A quality detection system and method for nuclear engineering complex construction site

Abstract

The application provides a kind of nuclear engineering complex construction air-ground integrated quality detection system and method, including air detection module, ground detection module, data processing center module, intelligent decision module and communication module.Air detection module is collected by unmanned aerial vehicle carrying multispectral camera, laser radar and other sensors image and data of construction area;Ground detection module uses high-precision positioning equipment, wireless sensor network and portable detection equipment for local fine detection;Data processing center module fuses air-ground data, three-dimensional modeling and feature extraction;Intelligent decision module analyzes data based on deep learning algorithm, identifies construction defects and evaluates quality risk.The application realizes flexible configuration and expansion of the system through modular design, combined with air-ground integrated detection technology, significantly improves the detection efficiency, accuracy and coverage rate of nuclear engineering construction quality, and provides strong guarantee for nuclear engineering safety.

Description

Technical Field

[0001] This invention relates to the field of safety technology for nuclear facility construction and quality inspection, and in particular to an integrated air-ground quality inspection system and method for complex nuclear engineering construction. Background Technology

[0002] Nuclear energy, as a crucial component of clean energy, prioritizes safety above all else. Nuclear engineering, especially the construction of nuclear power plants, is a massive systems engineering project characterized by extreme complexity, high specialization, long construction periods, and stringent safety requirements. Its construction quality directly impacts the safe, stable, and economical operation of the power plant for decades to come, and even public and environmental safety. Nuclear engineering construction encompasses thousands of critical processes, including civil engineering (such as containment dome pouring), installation (such as reactor pressure vessel hoisting), piping welding, electrical installation, and radiation shielding layer construction.

[0003] Currently, quality inspection and monitoring during nuclear engineering construction mainly rely on a combination of traditional and modern methods, but significant shortcomings still exist: Routine inspections, primarily conducted manually, involve quality inspectors using simple tools such as tape measures, levels, and probes, combined with visual inspection, to sample and test key construction points. This method is highly subjective, inefficient, and has limited coverage. Furthermore, it is difficult to implement inspections in high-altitude, concealed, or high-risk areas (such as high-radiation zones and confined spaces), resulting in significant safety blind spots. Inspection results heavily rely on the experience of engineers, and data records are mostly paper-based or in simple spreadsheets, making systematic analysis and traceability difficult.

[0004] Application of fixed-point automated monitoring technology: In recent years, total stations, GPS monitoring stations, and fixed strain gauges have been introduced into some major projects to continuously monitor settlement, displacement, and stress. However, these devices are usually located at fixed points, have limited coverage, high deployment costs, and poor flexibility, making it difficult to achieve dynamic and rapid response monitoring across the entire construction area. They focus on macroscopic deformation and are insufficient in detecting localized, microscopic quality problems such as concrete cracks, welding defects, and surface damage.

[0005] Preliminary attempts with emerging technologies (such as drones): The widespread adoption of consumer and industrial drones has led to their application in engineering surveying. In nuclear engineering, there have been attempts to use drones for progress photography and topographic surveying. However, existing applications are mostly single-function and fragmented, presenting the following problems: Limited functionality: Often equipped only with a visible light camera, lacking the ability to detect multiple dimensions of quality such as thermal properties, microscopic deformation, and material characteristics.

[0006] Data isolation: Data collected by drones is independent of ground inspection data and design models (BIM), forming "data silos" and lacking effective integration and linkage analysis.

[0007] Low level of intelligence: Data analysis still relies heavily on manual interpretation and fails to fully utilize artificial intelligence technology to achieve automatic identification and assessment of defects, and the value of massive amounts of data has not been deeply explored.

[0008] Insufficient adaptability: The system has not been designed and reinforced at the system level to address the unique environment of nuclear engineering, such as high radiation, strong electromagnetic interference, complex structures, and ultra-high safety standards.

[0009] Systemic deficiency: The system failed to build a complete closed-loop system from data acquisition, transmission, processing, analysis to decision support. The links were disconnected, making it impossible to achieve real-time, accurate, and forward-looking quality control.

[0010] In addition, nuclear engineering construction is progressive, and the focus of quality control varies at different stages (e.g., the focus is on earthwork and foundation in the early stage, on structure and installation in the middle stage, and on commissioning and sealing in the later stage). Traditional systems lack a modular and flexible architecture, making it difficult to flexibly adjust testing strategies and resource allocation according to project progress.

[0011] Therefore, there is an urgent need in this field for a next-generation quality inspection system and method that can overcome the above-mentioned defects, achieve full coverage, high precision, intelligence, adaptability to the special requirements of nuclear engineering, and dynamic evolution with the project stage. Summary of the Invention

[0012] The purpose of this invention is to provide an integrated air-ground quality inspection system and method for complex nuclear engineering construction, aiming to solve the problems of numerous blind spots in the existing technology, difficulty in data fusion, low level of intelligence, and insufficient system flexibility. It realizes full-process automation and intelligence from planning, data collection, fusion, analysis to decision feedback, significantly improving the efficiency, accuracy and reliability of nuclear engineering construction quality control.

[0013] According to one objective of the present invention, an integrated air-ground quality inspection system for complex nuclear engineering construction is provided, comprising an airborne inspection module, a ground inspection module, a data processing center module, an intelligent decision-making module, and a communication module. The airborne inspection module is used to collect macroscopic geometric, spectral, and thermodynamic data of the construction area. The ground inspection module is used to collect high-precision local deformation, material properties, and environmental parameter data of the construction area. The data processing center module is communicatively coupled to the airborne and ground inspection modules, serving as a data fusion and intelligent computing hub, used to process multi-source heterogeneous data and generate a digital quality status model dynamically associated with a building information model. The intelligent decision-making module is computationally coupled to the data processing center module, serving as an analysis and decision-making brain, used to intelligently analyze the digital quality status model and generate quality assessment reports and disposal recommendations. The communication module is network-coupled to the airborne inspection module, the ground inspection module, the data processing center module, and the intelligent decision-making module, serving as an information transmission link, used to establish a highly reliable, low-latency, and interference-resistant data link.

