Routing inspection data acquisition and contrastive analysis method based on augmented reality

Abstract

The invention belongs to the technical field of intelligent inspection, and discloses an inspection data acquisition and contrastive analysis method based on augmented reality, which comprises the following steps: carrying out multi-source data acquisition on an inspection site through augmented reality equipment to obtain multi-source inspection data and carrying out fusion processing to generate a three-dimensional inspection data atlas; performing dynamic behavior analysis and abnormal trend prediction based on the three-dimensional inspection data graph to obtain an inspection data feature model; based on the inspection data feature model, in combination with preset inspection standard data and historical inspection records, generating an inspection anomaly recognition and classification result; based on the inspection abnormity identification and classification result, combining with augmented reality equipment to carry out spatial positioning and visual labeling of an abnormal area, and obtaining inspection abnormity processing guidance; performing expert guidance and real-time feedback on the basis of the inspection abnormity processing guidance in combination with a remote cooperation function, and generating an inspection closed-loop data report; according to the invention, intelligent and closed-loop management of nuclear power station equipment inspection is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent inspection, more particularly, the present application relates to a kind of based on augmented reality's inspection data acquisition and comparative analysis method. BACKGROUND

[0002] Nuclear power plant equipment inspection is a key link to ensure nuclear safety. In traditional inspection, the operator needs to check the equipment regularly and record the operating parameters such as temperature, noise and vibration of the equipment. Due to the special nature of the nuclear power plant environment, such as high radiation area, dense distribution of equipment and complex electromagnetic environment, the inspection personnel face challenges such as low data collection accuracy, high safety risk and high operation difficulty when performing tasks. At the same time, the data generated in the traditional inspection process often exists in the form of paper records or simple electronic spreadsheets, lacking effective spatio-temporal correlation and multi-source data fusion capability, making it difficult to discover and accurately locate the abnormal state of the equipment in a timely manner.

[0003] The existing technology mainly relies on single sensor to collect equipment operation data, or through manual observation to record equipment state. These methods are greatly affected by human factors, and the data collection is not comprehensive, which cannot form a stereoscopic and dynamic monitoring of the equipment state. At the same time, due to the lack of multi-source data fusion analysis capability, the traditional inspection method cannot establish a correlation model between the equipment state and its operating environment, resulting in low abnormality recognition accuracy, delayed early warning, and inability to support accurate abnormal diagnosis and processing guidance. In addition, the traditional inspection results cannot form standardized knowledge accumulation, making experience inheritance and continuous optimization challenging, and unable to meet the demand for high reliability of nuclear power plant equipment operation.

[0004] In view of this, the present application proposes a kind of based on augmented reality's inspection data acquisition and comparative analysis method to solve the above problems. SUMMARY

[0005] In order to overcome the above-mentioned defects of the prior art, in order to achieve the above-mentioned purpose, the present application provides the following technical scheme: A kind of based on augmented reality's inspection data acquisition and comparative analysis method, comprising: Step S1: multi-source data collection of inspection site is carried out through augmented reality equipment, corresponding multi-source inspection data is obtained, and the collected multi-source inspection data is fused to generate a three-dimensional inspection data atlas; Step S2: based on the generated three-dimensional inspection data atlas, dynamic behavior analysis and abnormal trend prediction are carried out to obtain an inspection data feature model; Step S3: based on the obtained inspection data feature model, and combining the preset inspection standard data and historical inspection records, multi-dimensional data comparison analysis is carried out to generate inspection anomaly recognition and classification results; Step S4: Based on the inspection anomaly identification and classification results, and combined with the spatial marking function of the augmented reality device, the spatial positioning and visual marking of the abnormal area are carried out to obtain the inspection anomaly processing guide; Step S5: Based on the inspection anomaly processing guide, combined with the remote collaboration function, expert guidance and real-time feedback are carried out to generate an inspection closed-loop data report.

[0006] Further, the acquisition process of the multi-source inspection data includes: The depth image of the target inspection device is collected through the depth sensing terminal in the augmented reality device; the temperature distribution data of the target inspection device is collected through the thermal imaging sensing terminal of the augmented reality device; and the acoustic characteristic data of the target inspection device is collected through the acoustic sensing terminal of the augmented reality device. The spatial position data of the target inspection device is determined through the inertial measurement unit and the spatial positioning algorithm of the augmented reality device; and the collected data is associated to a unified spatial coordinate system to form multi-source inspection data.

[0007] Further, the construction process of the three-dimensional inspection data graph includes: The depth image is denoised and completed, and the processed depth image is converted to generate corresponding three-dimensional point cloud data; the temperature distribution data and acoustic characteristic data are adaptively denoised according to the scene; and the three-dimensional point cloud data and the processed temperature distribution data and acoustic characteristic data are fused to generate a three-dimensional inspection data graph.

[0008] Further, the three-dimensional inspection data graph includes the geometric shape, spatial position, running parameter and environmental state of the device.

[0009] Further, the implementation process of the scene adaptive denoising includes: A feature library containing common noise types is established; the time-frequency features of the input temperature distribution data and acoustic characteristic data are extracted; the similarity of each noise type in the feature library is obtained, the current main noise type and mixed noise composition are identified based on the similarity, and the optimal denoising algorithm is selected from the pre-constructed denoising database based on the denoising processing.

