Search

CN-122024035-A - Nondestructive accurate detection and three-dimensional reconstruction system and method for hidden engineering of transformer substation

CN122024035ACN 122024035 ACN122024035 ACN 122024035ACN-122024035-A

Abstract

The application discloses a nondestructive accurate detection and three-dimensional reconstruction system and method for hidden engineering of a transformer substation, and relates to the technical field of engineering detection of the transformer substation, wherein the system comprises a multi-source full-period acquisition and transmission module, a multi-class device, a three-stage acquisition and encryption transmission module and a non-contact signal data; the method comprises the steps of carrying out targeted preprocessing on data by a multi-source heterogeneous fusion module to generate a standardized data set, positioning a classification component by a high-precision recognition metering module, calculating parameters and outputting a report, constructing and adjusting theoretical and actual models by a three-dimensional modeling calculation amount correction module, triggering early warning and archiving data by a model comparison early warning chain storage module comparison model, carrying out collaborative acquisition and encryption transmission by multiple devices, combining targeted preprocessing and weighted fusion to generate the standardized data set, realizing accurate positioning and metering of the component by means of a lightweight detection and classification architecture, and combining bidirectional closed loop correction by a BIM and multi-view reconstruction technology to ensure the accuracy of the model.

Inventors

  • ZHANG JINRUI
  • CHEN YONG
  • ZHANG YING
  • ZHANG BOLANG
  • FANG TAO

Assignees

  • 云南电网有限责任公司文山供电局

Dates

Publication Date
20260512
Application Date
20251211

Claims (10)