[0014] Furthermore, the aerial detection module includes a drone platform with a radiation shielding layer and a cluster of mission payloads with standardized quick-release interfaces. The drone platform is an industrial-grade multi-rotor model with high stability, dust and water resistance, and electromagnetic interference resistance. The mission payload cluster includes at least a multispectral camera, lidar, and infrared thermal imager, and also integrates one or more of a high-precision positioning system, a gamma-ray spectrometer, and a high-precision gas sensor.

[0015] Furthermore, the ground detection module includes a high-precision ground control network, a wireless sensor network, a portable intelligent detection terminal, and a ground mobile robot; the high-precision ground control network consists of multiple high-precision positioning base stations, the wireless sensor network includes at least one of strain sensors, crack gauges, temperature and humidity sensors, and vibration sensors, and the nodes of the wireless sensor network adopt energy harvesting technology and have a sleep and wake-up mechanism; the ground mobile robot is a wheeled or tracked platform equipped with a local fine scanning LiDAR, a robotic arm, and a high-definition camera.

[0016] Furthermore, the data processing center module includes a data receiving and preprocessing unit, an air-to-ground data fusion unit, a real-scene 3D reconstruction unit, a quality feature extraction unit, and a BIM comparison and analysis unit, and also has a digital delivery interface; the data receiving and preprocessing unit uses Kalman filtering or wavelet transform algorithms for data noise reduction, the air-to-ground data fusion unit uses point cloud registration algorithms and image feature matching algorithms to achieve data fusion, and the BIM comparison and analysis unit is used to overlay and compare the current construction status 3D model with the preset design BIM model and identify and quantify the deviation.

[0017] Furthermore, the intelligent decision-making module includes a defect knowledge base, a machine learning analysis engine, a risk quantification and assessment unit, an automatic report generation unit, and a decision support dashboard; the machine learning analysis engine is loaded with at least one of a convolutional neural network model, a point cloud deep learning model, and a time series data analysis model; the risk quantification and assessment unit uses fuzzy comprehensive evaluation or risk matrix method to assign severity and urgency levels to defects; the automatic report generation unit is used to generate structured and visualized quality inspection reports, and the decision support dashboard is a graphical interface to dynamically display the construction quality status.

[0018] Furthermore, the communication module adopts a heterogeneous converged network architecture, including a wireless broadband subnet, an optical fiber backbone subnet, and a disaster recovery backup link, and also integrates a data security unit; the wireless broadband subnet deploys 5G / 6G millimeter wave small base stations, the disaster recovery backup link adopts satellite communication or a long-distance wireless mesh network, and the data security unit adopts national cryptographic algorithms to achieve end-to-end data encryption and performs identity authentication and access control for access devices.

[0019] A method for integrated air-ground quality inspection of complex nuclear engineering construction, utilizing the aforementioned integrated air-ground quality inspection system for complex nuclear engineering construction, includes the following steps executed in an orderly manner: S1: System initialization and testing scheme planning stage. Based on the construction stage and acceptance specifications, the target area, accuracy index, sensor combination and flight / movement path are set, and a testing task list is generated and distributed. S2: In the air-ground collaborative data acquisition phase, the air detection module first performs a wide-area air scan and transmits the data back, and then the ground detection module performs fine-grained ground-based supplementary measurements on suspected defect areas or key nodes. S3: Multi-source data fusion and 3D modeling stage, spatiotemporal registration and fusion of air and ground data, construction of real-scene 3D model and extraction of quality feature parameters; S4: Intelligent analysis and risk assessment stage, using deep learning models to identify and classify construction defects, and combining the defect knowledge base to conduct risk quantification assessment of defects; S5: Results visualization and decision feedback stage. A comprehensive test report is generated and presented through a visual interface. At the same time, key conclusions and early warning information are pushed to relevant engineering management platforms to form a closed-loop management system from testing to handling.

[0020] Furthermore, the air-to-ground data fusion in the S3 stage adopts a two-stage registration strategy, namely, a combination of feature-guided coarse registration and iterative nearest-point algorithm-based fine registration. In addition, spatial constraints based on the BIM model are used during the fusion process to improve the fusion accuracy and reliability.

[0021] Furthermore, the intelligent analysis process in the S4 stage includes time-series data trend prediction, using historical monitoring data to train a prediction model to predict future trends such as crack propagation and settlement changes, and triggering an early warning when the predicted value exceeds the safety threshold; the detection report generated in the S5 stage supports electronic signatures and archiving that comply with nuclear safety regulations, and can generate summary or detailed reports suitable for different levels of management personnel as needed.

[0022] Furthermore, it is applied to the construction quality inspection of nuclear power plants, nuclear fuel cycle facilities, and nuclear technology utilization devices.

[0023] This invention addresses the industry pain points of quality inspection in nuclear engineering construction. Through an integrated air-ground collaborative inspection model, it upgrades inspection from point sampling to full-area coverage, effectively overcoming inspection blind spots in high-risk areas such as high altitudes, high radiation zones, and enclosed spaces, significantly improving inspection efficiency and coverage. The precise fusion of multi-source heterogeneous data combined with deep learning algorithms eliminates the uncertainty of single data sources, enabling automatic identification, classification, and risk trend prediction of construction defects, shifting quality control from passive rectification to proactive prevention. The modular and scalable architecture flexibly adapts to the needs of each construction stage of nuclear engineering. Simultaneously, the constructed digital quality model creates a digital twin foundation for nuclear engineering, forming a closed-loop management system of inspection, assessment, and handling. This provides core data support for the entire lifecycle operation of nuclear engineering, comprehensively strengthening the quality and safety defenses of nuclear engineering construction. Attached Figure Description

[0024] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0025] Figure 1 This is a system module architecture diagram of an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the composition of the airborne detection module according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the composition of the ground detection module according to an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the composition of the data processing center module according to an embodiment of the present invention; Figure 5 This is a schematic diagram illustrating the composition of the intelligent decision-making module in an embodiment of the present invention; Figure 6 This is a schematic diagram of the composition of the communication module in an embodiment of the present invention; Figure 7This is a complete flowchart of the AI ​​image recognition quality detection method according to an embodiment of the present invention.