[0010] Further, the construction process of the inspection data feature model includes: The three-dimensional inspection data graph is time-sliced and evolution pattern extracted to obtain a device state change sequence, and time series analysis and periodic pattern recognition are performed to obtain a device running cycle graph; Based on the device running cycle graph, state transition probability calculation and Markov chain modeling are performed, and future state prediction and uncertainty quantification are performed to obtain a state prediction probability distribution; Based on the state prediction probability distribution and the pre-defined spatial correlation rule, the device group behavior is analyzed and the mutual influence is evaluated, and the abnormal propagation path is analyzed and the risk area is identified, and the risk propagation graph is obtained; Based on the risk propagation graph and the pre-defined historical abnormal pattern library, similarity matching and case reasoning are performed to obtain an abnormal type feature library, and based on the abnormal type feature library, multi-source abnormal pattern fusion and classifier training are performed to obtain an abnormal recognition model; The abnormal recognition model is integrated with the spatio-temporal features for ensemble learning and parameter optimization to obtain a spatio-temporal sensitive feature expression framework, and based on the spatio-temporal sensitive feature expression framework, multi-dimensional feature integration and model packaging are performed to obtain a patrol data feature model.

[0011] Further, the process of obtaining the patrol abnormal recognition and classification result includes: A dynamic threshold adjustment mechanism is established, and a baseline threshold is adjusted in combination with the multi-source patrol data to obtain a set of adaptive abnormal thresholds; based on the set of adaptive abnormal thresholds, the current multi-source patrol data is subjected to multi-level screening and priority sorting, and feature correlation analysis and root cause inference are performed to obtain an abnormal correlation graph; Based on the abnormal correlation graph, an abnormal type recognition and classification model is applied to obtain a preliminary classification result, and a classification label with confidence weighting is constructed based on the preliminary classification result; Based on the classification label, an abnormal severity assessment and impact range analysis are performed to obtain a risk level matrix, and abnormal information aggregation and structured description are performed to generate a patrol abnormal recognition and classification result.

[0012] Further, the patrol abnormal recognition and classification result includes abnormal type, location, severity, root cause and treatment priority.

[0013] Further, the process of obtaining the patrol abnormal treatment guide includes: Based on the patrol abnormal recognition and classification result, a virtual anchor point is generated and a position is calibrated in an augmented reality space to obtain a calibrated abnormal area boundary; Based on the calibrated abnormal area boundary, a visual marker design and transparency optimization are performed to obtain a hierarchical visual marker scheme, and based on the hierarchical visual marker scheme, an abnormal type symbol design and color coding are performed to obtain an abnormal identification system; Based on the abnormal identification system, in combination with the interaction function of the augmented reality device, an interaction response rule design and gesture recognition configuration are performed to obtain an abnormal area interaction framework, and based on the abnormal area interaction framework, a multi-level information display is performed; Based on the multi-level information display result, a processing flow visualization is performed to obtain a visual operation guide, and a safety operation constraint rule embedding and a risk prompt strategy configuration are performed to obtain a safety handling specification; Based on the safety processing specification, the abnormal processing knowledge base association and expert experience extraction are carried out, and the processing scheme generation and priority sorting are carried out, and the inspection abnormal processing guide is generated.

[0014] Further, the acquisition process of the inspection closed loop data report includes: Based on the inspection abnormal processing guide, and combined with the remote collaboration session, a collaborative space sharing framework for on-site view sharing and augmented reality marker synchronization is established; Based on the collaborative space sharing framework, expert remote marker generation and three-dimensional space guide projection are carried out, and a remote guidance marker set is obtained, and real-time interaction and operation confirmation are carried out with the on-site operator through the remote guidance marker set, and operation execution feedback is obtained; Based on the operation execution feedback, the processing process record and key step screenshot are carried out, the processing process archive is obtained, and the processing effect evaluation and before and after comparison analysis are carried out based on the processing process archive, and the processing effect verification data is obtained; Based on the processing effect verification data, the abnormal state repair confirmation and reinspection data acquisition are carried out, the closed loop verification result is obtained, and the processing measure effectiveness evaluation and experience summary are carried out through the closed loop verification result, and the processing experience knowledge item is obtained; Based on the processing experience knowledge item, the report automatic generation and key information extraction are carried out, and the data visualization processing and multi-format export are carried out, and the inspection closed loop data report is generated.

[0015] The technical effects and advantages of the inspection data acquisition and comparative analysis method based on augmented reality of the present application are as follows: The present application can comprehensively perceive the running state of the nuclear power plant equipment, greatly improves the intelligent level of the inspection, acquires multi-source inspection data such as depth image, temperature distribution and acoustic characteristics through the augmented reality equipment, not only covers the full range of monitoring requirements of the equipment, but also obtains high-precision space-time correlation information, adopts the method based on three-dimensional data graph construction and feature model analysis, can accurately identify the abnormal state of the equipment, and real-time track the parameter change trend, this precise equipment state perception ability, for realizing the intelligent abnormal identification and processing lays the key foundation, when constructing the inspection data feature model, not only considers the time sequence characteristics and spatial correlation of the equipment, such as running period, state transition and mutual influence between equipment, etc. Key factors, but also combines dynamic threshold adjustment and root cause analysis technology, constructs a comprehensive abnormal identification and processing system, can generate intuitive and easy-to-understand augmented reality processing guide, not only greatly improves the efficiency and accuracy of the inspection, also greatly enhances the safety and reliability of the nuclear power plant equipment operation; At the same time, through the functions of three-dimensional data acquisition, space marking, remote collaboration of AR technology, the intelligent, spatial and closed loop of inspection data acquisition and analysis are realized. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 This is a schematic diagram of an augmented reality-based inspection data acquisition and comparative analysis method according to the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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.