  1. 1. Nondestructive accurate detection and three-dimensional reconstruction system for hidden engineering of transformer substation is characterized in that the system comprises: the multi-source full-period acquisition and transmission module is used for deploying a zoom tripod head dome camera, a 5G double-shot mobile deployment and control ball, an electromagnetic induction detector and a geological radar, acquiring image data and non-contact signal data in three stages before construction, before concealment and after concealment, processing the data in a lossless compression mode, and transmitting the processed data through an AES-256 encryption protocol; The multi-source heterogeneous fusion module is used for performing Gaussian filtering, self-adaptive histogram equalization and Canny edge detection pretreatment on image data, performing wavelet transformation filtering and mean filtering pretreatment on non-contact signals, and correlating data before and after concealment of the same component through a weighted fusion algorithm to generate a standardized fusion data set; The high-precision recognition and measurement module is used for adopting a combined framework of a lightweight target detection model and a depth fine-granularity classification model to position and classify components, calculating the length, the size, the burial depth and the deformation of the components and generating a structural analysis report; The three-dimensional modeling calculation correction module is used for constructing a theoretical three-dimensional model through BIM modeling software, fusing the structural analysis report, generating an initial actual construction three-dimensional model in a mode of combining a multi-view image three-dimensional reconstruction technology with a non-contact signal characteristic mapping technology, and adjusting the model size and the engineering calculation value through a two-way closed loop correction algorithm; And the model comparison early warning chain storage module is used for comparing the theoretical model with the corrected actual model by adopting an ICP iterative closest point algorithm, triggering early warning according to a grading mechanism and archiving the whole-flow data through a Minio distributed storage system.
  2. 2. The nondestructive accurate detection and three-dimensional reconstruction system for the hidden engineering of the transformer substation is characterized in that in the multi-source full-period transmission module, 4 zoom tripod head ball machines are deployed at a construction site high point and fixed on an anti-shake support, the installation height is not lower than 15 meters, 220kV and 110kV field areas are covered, 6 5G double-shot mobile deployment control balls are deployed on wire coil pay-off areas and grounding body welding points, a magnetic attraction and tripod double-installation mode is adopted, an electromagnetic induction detector is deployed at 1 station every 5 meters along a component path, the turning positions are encrypted to 3 meters and are closely attached to the ground, a geological radar and the electromagnetic induction detector are alternately arranged every 10 meters along the same path, an antenna is perpendicular to the ground and has a ground clearance of not more than 5 centimeters, the construction area topography and the component initial state are collected before construction, 8 angle detail images of a component are collected before concealment, 4 illumination scenes are covered, component path coordinates, deformation and buried signals are collected, and coordinate calibration is completed by standard references before all equipment collection.
  3. 3. The transformer substation hidden engineering lossless accurate detection and three-dimensional reconstruction system according to claim 1 is characterized in that in the multi-source heterogeneous fusion module, gaussian filtering, adaptive histogram equalization and Canny edge detection preprocessing operations are carried out on image data, wherein the Gaussian filtering adopts a5×5 kernel size, image noise is filtered by calculating gray average values of target pixels and surrounding 24 adjacent pixels and weighting average, the adaptive histogram equalization divides an image into 8×8 pixel blocks, gray distribution in each block is counted, dynamic range of gray level in each block is adjusted, overall contrast of the image is balanced, local over-brightness or over-darkness is avoided, canny edge detection firstly calculates image gray gradient amplitude and direction, gradient pixels are selected by adopting a high threshold 150 and a low threshold 50, pixels with gradient amplitude larger than the high threshold are directly judged as edges, pixels which are between the high threshold and the low threshold and are communicated with the edge pixels are judged as edges, and finally the complete contour of a component is extracted.
  4. 4. The transformer substation hidden engineering nondestructive accurate detection and three-dimensional reconstruction system according to claim 1 is characterized in that in the multi-source heterogeneous fusion module, wavelet transformation filtering and mean filtering preprocessing are carried out on non-contact signals, specifically, for electromagnetic induction signals and geological radar signals, mean filtering processing is carried out firstly, sliding windows of 3×3 are adopted to slide point by point along a signal data sequence, arithmetic average values of all signal amplitudes in each window are calculated to replace original signal values of central positions of the windows, hierarchical filtering processing is carried out, signals are decomposed into multiple layers of components according to frequency characteristics, low-frequency components and effective intermediate-frequency components which reflect actual states of components are screened out, high-frequency clutter components generated due to electromagnetic interference and soil medium nonuniformity are removed, pure signals are reconstructed through component recombination, and random fluctuation and environmental interference components in the signals are removed.
  5. 5. The transformer substation hidden engineering nondestructive accurate detection and three-dimensional reconstruction system is characterized in that in the high-precision recognition metering module, a lightweight target detection model and a deep fine-grained classification model are adopted for combined architecture positioning and classifying components, the method comprises the specific steps of performing feature extraction through a 32-layer convolutional neural network after an input image is adjusted to 640 x 640 pixels, reserving deep features through residual connection in the first 16 layers to generate a multi-scale feature map, determining component positions through anchor frame matching and boundary frame regression based on feature map sliding window traversal images of the target detection model, outputting candidate areas containing component coordinates and confidence, performing feature enhancement on the candidate areas through a deep fine-grained classification model, focusing component key parts through a attention mechanism, performing similarity matching on the extracted feature vectors and a preset component feature library, distinguishing component types, materials and specification parameters, performing iterative training through a mode of random rotation, scaling, overturning and color disturbance by adopting a small-scale gradient descent method, setting 1000 rounds of training period, and performing verification performance of 20% recognition models to be lower than 95% of recognition results.