[0026] Figure 8 This is a roadmap for SLAM-based autonomous flight control technology for unmanned aerial vehicles (UAVs) according to embodiments of the present invention.

[0027] In the diagram, 1. Aerial detection module; 11. Unmanned aerial vehicle platform; 12. Multispectral camera; 13. LiDAR; 14. Infrared thermal imager; 15. Radiation shielding layer; 16. Mission payload cluster; 17. High-precision positioning system; 18. Gamma-ray spectrometer; 19. High-precision gas sensor. 2. Ground detection module; 21. High-precision positioning base station; 22. Wireless sensor network; 221. Strain sensor; 222. Crack gauge; 223. Temperature and humidity sensor; 224. Temperature and humidity sensor; 23. Portable intelligent detection terminal; 24. High-precision ground control network; 25. High-precision ground control network; 3. Data Processing Center Module; 31. Air-Ground Data Fusion Unit; 32. Real-Scene 3D Reconstruction Unit; 33. Quality Feature Extraction Unit; 34. BIM Comparison and Analysis Unit; 35. Data Receiving and Preprocessing Unit; 36. Digital Delivery Interface; 4. Intelligent Decision-Making Module; 41. Defect Knowledge Base; 42. Machine Learning Analysis Engine; 421. Convolutional Neural Network Model; 422. Point Cloud Deep Learning Model; 423. Time Series Data Analysis Model; 43. Risk Quantification and Assessment Unit; 44. Automatic Report Generation Unit; 45. Decision Support Dashboard; 5. Communication module; 51. Wireless broadband subnet; 52. Fiber optic backbone subnet; 53. Disaster recovery backup link; 54. Disaster recovery backup link. Detailed Implementation

[0028] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention.

[0030] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified. Furthermore, the terms "installed," "connected," and "linked" should be interpreted broadly; for example, they may refer to a fixed connection, a detachable connection, or an integral connection; they may refer to a mechanical connection or an electrical connection; they may refer to a direct connection or an indirect connection through an intermediate medium; and they may refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0031] Example 1 like Figures 1-8 As shown, an integrated air-ground quality inspection system for complex nuclear engineering construction adopts a distributed, pluggable modular architecture to achieve comprehensive quality monitoring and evaluation of multiple stages and disciplines such as civil engineering, installation, welding, and shielding layer construction during the construction of nuclear power plants. This invention belongs to the interdisciplinary field of nuclear engineering technology, construction engineering quality management and advanced non-destructive testing technology. Specifically, it relates to a quality inspection system and method that is particularly suitable for nuclear power plants, nuclear fuel cycle facilities, nuclear technology utilization devices and other nuclear engineering projects in the complex construction stage. It integrates modular design concept, comprehensively utilizes aerial unmanned aerial vehicle platforms and ground sensor networks for integrated collaborative operation, and integrates big data processing and artificial intelligence decision support.

[0032] This invention relates to an integrated air-ground quality inspection system for complex nuclear engineering construction. It adopts a top-level architecture design of "sensing layer - transmission layer - platform layer - application layer," and decouples the functions of each layer through a modular approach. This allows each module to have a standard interface and can be flexibly combined and expanded in a "building block" manner according to the scale, stage, and specific needs of a particular nuclear engineering project. The core of the system includes five major functional modules: Aerial detection module 1, as a large-scale, fast-response detection unit, is configured to perform aerial scanning of the construction area according to a predetermined flight path or real-time command, and to collect macroscopic geometric, spectral and thermodynamic data in a non-contact manner. Ground detection module 2, as a refined, fixed-point continuous detection unit, is configured to be deployed in the area of ​​interest or key quality control points specified in the construction specifications as defined by aerial detection module 1, and collect high-precision data on local deformation, material properties and environmental parameters. Ground detection module 2 is the "ground tentacles" of the system, responsible for refined, close-range and continuous monitoring of suspected areas or key points specified in the design as discovered by the aerial module.

[0033] Data processing center module 3, as the data fusion and intelligent computing hub of the system, is communication-coupled to air detection module 1 and ground detection module 2. It is configured to perform spatiotemporal registration, filtering and denoising, three-dimensional reconstruction and feature extraction on the received multi-source heterogeneous data, and generate a digital quality status model that is dynamically associated with the building information model. The intelligent decision-making module 4, serving as the system's analysis and decision-making brain, is coupled to the data processing center module 3 and is configured to perform intelligent analysis on the digital quality status model based on a machine learning model, automatically identify, classify, and locate construction defects, assess their safety risk level, and generate a structured quality assessment report and disposal recommendations. Communication module 5, as the information transmission link of the system, is network-coupled to air detection module 1, ground detection module 2, data processing center module 3 and intelligent decision-making module 4. It is configured to establish a highly reliable, low-latency and anti-interference data link between each module to ensure bidirectional real-time transmission of commands and data and collaborative operation of the system.

[0034] Specifically, the airborne detection module 1 further includes: The UAV platform 11, employing a multi-rotor design, possesses high stability, dust and water resistance, and a certain degree of electromagnetic interference resistance. Its fuselage and key electronic components are equipped with a radiation shielding layer 15 to adapt to the unique environment of nuclear engineering construction. The UAV platform 11 utilizes an industrial-grade multi-rotor UAV (suitable for precise operations in complex spaces). The platform features a high-precision flight control system, strong wind resistance, and IP67 dust and water resistance. Key innovations include: radiation-hardened design for the fuselage, flight control system, battery compartment, and mission payload interfaces, specifically for nuclear environments. Lead-based composite materials or tungsten alloys are used for partial shielding, and electronic components are selected for radiation hardening to ensure short-term normal operation without interference in low-intensity radiation areas.