[0018] Example 1 Please see Figure 1 As shown in this embodiment, a method for collecting and comparing inspection data based on augmented reality includes: Step S1: Collect multi-source data from the inspection site using augmented reality equipment to obtain corresponding multi-source inspection data, and fuse the collected multi-source inspection data to generate a three-dimensional inspection data map containing equipment status, environmental parameters, and spatial location. Step S2: Based on the generated 3D inspection data map, perform dynamic behavior analysis and anomaly trend prediction to obtain an inspection data feature model that includes time and space dimensions; Step S3: Based on the obtained inspection data feature model, and combined with the preset inspection standard data and historical inspection records, perform multi-dimensional data comparison and analysis to generate inspection anomaly identification and classification results; Step S4: Based on the inspection anomaly identification and classification results, and combined with the spatial marking function of the augmented reality device, spatial positioning and visual annotation of the abnormal area are performed to obtain inspection anomaly handling guidelines; Step S5: Based on the inspection anomaly handling guidelines and combined with remote collaboration functions for expert guidance and real-time feedback, generate an inspection closed-loop data report containing handling measures and verification results.

[0019] In one embodiment of the present invention, the multi-source inspection data can be acquired through various sensing terminals configured on the corresponding augmented reality device, including depth sensing terminals, thermal imaging sensing terminals, acoustic sensing terminals, and UWB positioning devices, etc., wherein the acoustic sensing terminal is used to collect acoustic feature data; the UWB positioning device determines the spatial position data of the target inspection device based on an inertial measurement unit and a spatial positioning algorithm; the augmented reality device can be a head-mounted AR glasses or a handheld AR terminal, etc. The three-dimensional data inspection map is generated by spatiotemporal fusion processing of the collected multi-source inspection data. The three-dimensional inspection data map is a mapping model between the physical space of the equipment and the data feature space, including the geometric shape, spatial location, operating parameters and environmental conditions of the equipment. Both the inspection data feature model and the preset inspection standard data are trained using historical inspection data collected during equipment inspection. By analyzing the historical inspection data, the status characteristic data of different equipment are statistically analyzed, including information such as temperature distribution, vibration characteristics, acoustic characteristics and spatial location. Corresponding equipment status templates are established, each feature parameter is labeled, normal samples are identified, and the equipment status templates and feature samples are saved to the database. The inspection data feature model and the preset inspection standard data are established respectively. The inspection data feature model can be used to support the correlation mapping between equipment status evolution over time and spatial distribution, and can predict the propagation path and speed of abnormal equipment status. In the process of fusing multi-source inspection data, it is necessary to perform depth image denoising and completion processing on the 3D point cloud data collected by the corresponding depth sensing terminal. This includes adaptively adjusting the computational complexity of the denoising algorithm based on the current power consumption state of the device, and selecting the optimal image denoising model parameters in combination with the spatial characteristics of the inspection scene. Then, the depth image is progressively denoised through a block processing strategy, prioritizing the processing of high information entropy regions, and the missing depth regions are completed through an edge-preserving structured interpolation algorithm.

[0020] It should be noted that due to the complex environment and multiple noise sources within a nuclear power plant, the data quality of augmented reality devices during data acquisition can be affected by the environmental complexity. Therefore, scene-adaptive denoising processing is required for the temperature distribution data and acoustic feature data within the multi-source inspection data. The corresponding scene-adaptive denoising process determines the noise type corresponding to the current scene through real-time noise pattern recognition, selects the optimal denoising algorithm based on the noise type, and dynamically adjusts the processing intensity of the denoising algorithm according to the signal-to-noise ratio. Furthermore, wavelet transform is used to separate noise of different frequencies, and spatial correlation verification is performed on the denoised temperature distribution data and acoustic feature data to eliminate artifacts introduced during the denoising process. At the same time, to ensure real-time performance on resource-constrained augmented reality devices, the algorithm complexity of scene-adaptive denoising processing does not exceed a pre-set fixed upper limit.

[0021] A dynamic threshold adjustment mechanism is established based on the feature model of inspection data. This mechanism automatically adjusts the equipment status assessment threshold according to environmental parameters, equipment operating conditions, and historical data distribution, resulting in an adaptive anomaly threshold set. By performing multi-level filtering and priority sorting on current multi-source inspection data, an anomaly candidate list is determined. An anomaly correlation map is obtained through feature correlation analysis and root cause inference. Based on the anomaly correlation map, anomaly type identification and classification models are applied to obtain preliminary classification results. Through uncertainty assessment and expert rule verification, confidence-weighted classification labels are obtained. The severity of the anomaly is comprehensively assessed based on the degree to which the anomaly parameters deviate from the dynamic threshold and the potential risk level, categorized into four levels: urgent, severe, moderate, and minor, marked with red, orange, yellow, and blue, respectively. When anomaly identification and classification fail, the augmented reality device is restarted to acquire multi-source inspection data for analysis. An alarm is triggered when the number of repeated detections exceeds the system's preset number.