  6. 6. The nondestructive accurate detection and three-dimensional reconstruction system for hidden engineering of a transformer substation according to claim 1 is characterized in that the specific steps of constructing a theoretical three-dimensional model by BIM modeling software in the three-dimensional modeling calculation amount correction module are that design data are imported, space coordinate material parameter installation specifications of component design dimensions are analyzed, the model is built in layers according to component type region numbers, a cable layer labeling path trend section size, a grounding body layer is restored to grid distribution welding nodes, a cable trench layer is clear in length, width and depth and supporting structures, each component binds with a design dimension material density calculation amount unit installation standard, after the modeling is completed, collision detection component spacing is executed, standard check, installation angle connection mode burial depth requirements are developed, after verification is passed, FBX format model files are output, and a complete theoretical model system is formed by correlating engineering settlement calculation amount parameters.
  7. 7. The transformer substation hidden engineering nondestructive accurate detection and three-dimensional reconstruction system is characterized in that in the three-dimensional modeling calculation correction module, an initial actual construction three-dimensional model is generated by combining a multi-view image three-dimensional reconstruction technology with a non-contact signal feature mapping technology, and the method comprises the specific steps of firstly extracting feature points of a multi-view image, capturing key feature points of the surface of a component based on image texture and edge information, matching the feature points of different views, generating preliminary point cloud data of the component through space triangulation, denoising and registering the preliminary point cloud, removing redundant points and abnormal points, optimizing the density and uniformity of the point cloud, analyzing feature information in non-contact signals, mapping signal features to corresponding space positions, supplementing geometric information of hidden areas in the point cloud data, fusing component types, specifications and size parameters in a structural analysis report, triangulating the optimized point cloud data, constructing a component surface grid model, binding component material physical properties, reducing component connection relations and space postures, and finally generating the initial actual construction three-dimensional model consistent with the actual state of a construction site.
  8. 8. The nondestructive accurate detection and three-dimensional reconstruction system for hidden engineering of a transformer substation according to claim 1, wherein in the three-dimensional modeling calculation correction module, a mathematical expression of a bidirectional closed loop correction algorithm is as follows: ; Wherein, the method comprises the steps of, Is the first The three-dimensional model size after the correction of the components; Is the first The engineering calculation value after the correction of each component, Is the first The design theoretical model dimensions of the individual components, Is the first The theoretical engineering calculation value of each component, Is the first Initial model to theoretical dimensional deviations of individual components, Is the first The deviation rate of the manual calculation quantity and the theoretical calculation quantity of each component, Is a two-way correction coefficient and, Is the characteristic weight coefficient of the fusion, Is the size correction factor of the first member, Is the first A set of measured dimensional data for each component, Is the first The calculation amount correction coefficient of each component, Is the first A set of measured quantities of the individual components influence the factors, Is the first The length class of each component fuses the characteristic values, Is the first The volume class of the individual components fuses the eigenvalues.
  9. 9. The nondestructive accurate detection and three-dimensional reconstruction system for hidden engineering of a transformer substation according to claim 1, wherein in the model comparison early warning chain memory module, a mathematical expression of an ICP iterative nearest point algorithm is as follows: wherein min is the calculated minimum value, Is the first point cloud of the actual model A point of the light-emitting diode is located, Is the first point cloud of theoretical model A point of the light-emitting diode is located, Is a rotation matrix of the rotation, Is the translation vector of the motion vector, The method comprises the steps of squaring Euclidean distance, a grading early warning mechanism, a platform popup window and short message warning and correction within 48 hours, and a three-stage early warning triggering condition, wherein the first-stage early warning triggering condition is that spatial position deviation is greater than 5 cm or size deviation is greater than 3% or deformation exceeds an engineering allowable value, audible and visual warning is triggered, and a freeze acceptance checking flow is frozen, the second-stage early warning triggering condition is that spatial position deviation is 3-5 cm or size deviation is 2-3% or deformation is 80-100% of the engineering allowable value, the platform popup window and short message warning is triggered, and correction is required within 48 hours, and the third-stage early warning triggering condition is that spatial position deviation is 1-3 cm or size deviation is 1-2%, and only the platform early warning is triggered and deviation information is recorded.
  10. 10. A nondestructive accurate detection and three-dimensional reconstruction method for hidden engineering of a transformer substation, which is suitable for the nondestructive accurate detection and three-dimensional reconstruction system for hidden engineering of a transformer substation according to any one of claims 1 to 9, and is characterized by comprising the following specific steps: The multi-source full-period data acquisition and transmission comprises the steps of deploying a zoom tripod head dome camera, a 5G double-shot mobile deployment and control ball, an electromagnetic induction detector and a geological radar, acquiring image data and non-contact signal data in three stages of before construction, before concealment and after concealment, and transmitting through an AES-256 encryption protocol after lossless compression treatment; Performing Gaussian filtering, self-adaptive histogram equalization and Canny edge detection preprocessing on image data, performing wavelet transformation filtering and average filtering preprocessing on non-contact signals, and correlating data before concealment and data after concealment of the same component through a weighted fusion algorithm to generate a standardized fusion data set; The component high precision identification and measurement, namely adopting a lightweight target detection model and depth fine granularity classification model combined architecture to locate and classify the component, calculating the length, the size, the burial depth and the deformation of the component, and generating a structural analysis report; constructing a theoretical three-dimensional model through BIM modeling software, fusing a structural analysis report, generating an initial actual construction three-dimensional model in a mode of combining a multi-view image three-dimensional reconstruction technology with a non-contact signal characteristic mapping technology, and adjusting the model size and the engineering calculation value through a two-way closed-loop correction algorithm; and (3) model comparison, namely pre-warning and data archiving, wherein an ICP iterative closest point algorithm is adopted to compare a theoretical model with a corrected actual model, the pre-warning is triggered according to a grading mechanism, and full-flow data is archived through a Minio distributed storage system.