[0035] The mission payload cluster 16 employs a standardized quick-release interface and can be flexibly configured according to mission requirements. The mission payload cluster 16 is detachably mounted beneath the unmanned aerial vehicle platform 11 and includes at least: The multispectral camera 12 is configured to acquire high-resolution images in the visible light and specific near-infrared bands for identifying surface contamination, humidity distribution, and material types. In addition to acquiring high-definition RGB images, the multispectral camera 12 also captures information in specific near-infrared (NIR) and red edge bands for identifying vegetation stress (environmental assessment), uneven concrete moisture content, and boundaries between different materials.

[0036] The lidar 13 is configured to emit laser pulses and receive echoes to generate three-dimensional point cloud data of the construction area with centimeter-level accuracy, which is used to calculate earthwork volume, detect installation deviations of structural components and overall deformation. The lidar 13 emits laser beams with millions of points per second to generate high-density three-dimensional point clouds, which accurately reflect the geometry of the structure and are used to detect flatness, verticality, volume calculation and collision detection with other components.

[0037] Infrared thermal imager 14 is configured to capture infrared radiation from the surface of an object and generate a thermal image. It is used to detect anomalies such as heat of hydration in concrete, defects in pipe insulation layers, and overheating of electrical joints. Infrared thermal imager 14 senses minute temperature differences on the surface of an object and generates thermal radiation images. In nuclear engineering construction, it is used to detect missing pipe insulation layers, overheating of electrical connection points, uneven distribution of heat of hydration after concrete pouring (to prevent temperature cracks), and the working status of floor heating systems.

[0038] The mission payload cluster 16 also includes: High-precision positioning system 17: Integrating differential GNSS (such as RTK), high-precision IMU and high-precision SLAM, it provides centimeter-level outdoor absolute geographic coordinates, indoor relative coordinates and precise attitude angle information for UAVs and all sensor data. This is the prerequisite for realizing high-precision fusion of indoor and outdoor integrated air-space-ground data.

[0039] The Gamma Ray Spectrometer 18 is configured to detect the environmental radiation level around the construction area during specific inspection tasks, and overlay it with geographic information to draw a radiation dose distribution map for evaluating the shielding construction effect and regional radiation safety. The high-precision gas sensor 19 is configured to detect the concentration of specific volatile organic compounds or harmful gases in the air to help determine the quality of sealing work or the risk of leakage.

[0040] Specifically, the ground detection module 2 further includes: The high-precision ground control network 25 consists of multiple high-precision positioning base stations 21 distributed around the construction area. The base stations use total stations or are equipped with continuously operating reference station technology and are configured to transmit differential correction signals to provide sub-centimeter-level positioning enhancement services for the aerial detection module 1 and the mobile detection equipment. The wireless sensor network 22 consists of a large number of low-cost, low-power miniature sensor nodes. These nodes form a multi-hop network through self-organization and are pre-encapsulated in protective housings. They are deployed in locations that are difficult to observe directly, such as inside critical concrete structures, around the foundations of important equipment, and near pipe welds. Sensor node types include, but are not limited to: Strain sensor 221 is used to monitor changes in structural stress; Crack gauge 222, used to monitor crack width development; Temperature and humidity sensor 223 is used to monitor the concrete curing environment; Vibration sensor 224 is used to monitor the stability of the equipment installation foundation; The portable intelligent inspection terminal 23 is a handheld device that integrates a high-resolution camera, an ultrasonic flaw detector, a laser rangefinder, and a touch screen. It is configured to allow quality inspectors to perform close-range verification inspections of specific parts and can interact with the data processing center module 3 in real time through the communication module 5 to upload inspection data and receive guidance information. The ground mobile robot 24 is a wheeled or tracked robot platform equipped with a local fine scanning LiDAR, a robotic arm and a high-definition camera. It is configured to enter areas that are inaccessible to personnel or pose certain risks (such as inside pipelines or narrow spaces) to perform automated close-range inspection tasks under preset paths or remote control.

[0041] Among them, the nodes of the wireless sensor network 22 adopt energy harvesting technology to obtain some of the energy required for operation from environmental vibration, temperature difference or solar energy, and have a sleep and wake-up mechanism to significantly extend the overall life of the network and adapt to the characteristics of long-term construction of nuclear engineering.

[0042] Specifically, the data processing center module 3 further includes: The data receiving and preprocessing unit 35 is configured to parse, decode, timestamp align and unit standardize raw data from different sources, in different formats and at different times, and to perform noise reduction processing on the data using Kalman filtering or wavelet transform algorithms. The air-to-ground data fusion unit 31 is configured to apply point cloud registration algorithms (such as iterative nearest point algorithms) and image feature matching algorithms to accurately fuse the wide-area point cloud and image data acquired by the air detection module 1 with the fine point cloud and detection data acquired by the ground detection module 2 in a unified coordinate system, thereby eliminating gaps and contradictions between the data. The real-scene 3D reconstruction unit 32 is configured to generate a high-precision, highly realistic 3D model of the current construction status based on the fused point cloud data and texture images through a surface reconstruction algorithm. The quality feature extraction unit 33 is configured to automatically extract quality-related feature parameters from the 3D model and sensor time series data, including but not limited to: flatness, perpendicularity, smoothness, elevation deviation, structural spacing, crack length and width, temperature gradient, and strain history curve. BIM comparison and analysis unit 34 is configured to overlay and compare the current construction status 3D model with the preset design stage BIM model. Through the difference analysis algorithm, it automatically identifies and quantifies the deviation between construction and design, and associates the deviation results with the attribute information of BIM components.

[0043] Specifically, the data processing center module 3 is also equipped with a digital delivery interface 36, which is configured to encapsulate the three-dimensional model of the current construction status, quality feature data and BIM comparison results in accordance with the nuclear engineering industry standard format, and support seamless digital delivery to the owner or operation and maintenance system.