[0022] It should be further explained that, in the specific implementation process, the acquisition of multi-source inspection data includes the following steps: Step S111: Acquire depth images of the target inspection equipment through the depth sensing terminal in the augmented reality device, and generate corresponding three-dimensional point cloud data by performing coordinate transformation on the depth images; Step S112: Collect temperature distribution data of the target inspection equipment through the thermal imaging sensor terminal of the augmented reality device; Step S113: Collect acoustic feature data of the target inspection equipment through the acoustic sensing terminal of the augmented reality device; Step S114: Determine the spatial location data of the target inspection equipment using the inertial measurement unit and spatial positioning algorithm of the augmented reality device; Step S115: Associate the three-dimensional point cloud data, temperature distribution data, acoustic feature data and spatial location data into a unified spatial coordinate system to form spatiotemporally correlated multi-source inspection data; Specifically, due to the complexity of the nuclear power plant environment, such as dense equipment distribution, strong electromagnetic interference, and radiation interference, noise is present in the acquired multi-source inspection data, resulting in decreased data quality and affecting subsequent data analysis and anomaly identification. Therefore, targeted processing of the collected multi-source inspection data is required. For example, noise removal can be performed on depth images; temperature distribution data can be corrected for temperature anomalies using heatmap smoothing algorithms; and acoustic feature data can be filtered out for environmental background noise through spectral analysis. Due to the influence of equipment location and the perspective of inspection personnel, the collected data may be obstructed or have angular deviations. Spatial registration technology is required to perform unified coordinate transformation on the collected multi-source data to ensure that the data are fused under the same spatial reference system.

[0023] It should be further explained that, in the specific implementation process, the construction of the 3D inspection data map includes the following steps: Step S121: Denoise and complete the depth image, and perform coordinate transformation on the processed depth image to generate the corresponding 3D point cloud data; Step S122: Perform scene-adaptive denoising processing on temperature distribution data and acoustic feature data; Step S123: The three-dimensional point cloud data, the processed temperature distribution data, and the acoustic feature data are fused together to generate a three-dimensional inspection data map. Specifically, depth image denoising and completion processing is designed for resource-constrained augmented reality devices. It involves real-time monitoring of the device's current battery level and processor load, dynamically adjusting the computational complexity of the selected denoising algorithm based on the monitoring results. For example, when the device's battery level falls below a preset threshold, it automatically switches to a low-power mode to reduce computation. Depth image denoising employs a block-based processing strategy, dividing the entire depth image into multiple sub-regions and prioritizing high-information-entropy regions, i.e., regions containing critical device components, to ensure data quality in important areas while reducing overall computation. For missing regions in the depth image, edge-preserving structured interpolation algorithms are used for completion, preserving depth boundary information and avoiding blurring at structural edges. The computational complexity is adjusted linearly with the remaining battery power of the augmented reality device, corresponding to the parameters of the denoising algorithm. Scene-adaptive denoising processing can automatically adjust denoising strategies according to different noise environments. It uses a real-time noise pattern recognition algorithm to determine the noise type of the current environment, such as Gaussian noise, impulse noise, or mixed noise. Then, for the identified noise type, it selects the optimal denoising method from a preset denoising algorithm library, such as median filtering, wavelet denoising, or nonlocal mean filtering. The processing intensity of the selected denoising algorithm is dynamically adjusted according to the signal-to-noise ratio of the current temperature distribution data and acoustic feature data. The processing intensity is increased when the signal-to-noise ratio is low and decreased when the signal-to-noise ratio is high, achieving a balance between noise suppression and information preservation. At the same time, to ensure real-time processing on resource-constrained augmented reality devices, the complexity of all denoising algorithms is set with a fixed upper limit on the complexity processing capacity, which is achieved through algorithm optimization and parallel computing technology. Furthermore, the three-dimensional point cloud data, the processed temperature distribution data, and the acoustic feature data are characterized, and three-dimensional modeling is performed using the characterized data. The characterized data is then partitioned and mapped using spatial features to obtain a three-dimensional inspection data map. The characteristic processing includes steps such as data dimensionality reduction, feature extraction, and feature selection. The corresponding processing procedures are all publicly available technologies and will not be elaborated upon in this application.

[0024] It should be further explained that, in the specific implementation process, the construction of the inspection data feature model includes the following steps: Step S201: Perform time-series slicing and evolution pattern extraction on the three-dimensional inspection data map to obtain the equipment state change sequence, and perform time-series analysis and periodic pattern recognition based on the equipment state change sequence to obtain the equipment operation cycle map; Step S202: Calculate the state transition probability and model the Markov chain based on the equipment operation cycle map to obtain the equipment state transition matrix. Then, predict the future state and quantify the uncertainty based on the equipment state transition matrix to obtain the state prediction probability distribution. Step S203: Based on the state prediction probability distribution and spatial association rules, conduct collaborative analysis and mutual influence assessment of the behavior of the equipment group to obtain the equipment association network model, and use the equipment association network model to analyze the abnormal propagation path and identify risk areas to obtain the risk propagation map; Step S204: Based on the risk propagation map and historical anomaly pattern library, perform similarity matching and case reasoning to obtain an anomaly type feature library, and use the anomaly type feature library to perform multi-source anomaly pattern fusion and classifier training to obtain an anomaly recognition model; Step S205: Integrate the anomaly recognition model with spatiotemporal features for learning and parameter optimization to obtain a spatiotemporally sensitive feature representation framework. Based on the spatiotemporally sensitive feature representation framework, perform multi-dimensional feature integration and model encapsulation to construct an inspection data feature model that includes time and space dimensions.