Description

Nondestructive accurate detection and three-dimensional reconstruction system and method for hidden engineering of transformer substation Technical Field The application relates to the technical field of transformer substation engineering detection, in particular to a nondestructive accurate detection and three-dimensional reconstruction system and method for a transformer substation concealed engineering. Background The hidden engineering of the transformer substation is a core component of the infrastructure construction of the transformer substation, relates to the arrangement and installation of various key components, the construction quality of the hidden engineering is directly related to the long-term safe and stable operation of the transformer substation, the detection, quality tracing and maintenance work of the states of the components are extremely difficult in the follow-up process after the construction is completed, the accurate acquisition, efficient integration and visual management of data in the whole period of the hidden engineering are realized along with the continuous improvement of the construction standard and the safe operation requirement of the transformer substation in the power industry, the important problem to be solved in the development of the industry is becoming the urgent problem, and the application of the multi-source sensing technology, the digital image processing technology, the artificial intelligent identification technology, the building information model technology and the like in the engineering field is gradually deepened at present, so that the possibility is provided for breaking through the technical bottleneck of the hidden engineering management, and the industry is increasingly pressing the requirements on nondestructive detection and the three-dimensional reconstruction solution which can integrate the advantages of multiple technologies and the whole flow of the covered engineering. The detection and management technology of the traditional transformer substation concealed engineering is obviously limited, the high quality requirement of the modern electric engineering is difficult to meet, in a data acquisition link, the traditional method depends on contact detection means, the operation flow is complex, unnecessary damage is possibly caused to constructed concealed components, data acquisition is often focused on a specific stage of the engineering, continuous data coverage of the whole period before construction, before concealment and after concealment cannot be realized, the whole process of the engineering is incomplete, the subsequent quality is lack of sufficient basis, in the aspects of data processing and component identification, the traditional technology is lack of an effective multi-source heterogeneous data fusion mechanism, image data and non-contact signal data are difficult to cooperatively play a role, component identification depends on manual experience judgment, the efficiency is low, identification deviation is easily generated due to subjective factors, the types, materials and specifications of components are difficult to accurately distinguish, the model construction and quality management link are not enough, the traditional three-dimensional modeling is mostly based on a theoretical model established on design drawing, a large gap exists between the actual construction working condition, the accuracy of engineering calculation is not sufficient, meanwhile, the model comparison and deviation early warning mechanism of a system is lacked, quality is difficult to be dispersed, the quality is difficult to realize, the complete and the quality is difficult to control and the subsequent quality is difficult to control and the quality is difficult to control and has a complete and is difficult to store. Disclosure of Invention The application aims to make up the defects of the prior art and provides a nondestructive accurate detection and three-dimensional reconstruction system and method for hidden engineering of a transformer substation, wherein the system acquires images and non-contact signals through a multi-source full-period acquisition and transmission module and generates a standardized data set through processing of a multi-source heterogeneous fusion module; the method comprises the steps of positioning a classification component by a high-precision recognition metering module and calculating parameters, constructing a theoretical and actual construction three-dimensional model by a three-dimensional modeling calculation correction module, adjusting by a two-way closed loop correction algorithm, comparing the theoretical model with the actual model by a model comparison early-warning chain storage module, triggering early warning according to a grading mechanism, and archiving data. The application provides a technical scheme for solving the technical problems, which comprises, on one hand, a nondestructive accurate de