[0044] Specifically, the intelligent decision-making module 4 further includes: The Defect Knowledge Base 41 stores various typical defect cases, characteristic data, cause analysis and treatment solutions that have occurred in nuclear engineering construction throughout history, and supports continuous learning and updating based on newly discovered defects. The machine learning analysis engine 42, as a core analysis component, is loaded with a deep learning model trained on a large amount of labeled data. The model includes at least: The convolutional neural network model 421 is specifically designed for pixel-level segmentation of optical images and infrared thermal images to automatically identify surface defects such as cracks, peeling, honeycomb pitting, and corrosion. The point cloud deep learning model 422 is specifically designed to process 3D point cloud data and directly identify geometric spatial defects such as structural deformation, component misalignment, and installation deviation. The time-series data analysis model 423, based on a long short-term memory network or a time-series convolutional network, is used to analyze the continuous monitoring data transmitted back by the wireless sensor network 22, predict the crack development trend and structural stability changes, and realize early warning. The risk quantification and assessment unit 43 is configured to assign a severity level and an urgency level to each defect based on the defect type, size, location and development trend identified by the machine learning analysis engine 42, combined with historical experience in the defect knowledge base 41, using the fuzzy comprehensive evaluation method or the risk matrix method. The report automatic generation unit 44 is configured to automatically generate a structured and visualized quality inspection report based on the risk assessment results. The report content includes a defect distribution map, a detailed defect list (including location coordinates, pictures, descriptions, and risk levels), an overall quality score, maintenance priority suggestions, and potential risk warnings.

[0045] Specifically, the intelligent decision-making module 4 also integrates a decision support dashboard 45, which dynamically displays the overall quality status of the construction area, alarms for high-risk areas, the progress of testing tasks, and key performance indicators in a graphical interface, assisting project management in making macro-level decisions.

[0046] Specifically, communication module 5 adopts a heterogeneous converged network architecture, which includes: Wireless broadband subnet 51 deploys dedicated 5G / 6G millimeter wave micro base stations in the construction area to provide a high-speed, low-altitude coverage data transmission channel for the air detection module 1 and the mobile terminal; The fiber optic backbone subnet 52 connects various fixed ground detection equipment, high-precision positioning base stations 21 and the central server to ensure the stable transmission of massive amounts of data. Disaster recovery backup link 53 uses satellite communication or long-distance wireless mesh network as emergency communication support in extreme situations.

[0047] Specifically, the communication module 5 also integrates a data security unit 54, which is configured to perform end-to-end encryption of the transmitted data using national cryptographic algorithms, and to perform identity authentication and access management for devices accessing the system, so as to meet the stringent information security requirements of the nuclear engineering field.

[0048] A method for integrated air-ground quality inspection of complex nuclear engineering construction, utilizing the aforementioned integrated air-ground quality inspection system for complex nuclear engineering construction, includes the following sequentially executed steps: S1: System initialization and testing plan planning stage: Based on the current construction stage, acceptance specifications, and key focus areas, the target area, testing accuracy indicators, sensor combinations, and flight / movement paths for this test are set in module 3 of the data processing center; the system automatically generates a test task list and distributes it to the corresponding modules; S2: Air-Ground Collaborative Data Acquisition Phase: S2.1: Aerial wide-area scanning sub-step: Start the aerial detection module 1, perform automated flight scanning of the construction area according to the predetermined path, and simultaneously collect high-resolution optical images, laser point clouds and infrared thermal image data, and transmit them back to the data processing center module 3 in real time through the communication module 5; S2.2: Ground-based fine-tuning sub-step: Based on the preliminary results or preset plan of S2.1, the wireless sensor network (22) in the ground detection module 2 is activated for continuous monitoring, and a portable detection terminal (23) or a ground mobile robot (24) is dispatched to conduct close-range, high-precision supplementary detection of suspected defect areas and key nodes. S3: Multi-source data fusion and 3D modeling stage: In the data processing center module 3, all the air and ground data obtained in the S2 stage are spatiotemporally registered and fused to construct a real-world 3D model that reflects the actual construction status, and to accurately extract various geometric and physical quality characteristic parameters from it. S4: Intelligent Analysis and Risk Assessment Stage: In the intelligent decision-making module 4, the pre-trained deep learning model is used to analyze the 3D model and feature data generated in the S3 stage, automatically identifying and classifying construction defects; then, combined with the defect knowledge base, the identified defects are quantitatively assessed for risk, and their severity and priority of handling are determined. S5: Results Visualization and Decision Feedback Stage: Generate a comprehensive inspection report containing defect details, risk maps, quality scores, and maintenance recommendations, and present it to users through a visual interface; at the same time, key conclusions and early warning information are automatically pushed to the Engineering Management System (ERP), project collaboration platform, or mobile inspection App through an interface, forming a closed-loop management from inspection to handling.

[0049] Specifically, in the S3 phase, the air-to-ground data fusion adopts a two-stage strategy that combines feature-guided coarse registration with fine registration based on the iterative nearest point algorithm, and uses spatial constraints based on the BIM model to improve the fusion accuracy and reliability.

[0050] Specifically, in the S4 stage, the intelligent analysis process includes trend prediction of time series data, that is, using historical monitoring data to train a prediction model to predict crack expansion, settlement changes, etc. in a specific future time period, and triggering an early warning when the predicted value exceeds the safety threshold.

[0051] Specifically, reports generated in the S5 phase support electronic signatures and archiving that comply with nuclear safety regulations, and can generate summary or detailed versions of reports as needed for different levels of management personnel (such as site engineers, project managers, supervisors, and owner representatives).

[0052] Example 2 like Figures 1-8 As shown, this is a comprehensive quality inspection of the concrete pouring for the containment dome of the nuclear island. Background: The containment dome of the reactor building at an EPR nuclear power plant was successfully poured in a single, monolithic manner. This process is crucial, requiring comprehensive inspection of the concrete pouring quality to prevent temperature cracks and structural defects.

[0053] System Configuration: Aerial module: Radiation-hardened drone equipped with a high-resolution optical camera and an infrared thermal imager.