[0025] Specifically, the time series data of the 3D inspection data map is acquired, and time-series slicing and evolution pattern extraction are performed on it to obtain the equipment state change sequence. Time-series slicing refers to systematically sampling the 3D data map according to a predefined time interval. Evolution pattern extraction is used to identify the rules and patterns of equipment state changes over time and generate a state change sequence that characterizes the dynamic behavior of the equipment. The equipment operation cycle map is obtained by matching the equipment state change sequence with time series analysis algorithms and identifying periodic patterns. Matching refers to the sequence analysis of the corresponding state change sequence using time series analysis methods. The sequence analysis process needs to consider the differential processing of sequence characteristics, including regression analysis for linear trends, Fourier transform for periodic changes, and wavelet analysis for nonlinear patterns. Periodic pattern recognition is based on spectrum analysis and autocorrelation technology to identify the implicit periodic patterns and generate the equipment operation cycle map containing periodic information such as the main period, amplitude characteristics, phase relationship, and time dependence. Based on the equipment's operating cycle map, state transition probabilities are calculated and Markov chain modeling is performed to obtain the equipment state transition matrix. State transition probability calculation involves statistically analyzing the transition frequencies between equipment states and setting corresponding probability models based on the transition patterns between different states to generate the transition probabilities between each pair of states. Markov chain modeling constructs a state transition model based on the state space definition and transition probabilities, and uses maximum likelihood estimation or Bayesian inference techniques for parameter optimization. Based on the results of state transition probability calculation and Markov chain modeling, a matrix is ​​constructed to obtain a state transition matrix that can describe the dynamic changes in equipment states. Based on the equipment state transition matrix, future state prediction and uncertainty quantification are performed to obtain the state prediction probability distribution. Future state prediction refers to generating state distribution predictions for multiple future time points by iteratively applying the state transition matrix. The prediction process uses matrix iterative calculation or Monte Carlo simulation methods. Uncertainty quantification evaluates the reliability and accuracy of the prediction results through entropy calculation or confidence interval estimation, generating a state prediction probability distribution that includes uncertainty indicators such as each possible state and its probability, prediction time window, confidence level, and entropy value. Based on state prediction probability distribution and spatial association rules, device group behavior modeling and mutual influence evaluation are performed to obtain a device association network model. Device group behavior modeling refers to establishing a collective behavior model for interrelated device groups and constructing an association relationship network using graph theory or complex network theory. Mutual influence evaluation, on the other hand, uses association rule mining or Bayesian network technology to quantify the strength and direction of mutual influence between devices and generate a device association network model that represents the dynamic interaction relationship between devices. Anomaly propagation path analysis and risk area identification are performed based on the device association network model to obtain a risk propagation map. Anomaly propagation path analysis refers to simulating the propagation process and diffusion path of anomalies in the device network using network flow algorithms or infectious disease models. Risk area identification is based on key node identification and community detection algorithms to determine high-risk areas and key devices affected by anomalies, and to generate a risk propagation map containing risk information such as propagation path, scope of impact, time evolution, and key nodes. By integrating risk propagation maps and historical anomaly case libraries, and training models based on these, an inspection data feature model containing both time and spatial dimensions is obtained.

[0026] It should be further explained that, in the specific implementation process, the process of obtaining the inspection anomaly identification and classification results includes the following steps: S301: Establish a dynamic threshold adjustment mechanism and automatically adjust the equipment status assessment threshold according to environmental parameters, equipment operating conditions and historical data distribution to obtain an adaptive abnormal threshold set; S302: Based on the adaptive anomaly threshold set, perform multi-level screening and priority sorting on the current multi-source inspection data to obtain an anomaly candidate list, and perform feature correlation analysis and root cause inference through the anomaly candidate list to obtain an anomaly correlation map; S303: Based on the anomaly association graph, anomaly type identification and classification model application is performed to obtain preliminary classification results. Uncertainty assessment and expert rule verification are then performed based on the preliminary classification results to obtain confidence-weighted classification labels. S304: Based on confidence-weighted classification labels, assess the severity of anomalies and analyze the scope of impact to obtain a risk level matrix. Then, aggregate and structure the anomaly information using the risk level matrix to generate inspection anomaly identification and classification results. Specifically, the dynamic threshold adjustment mechanism includes environmental factor influence models, operating condition influence models, and historical trend influence models, as well as the influence weights and adjustment ranges of each model. Based on the current distribution of multi-source inspection data and historical data, the pre-set benchmark thresholds are dynamically adjusted. For example, the environmental factor influence model calculates the threshold adjustment amount caused by the environment based on current environmental parameters (such as ambient temperature, humidity, radiation intensity, etc.), the operating condition influence model calculates the threshold adjustment amount caused by operating conditions based on current equipment load and operating status, and the historical trend influence model calculates the threshold adjustment amount caused by historical changes based on the historical trend of parameter changes. Then, based on the threshold adjustment amounts caused by the environment, operating conditions, and history, the corresponding benchmark thresholds are dynamically adjusted to obtain an adaptive abnormal threshold set, which includes threshold information such as anomaly judgment thresholds and confidence intervals for each dimension. The multi-level screening process first performs a coarse screening based on a single-parameter threshold, and then a fine screening based on a multi-parameter joint probability distribution. The screening process adopts a cascaded screening architecture and multi-threshold judgment logic. Prioritization considers factors such as anomaly severity, equipment importance, and system impact, prioritizing the screened anomalies and generating an anomaly candidate list with a clear processing order. Feature correlation analysis uses methods such as mutual information, partial correlation coefficient, and Granger causality test to identify statistical correlations and causal relationships between anomaly parameters. Root cause inference is based on Bayesian network or decision tree models to trace the source cause of the anomaly and evaluate the probability distribution of multiple possible causes. The anomaly correlation graph uses a directed graph structure to represent the propagation relationship and impact path between anomaly events. Nodes in the graph represent anomaly states, and edges represent impact relationships and their strength. It not only shows the spatial distribution and propagation trend of anomalies, but also intuitively reflects the severity and priority of anomalies through color coding and icon size. The anomaly type identification and classification model integrates multiple machine learning algorithms to categorize anomaly features into predefined types, such as "equipment overheating," "vibration anomaly," and "radiation leakage." It automatically selects the most suitable classifier, such as decision trees, support vector machines, or deep neural networks, based on the complexity of the anomaly features. During classification, an imbalanced sample handling technique is introduced, synthesizing minority class samples and majority class undersampling to balance the training data distribution across different anomaly types. In the model application phase, a hierarchical ensemble strategy is employed, first performing coarse-grained broad anomaly category classification, and then fine-grained classification for specific subcategories to improve the efficiency of classification. The method achieves a high anomaly identification rate. For the reliability assessment of classification results, a Bayesian uncertainty quantification method is introduced, and Monte Carlo sampling is used to calculate the predicted confidence interval. Simultaneously, expert knowledge rules are combined to perform secondary confirmation on low-confidence predictions, generating classification labels with confidence weights. These labels not only include the anomaly type but also include confidence scores and uncertainty indicators, providing more comprehensive information support for subsequent decision-making. This method significantly improves the accuracy and interpretability of anomaly classification, especially for complex or rare fault types, providing reasonable confidence assessments and avoiding inappropriate handling due to misdiagnosis. The severity assessment and impact range analysis of anomalies are conducted by constructing a risk quantification framework based on three dimensions: equipment functional importance, probability of fault propagation, and difficulty of repair. Risk scores are calculated by combining impact records of similar faults from a historical case library with key equipment indicators. Impact range analysis utilizes an anomaly correlation graph generated from the risk scores and simulates the fault propagation path using a graph propagation algorithm to identify potentially affected areas and critical nodes. This is then visually presented in the form of a risk level matrix, with the horizontal axis representing the probability of occurrence and the vertical axis representing the severity of the impact, and matrix units corresponding to different risk levels. Finally, through a structured information organization template, information such as anomaly type, severity, root cause analysis, and recommended measures are integrated into a standardized anomaly report. At the same time, customized views for different roles are generated to meet the differentiated information needs of operations personnel, managers, and experts. Ultimately, inspection anomaly identification and classification results containing comprehensive information such as anomaly type, location, severity, root cause, and processing priority are generated.

