensional spatial benchmark model of the tower. By setting the horizontal safety distance and the vertical layered scanning path, the UAV's circling route is dynamically generated and a real-time obstacle avoidance strategy is integrated.

3. The method for multi-device linkage at the edge of a transmission tower based on ad hoc networking and two- and three-dimensional fusion according to claim 1, characterized in that: The channel state information includes signal strength, phase offset, propagation time difference and subcarrier index.

4. The method for multi-device linkage at the edge of a transmission tower based on ad hoc networking and two-dimensional and three-dimensional fusion according to claim 1, characterized in that: The passive sensor deformation parameters include cantilever beam displacement and bolt loosening angle.

5. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for linking multiple devices at the edge of a transmission tower based on self-organizing networking and two- and three-dimensional fusion are implemented as described in any one of claims 1 to 4.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for linking multiple devices at the edge of a transmission tower based on self-organizing networking and two- and three-dimensional fusion as described in any one of claims 1 to 4 are implemented.

Images