[0054] Ground module: Temperature sensors and strain gauge networks were pre-embedded in key parts of the dome; a ground 3D scanner was deployed to obtain fine point clouds of complex nodes at the bottom.

[0055] Center and Decision Module: Enables BIM comparison and thermal imaging analysis AI models.

[0056] Method execution: Planning: Within 24 hours after the pouring is completed, the system will plan an infrared patrol mission for drones, flying once every 6 hours for 3 days.

[0057] Data Acquisition: The drone flies autonomously to collect panoramic infrared thermal images and visible light images of the dome surface. Ground sensors continuously record the internal temperature and strain.

[0058] Fusion and Analysis: The data processing center maps infrared thermal images onto the dome BIM model to generate temperature field cloud maps. The AI ​​model analyzes the thermal images to identify "cold spots" (too low temperature, slow strength development) caused by inadequate insulation measures and "hot spots" (too large temperature difference, high risk of cracking) caused by the accumulation of cement hydration heat.

[0059] Decision: The system determines that a certain "hotspot" area has a "high" risk level and immediately generates an alert. The report recommends immediately strengthening surface insulation and moisture retention maintenance in the area. The alert information is sent directly to the on-site engineer's mobile app. After the engineer takes measures, the system schedules the next flight to verify the effectiveness of the measures.

[0060] Results: Successfully provided early warning before cracks appeared, avoiding potential quality defects and costly rework, and ensuring the long-term integrity of the containment vessel.

[0061] Example 3 In the construction of a nuclear power plant, the hoisting and placement of the reactor pressure vessel (RPV) is a critical milestone. The RPV is connected to the nuclear island concrete structure through its massive lower support ring, and the welding quality of the RPV directly affects the structural safety and long-term operational stability of the entire reactor system.

[0062] Challenges faced: Complex structure and limited space: The weld is located at the bottom of the RPV, surrounded by dense pipes and cable trays, and the space is small and dark, making traditional manual inspection difficult and posing safety risks.

[0063] Extremely stringent quality requirements: This type of safety-grade weld requires 100% non-destructive testing and strict inspection standards (such as meeting ASME III standards). Any minor defects (such as porosity, slag inclusions, or lack of fusion) may lead to non-acceptance.

[0064] Inspection efficiency and traceability: Due to the long length of welds, traditional ultrasonic testing (UT) or radiographic testing (RT) is time-consuming, and paper records are not easy to manage and trace, making it difficult to form digital assets.

[0065] System Configuration and Deployment The "integrated air-ground quality detection system" described in this invention was activated and configured in a targeted modular manner: Aerial detection module: Platform: A small, highly maneuverable hexacopter UAV with 360° obstacle avoidance and precise indoor positioning capabilities (such as UWB ultra-wideband positioning) is selected to adapt to the complex environment inside the nuclear island.

[0066] Payload: Equipped with a high-resolution optical zoom camera (for global observation and preliminary visual inspection) and a high-precision structured light scanner (for rapidly acquiring dense 3D point cloud data of the weld area with an accuracy of up to 0.1mm).

[0067] Ground detection module: Core equipment: Deploy one tracked explosion-proof mobile robot. This robot is equipped with: A six-degree-of-freedom robotic arm with an integrated phased array ultrasonic testing (PAUT) probe at its end effector.

[0068] Laser rangefinders are used to help robots accurately locate weld seams.

[0069] High-definition cameras are used for real-time monitoring of the detection process.

[0070] Positioning assistance: Several UWB positioning base stations are set up around the detection area to provide centimeter-level absolute position information for drones and ground robots.

[0071] Data Processing Center Module & Intelligent Decision-Making Module: A dedicated weld knowledge base has been established, storing a large number of ultrasonic and three-dimensional morphological features of compliant welds and typical defects (porosity, cracks, incomplete penetration, etc.).

[0072] Load the pre-trained deep learning model: PointCloud-Net model: Used to process 3D point clouds generated by structured light scanning and identify macroscopic geometric defects in welds, such as undercut, uneven reinforcement, and misalignment.

[0073] CNN-LSTM fusion model: Used to analyze time-series image data such as B-scan and C-scan transmitted from PAUT equipment, and intelligently identify and classify internal defects.

[0074] Implementation process (method execution) The entire implementation process follows the five-stage method described in the patent, forming a complete closed loop: Phase 1: Intelligent Planning and Task Allocation Engineers import the BIM model of the RPV support ring into the system's task planning interface.

[0075] The system automatically identifies weld seam markers in the model and, based on their geometric information, automatically plans the global scanning path for the UAV and the precise detection path for the robot.

[0076] The task assignment is wirelessly transmitted to the control terminals of the UAV and ground robot via the communication module.

[0077] Phase Two: Air-Ground Collaborative Data Acquisition Aerial wide-area scanning: The drone flies autonomously along a predetermined path, using a structured light scanner to perform a rapid, non-contact, global 3D scan of the support ring weld. The scan data is transmitted back to the data processing center in real time. The entire process takes only 15 minutes, far faster than manual scaffolding measurement.

[0078] Ground-level fine-grained inspection: The data processing center performs rapid preprocessing on the point cloud transmitted back by the UAV, generates a preliminary three-dimensional model of the weld, and automatically locates the sections that need to be inspected.

[0079] The system issues commands to the ground robot. The robot autonomously navigates to the designated weld start point, and the laser sensor on the robotic arm precisely locates and guides the PAUT probe to scan along the weld centerline at a constant speed and pressure.

[0080] The raw ultrasound data collected by the PAUT device is streamed back to the data processing center in real time via a high-speed Wi-Fi 6 network.

[0081] Phase 3: Multi-source data fusion and modeling The data processing center performs spatiotemporal registration and fusion of the high-precision three-dimensional weld topography point cloud acquired by the UAV and the PAUT internal defect data collected by the ground robot.

[0082] Generate a "digital twin of the weld": In this three-dimensional model, not only can the external geometry of the weld be seen, but also information such as the location, size, and burial depth of internal defects can be accurately marked on the corresponding three-dimensional spatial location through color mapping, three-dimensional visualization and other technologies.