[0027] It should be further explained that, in the specific implementation process, the process of obtaining the inspection anomaly handling guidelines includes the following steps: Step S401: Based on the inspection anomaly identification and classification results, perform spatial coordinate mapping and precise positioning calculation to obtain the three-dimensional coordinate set of anomaly points. Then, use the three-dimensional coordinate set of anomaly points to generate virtual anchor points and perform position calibration in augmented reality space to obtain the calibrated anomaly area boundary. Step S402: Based on the calibrated abnormal region boundaries, visual marking design and transparency optimization are performed to obtain a hierarchical visual marking scheme. Then, abnormal type symbols are designed and color-coded using the hierarchical visual marking scheme to obtain an intuitive abnormal identification system. Step S403: Based on the intuitive anomaly identification system and the interactive function of the augmented reality device, design the interactive response rules and configure the gesture recognition for the marked area to obtain the anomaly area interaction framework. Then, design the information display hierarchy structure and define the view switching logic through the anomaly area interaction framework to obtain a multi-level information display mechanism. Step S404: Based on the multi-level information display mechanism, perform visual design and step decomposition of the processing flow to obtain a visual operation guide. Then, through the visual operation guide, embed safety operation constraint rules and configure risk warning strategies to obtain a safety processing specification. Step S405: Based on the safety handling specifications, perform anomaly handling knowledge base association and expert experience extraction to obtain a knowledge-assisted decision support system, and use the knowledge-assisted decision support system to generate handling solutions and prioritize them to generate inspection anomaly handling guidelines; Specifically, spatial coordinate mapping and precise positioning calculations employ multi-sensor fusion positioning technology, achieving centimeter-level positioning accuracy by combining visual SLAM algorithms and inertial measurement unit data. For example, when a densely packed area of ​​metal equipment is detected, the system automatically switches to ultrasonic sensing terminals for assisted positioning, avoiding positioning drift caused by magnetic interference. The coordinate mapping process uses a distributed reference point strategy, pre-deploying high-precision QR codes or RFID tags as spatial anchors in the industrial environment to establish a coordinate correspondence between physical space and digital twin space. For the precise calibration of anomalies, a progressive precision optimization algorithm is used. First, a coarse positioning is performed based on the global coordinate system, then fine-tuning is done using local feature matching and depth information, and finally, constraints from the equipment CAD model are combined to ensure that the marker points are strictly attached to the surface of the equipment entity, effectively solving the "floating marker" problem in augmented reality. The visual marker design and transparency optimization process involves using a hierarchical visual coding method for anomalies of varying severity to ensure intuitive identification of key information. The marker design employs semi-transparent outline highlighting technology, dynamically adjusting the marker edge clarity and fill transparency based on viewing distance to reduce visual obstruction at close range and enhance visibility at long distances. The designed visual markers consider various industrial lighting conditions, adjusting marker brightness and contrast in real-time through ambient light perception to ensure good legibility in both strong and low light environments. Simultaneously, the anomaly type symbol design utilizes a dual redundancy strategy combining shape and color coding, enabling staff to accurately identify anomaly types. Furthermore, the anomaly type symbol design conforms to international industrial safety standards and specifications, achieving universal understanding among multiple technical personnel. The interactive response rule design and gesture recognition configuration are designed to meet the specific needs of industrial scenarios, supporting robust gesture recognition even when wearing industrial gloves. The interactive framework supports multiple interaction methods, including gestures, voice, and eye tracking, which technicians can flexibly switch between based on ambient noise and operational constraints. The designed interaction rules follow the "principle of least action," reducing cognitive load by simplifying the operation process. Secondary confirmation mechanisms are set up for critical operations such as anomaly confirmation and completion of processing to prevent misoperation. The information display hierarchy adopts a "demand-driven" design concept. The basic layer displays key information such as anomaly type and severity, the advanced layer provides detailed technical parameters and historical comparison data, and the expert layer presents root cause analysis and system impact assessment. Different levels of information display are triggered by simple gestures or gaze duration to meet the needs of personnel with different technical levels. The process visualization design and step decomposition utilize cognitive task analysis to break down complex maintenance processes into intuitive sequences of steps, setting clear start states and completion criteria for each step. The visual operation guide employs augmented reality spatial registration technology to precisely overlay virtual operation instructions onto corresponding physical equipment locations, demonstrating correct operating postures and tool usage methods through animation. An adaptive difficulty control mechanism is incorporated into the process design, automatically adjusting the level of detail in the instructions based on the technician's experience level. The novice mode provides detailed step decompositions and tool usage instructions, while the expert mode only displays key points and precautions. Simultaneously, safety operation constraints employ a pre-verification approach, mandating the display of safety warnings and requiring confirmation before performing dangerous operations. For specific high-risk operations such as live-line maintenance or high-voltage system intervention, the system automatically checks whether safety protection measures are activated to ensure operational safety. Furthermore, a multi-dimensional knowledge network is constructed, encompassing equipment type, failure mode, handling methods, and successful cases. Knowledge extraction employs a semi-automated approach, combining natural language processing technology to extract structured knowledge from maintenance reports and technical documents, with expert review mechanisms ensuring knowledge quality. The decision support system utilizes a hybrid reasoning framework combining case-based and rule-based reasoning, retrieving matching cases based on the similarity between the current anomaly and historical cases, while adaptively adjusting to specific constraints of the current equipment. A multi-objective optimization algorithm is introduced during the processing solution generation process, considering multiple objectives such as repair time, resource consumption, downtime impact, and safety risks, generating multiple feasible solutions and prioritizing them based on current environmental conditions and resource availability. The final anomaly handling guide includes structured operation steps, a list of required tools, safety precautions, and expected repair results, presented intuitively through an augmented reality interface. It also supports real-time adjustments and decision support during the handling process, ensuring the efficiency and safety of anomaly handling.