[0083] Phase Four: AI-Driven Defect Identification and Risk Assessment Intelligent recognition: PointCloud-Net model analysis of point cloud reports: "Local excess height in weld section No. 3 exceeds the specification by 0.2mm."

[0084] The CNN-LSTM model analyzed the PAUT data and reported: "A cluster of pores with an equivalent diameter of 1.5 mm was found at the location (X1250 mm, Y350 mm) of the 'weld digital twin'."

[0085] risk assessment: The system retrieves the weld knowledge base and automatically evaluates the welds based on the defect type, size, and location (whether it is in a stress concentration area) according to relevant standards.

[0086] Output results: "Clustered pores" are judged as unacceptable defects according to the acceptance criteria (level: exceeding the standard); "Excessive height" is judged as a defect that needs to be ground (level: moderate).

[0087] The system automatically generates a risk map, highlighting "exceeding the standard" defects in red and "moderate" defects in yellow on the 3D model.

[0088] Phase 5: Decision Support and Closed-Loop Feedback Report Generation: The system automatically generates a detailed "Support Ring Weld Inspection Report", which includes a list of defects, accurate 3D location screenshots, risk level, and evaluation opinions based on standards.

[0089] Closed-loop management: The report is automatically pushed to the project department's "Non-conforming Product Handling Process" (NCR system) through the system interface.

[0090] The system automatically creates a "weld rework notice" for "exceeding standards" defects and assigns a responsible welding team.

[0091] After rework is completed, the welder reports the completion in the system. The system can automatically plan a review and inspection task, scanning only the reworked area to verify the repair effect, forming a quality control closed loop of "inspection-identification-treatment-verification".

[0092] like Figure 7 As shown in this embodiment, the software processing flow (actual detection execution flow), after the model is online, includes the following steps for real-time detection of the target (such as a weld): Camera image acquisition: The original image of the workpiece to be inspected (such as a weld) is acquired using an industrial camera.

[0093] Image cropping: The acquired image is cropped to focus on the target detection area and reduce interference from irrelevant information.

[0094] AI model for image segmentation and localization: The trained AI model is used to segment and localize the target in the cropped image, generating a result image similar to a black and white mask (the white area in the image is the target area recognized by the model).

[0095] Logical judgments are made based on the model's prediction results: based on the segmentation and localization results output by the model, logical processing such as quality assessment and defect analysis is performed.

[0096] Output results: The final output is a result image with recognition annotations (such as confidence level 0.93), which intuitively shows the target area and confidence level recognized by the model.

[0097] The example image visually illustrates this process: from the original weld image → a cropped image focusing on the weld → a black-and-white mask image generated by the model → the final result image with confidence level annotations.

[0098] like Figure 7 As shown in this embodiment, the AI ​​model training process (model building and optimization process) provides the model development steps that enable core AI capabilities for the software processing flow: Collect small batches of images: Collect small batches of labeled or unlabeled images as initial data for model training.

[0099] Model pre-training: Using small batches of collected data to pre-train the AI ​​model, allowing the model to initially learn the target features.

[0100] Deployment testing: Deploy the pre-trained model to a real-world environment for testing to verify its performance in real-world scenarios.

[0101] Model iteration and optimization: Based on feedback from online testing, adjust model parameters, supplement training data, and continuously improve model accuracy and robustness.

[0102] Model delivery: Once the model performance meets the requirements, the optimized model will be delivered for segmentation and localization tasks in the software processing flow.

[0103] The core of this process is to first build a reliable AI model through a training process, and then implement the model's capabilities into actual inspection tasks through a software processing process, thereby achieving automated and intelligent industrial vision inspection.

[0104] In summary, the integrated air-ground quality inspection system and method for complex nuclear engineering construction of this invention reduces the manual inspection process, which originally required several days, to just a few hours. It avoids personnel working in confined, high-risk spaces, and the inspection results are objectively determined by an AI model, eliminating human interference. It generates a "digital twin of the weld" containing complete geometric and quality information, which serves as an important digital asset transferred to the operator, providing accurate baseline data for in-service inspections throughout the power plant's lifespan. This achieves a shift from "post-event problem discovery" to "precise in-process location and immediate decision-making," greatly optimizing the construction process and quality control level.

[0105] Compared with the prior art, the present invention has the following significant advantages: Revolutionary improvement in detection range and efficiency: Through integrated air-ground collaboration, a leap from "point sampling" to "full coverage" has been achieved. Rapid scanning by drones greatly improves efficiency and overcomes the blind spots and risks of manual detection.

[0106] A qualitative leap in detection accuracy and reliability: Multi-source data fusion and cross-verification eliminate the uncertainty of a single data source. AI recognition avoids human subjective error and fatigue factors, resulting in more objective and traceable detection results.

[0107] Intelligent and forward-looking quality control: Based on deep learning and predictive models, the system can not only detect defects that have occurred, but also assess potential risks, realizing the transformation from "passive rectification" to "proactive prevention" and truly preventing problems before they occur.

[0108] The system's high adaptability and scalability: The modular design allows the system to adapt to the specific needs of different stages and sections of nuclear engineering, just like "building blocks," reducing the total cost of ownership and protecting the investment.

[0109] Building a digital twin foundation for nuclear engineering: The accurate, dynamic 3D reality model and rich attribute information generated by this system are the key foundation for building a digital twin of the entire life cycle of a nuclear power plant, providing invaluable data assets for subsequent operation, maintenance and decommissioning.

[0110] Comprehensive enhancement of nuclear safety culture: By using technical means to streamline, make transparent, and digitize quality control processes, the effectiveness of the quality assurance system has been greatly enhanced, building a solid technical defense line for nuclear safety.