[0028] It should be further explained that, in the specific implementation process, the acquisition of the inspection closed-loop data report includes the following steps: Step S501: Based on the inspection anomaly handling guidelines, establish a remote collaboration session and transmit real-time video streams to obtain a two-way communication channel. Then, use the two-way communication channel to share on-site views and synchronize augmented reality markers to obtain a collaborative space sharing framework. Step S502: Based on the collaborative space sharing framework, perform expert remote marker generation and 3D space guidance projection to obtain a remote guidance marker set, and use the remote guidance marker set to interact with on-site operators in real time and confirm operations to obtain operation execution feedback; Step S503: Based on the operation execution feedback, record the processing process and take screenshots of key steps to obtain the processing process archive. Then, evaluate the processing effect and conduct before-and-after comparison analysis through the processing process archive to obtain processing effect verification data. Step S504: Based on the processing effect verification data, confirm the abnormal state repair and collect re-inspection data to obtain closed-loop verification results. Then, evaluate the effectiveness of the processing measures and summarize the experience through the closed-loop verification results to obtain processing experience knowledge items. Step S505: Based on the knowledge items of processing experience, automatically generate reports and extract key information to obtain a structured report template. Then, use the structured report template to perform data visualization processing and export in multiple formats to generate a closed-loop inspection data report containing processing measures and verification results. Specifically, the establishment of remote collaboration sessions and real-time video streaming employs adaptive bandwidth management technology, dynamically adjusting video stream resolution and frame rate based on network conditions to ensure connection stability in limited network environments. For example, when a decrease in network bandwidth is detected, the system automatically reduces video quality in non-critical areas while maintaining high-definition transmission in areas of interest. Communication channel establishment utilizes a multi-path redundancy transmission strategy, simultaneously establishing dual-channel connections using both cellular and industrial Wi-Fi networks. Real-time latency monitoring automatically switches to the optimal path to ensure communication reliability. For augmented reality marker synchronization, a lightweight differential transmission protocol is used, transmitting only marker position and attribute change information, rather than complete scene data, significantly reducing synchronization bandwidth requirements. Furthermore, by introducing an automatic spatial anchor point calibration mechanism, the system periodically compares feature points on field devices with views from remote experts, automatically correcting spatial offsets caused by changes in viewing angle to ensure accurate overlay of augmented reality markers. Expert remote marker generation and 3D spatial guidance projection are based on an intuitive interactive design concept, allowing experts to quickly create spatial guidance markers through various input methods (touch, gestures, and voice). Marker generation uses semantic marker technology, with preset templates for common guidance types such as "rotation direction," "disassembly sequence," and "attention area." Experts only need to select the type and specify the location to generate standardized 3D markers. The created markers have spatial awareness characteristics, automatically attaching to the corresponding surface according to the shape of the equipment on site, and achieving precise tracking between the markers and the equipment through physical constraint recognition. The 3D guidance projection uses a semi-transparent layered visualization method, with different priority guidance using different depth layers and saliency codes to ensure that key guidance is visually highlighted. At the same time, it supports context-aware marker display control. When the operator looks at a specific area or performs a specific action, the relevant guidance is automatically highlighted and detailed information is displayed, while irrelevant guidance is automatically faded to reduce visual interference. The process recording and key step screenshotting process automatically detects critical operation moments through action recognition algorithms and triggers high-resolution screenshots and short video recordings. The recording process adopts a layered storage strategy, with continuous low-resolution background recording to ensure integrity, while key steps are preserved in high-definition detail, optimizing storage space utilization. Simultaneously, the system integrates audio recording functionality, capturing on-site and remote expert voice communication and automatically generating operation instructions text through speech recognition technology. Processing effect evaluation employs computer vision-based difference detection technology, automatically identifying changes in equipment status and quantifying the degree of improvement through structured comparison of images before and after processing. Comparative analysis also incorporates thermal imaging data and acoustic feature data to evaluate whether anomalies have been effectively resolved from multiple dimensions, generating a verification data package containing numerical indicators and visual evidence. The verification data package is managed using a version control mechanism, supporting historical comparison and traceability of the repair process. The abnormal state repair confirmation and re-inspection data collection automatically selects appropriate detection methods for re-inspection based on the anomaly type. The re-inspection process adopts a structured sampling strategy, including dense sampling of the repair area according to a preset pattern and gradient decreasing sampling of the surrounding area to ensure comprehensive coverage while optimizing efficiency. Through equipment parameter regression testing and functional verification testing, the operating status of the equipment after repair is comprehensively evaluated. The closed-loop verification results generate a comprehensive health index through data fusion algorithms, intuitively quantifying the repair effect. At the same time, the effectiveness evaluation of the treatment measures adopts a multi-cycle tracking method, arranging regular inspections during the continuous monitoring period after repair to track the stability and persistence of the repair effect. Furthermore, by using structured knowledge extraction technology, key decision points, effective tools and methods, and expert skills are extracted from successful repair cases to form reusable knowledge items, and application rules are established according to applicable conditions and constraints. Furthermore, the automatic report generation and key information extraction are based on templated document generation technology, combined with natural language processing capabilities to transform structured data into fluent technical narratives. Report generation employs a hierarchical content organization strategy, automatically adjusting content depth and terminology based on the target audience's role (on-site technicians, management, and technical experts). The key information extraction process identifies high-value information points based on dimensions such as anomaly severity, repair complexity, and innovativeness, ensuring their prominent display in the summary and highlights sections. Data visualization processing supports multi-dimensional interactive charts, allowing recipients to flexibly switch between different views on a unified data basis, such as time-series diagrams, comparison charts, and heatmaps. Simultaneously, it supports multi-format report export, including a lightweight mobile version suitable for on-site viewing, a detailed PDF version suitable for archiving, a structured data package suitable for data analysis, and a semantically tagged version suitable for knowledge base integration. All report versions maintain core data consistency, ensuring traceability through a unified identifier, ultimately forming a complete, professional, and easily disseminated closed-loop inspection data report, effectively supporting the full lifecycle management and continuous optimization and improvement of industrial equipment.