[0111] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A combined air-ground quality inspection system for complex nuclear engineering construction, characterized in that, The system includes an aerial detection module, a ground detection module, a data processing center module, an intelligent decision-making module, and a communication module. The aerial detection module collects macroscopic geometric, spectral, and thermodynamic data of the construction area. The ground detection module collects high-precision local deformation, material properties, and environmental parameter data of the construction area. The data processing center module is communicatively coupled to the aerial and ground detection modules, serving as the data fusion and intelligent computing hub. It processes multi-source heterogeneous data and generates a digital quality status model dynamically associated with the building information model. The intelligent decision-making module is computationally coupled to the data processing center module, acting as the analysis and decision-making brain. It intelligently analyzes the digital quality status model and generates quality assessment reports and disposal recommendations. The communication module is network-coupled to the aerial, ground, data processing, and intelligent decision-making modules, serving as an information transmission link to establish a highly reliable, low-latency, and interference-resistant data link.

2. The integrated air-ground quality inspection system for complex nuclear engineering construction as described in claim 1, characterized in that, The aerial detection module includes a drone platform with a radiation shielding layer and a cluster of mission payloads with standardized quick-release interfaces. The drone platform is an industrial-grade multi-rotor model with high stability, dust and water resistance, and electromagnetic interference resistance. The mission payload cluster includes at least a multispectral camera, lidar, and infrared thermal imager, and also integrates one or more of the following: a high-precision positioning system, a gamma-ray spectrometer, and a high-precision gas sensor.

3. The integrated air-ground quality inspection system for complex nuclear engineering construction as described in claim 1, characterized in that, The ground detection module includes a high-precision ground control network, a wireless sensor network, a portable intelligent detection terminal, and a ground mobile robot. The high-precision ground control network consists of multiple high-precision positioning base stations. The wireless sensor network includes at least one of strain sensors, crack gauges, temperature and humidity sensors, and vibration sensors. The nodes of the wireless sensor network employ energy harvesting technology and have sleep and wake-up mechanisms. The ground mobile robot is a wheeled or tracked platform equipped with a local fine-scanning LiDAR, a robotic arm, and a high-definition camera.

4. The integrated air-ground quality inspection system for complex nuclear engineering construction according to claim 1, characterized in that, The data processing center module includes a data receiving and preprocessing unit, an air-to-ground data fusion unit, a real-scene 3D reconstruction unit, a quality feature extraction unit, and a BIM comparison and analysis unit, and also has a digital delivery interface. The data receiving and preprocessing unit uses Kalman filtering or wavelet transform algorithms for data noise reduction. The air-to-ground data fusion unit uses point cloud registration algorithms and image feature matching algorithms to achieve data fusion. The BIM comparison and analysis unit is used to overlay and compare the current 3D model of the construction status with the preset design BIM model and identify and quantify the deviation.

5. The integrated air-ground quality inspection system for complex nuclear engineering construction as described in claim 1, characterized in that, The intelligent decision-making module includes a defect knowledge base, a machine learning analysis engine, a risk quantification and assessment unit, an automatic report generation unit, and a decision support dashboard. The machine learning analysis engine is loaded with at least one of a convolutional neural network model, a point cloud deep learning model, and a time series data analysis model. The risk quantification and assessment unit uses fuzzy comprehensive evaluation or risk matrix method to assign severity and urgency levels to defects. The automatic report generation unit is used to generate structured and visualized quality inspection reports, and the decision support dashboard is a graphical interface to dynamically display the construction quality status.

6. The integrated air-ground quality inspection system for complex nuclear engineering construction according to claim 1, characterized in that, The communication module adopts a heterogeneous converged network architecture, including a wireless broadband subnet, an optical fiber backbone subnet, and a disaster recovery backup link, and also integrates a data security unit. The wireless broadband subnet deploys 5G / 6G millimeter wave small base stations, the disaster recovery backup link adopts satellite communication or a long-distance wireless mesh network, and the data security unit uses national cryptographic algorithms to achieve end-to-end data encryption and performs identity authentication and access control for access devices.

7. A method for integrated air-ground quality inspection of complex nuclear engineering construction, employing the integrated air-ground quality inspection system for complex nuclear engineering construction as described in any one of claims 1-6, characterized in that, This includes the following steps, executed in an orderly manner: S1: System initialization and testing scheme planning stage. Based on the construction stage and acceptance specifications, the target area, accuracy index, sensor combination and flight / movement path are set, and a testing task list is generated and distributed. S2: In the air-ground collaborative data acquisition phase, the air detection module first performs a wide-area air scan and transmits the data back, and then the ground detection module performs fine-grained ground-based supplementary measurements on suspected defect areas or key nodes. S3: Multi-source data fusion and 3D modeling stage, spatiotemporal registration and fusion of air and ground data, construction of real-scene 3D model and extraction of quality feature parameters; S4: Intelligent analysis and risk assessment stage, using deep learning models to identify and classify construction defects, and combining the defect knowledge base to conduct risk quantification assessment of defects; S5: Results visualization and decision feedback stage. A comprehensive test report is generated and presented through a visual interface. At the same time, key conclusions and early warning information are pushed to relevant engineering management platforms to form a closed-loop management system from testing to handling.

8. The integrated air-ground quality inspection method for complex nuclear engineering construction according to claim 7, characterized in that, The air-to-ground data fusion in the S3 phase adopts a two-stage registration strategy, which combines feature-guided coarse registration with iterative nearest point algorithm-based fine registration. Furthermore, spatial constraints based on the BIM model are used during the fusion process to improve the accuracy and reliability of the fusion.

9. The integrated air-ground quality inspection method for complex nuclear engineering construction according to claim 7, characterized in that, The intelligent analysis process in the S4 stage includes time-series data trend prediction, using historical monitoring data to train a prediction model, predicting future trends such as crack propagation and settlement changes, and triggering an early warning when the predicted value exceeds the safety threshold. The test report generated in the S5 stage supports electronic signatures and archiving that comply with nuclear safety regulations, and can generate summary or detailed versions of the report as needed for different levels of management personnel.

10. The integrated air-ground quality inspection method for complex nuclear engineering construction according to claim 7, characterized in that, It is used for construction quality inspection of nuclear power plants, nuclear fuel cycle facilities, and nuclear technology utilization devices.

Images