[0029] An embodiment of the present invention further includes: adaptive learning and optimization based on the inspection closed-loop data report, updating the inspection standard database and the inspection data feature model, and generating an intelligent inspection data analysis strategy; It should be noted that by using the comparison data before and after processing, expert guidance records, and final processing effect evaluation information in the inspection closed-loop data report, the processing mode and parameter sensitivity characteristics within the corresponding information are effectively learned. The corresponding learning results are used to update the threshold settings, parameter weights, and anomaly classification standards in the inspection standard database. They are also used to optimize the structure and parameter configuration of the inspection data feature model, so that it can more accurately identify and predict similar anomalies in future inspections. Meanwhile, based on the accumulation of multi-source inspection data, the system automatically adjusts operating parameters such as sampling frequency, analysis depth, and early warning level to achieve adaptive optimization of the inspection system, reduce false alarm rate and missed alarm rate, and improve inspection efficiency.

[0030] An embodiment of the present invention further includes the following: the implementation process of scene adaptive denoising processing includes: Establish a feature library of common noise types in nuclear power plant environments; extract time-frequency features of input temperature distribution data and corresponding time-series data and acoustic feature data through short-time Fourier transform; obtain the similarity between these features and each noise type in the feature library; identify the main noise type and mixed noise composition of the current scene based on the obtained similarity; and select the optimal denoising algorithm for denoising processing based on the identification results. It should be noted that the noise pattern recognition method solves the "one-size-fits-all" problem of traditional denoising algorithms, and can select the most suitable denoising algorithm for different types of noise. Among them, common noise types in nuclear power plant environments include Gaussian noise, impulse noise, periodic interference, and random mutations. Each type of noise has its specific time-frequency characteristics. By converting the time-domain signal into a time-frequency domain representation through short-time Fourier transform, it is easier to identify different types of noise. For the main noise types identified, the system will select the corresponding optimal denoising algorithm from the preset denoising algorithm library. For mixed noise, a cascaded denoising strategy is adopted to apply multiple denoising algorithms in a specific order.

[0031] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0032] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0033] In the description of this invention, it should be understood that the terms "first," "second," etc., are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance.

[0034] In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0035] In the description of this invention, "several" means one or more, and "a large number" means two or more.

[0036] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0037] All formulas in this manual are dimensionless and calculated numerically. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0038] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for collecting and comparing inspection data based on augmented reality, characterized in that, include: Step S1: Use augmented reality devices to collect multi-source data from the inspection site, obtain corresponding multi-source inspection data, and perform fusion processing on the collected multi-source inspection data to generate a three-dimensional inspection data map. Step S2: Based on the generated 3D inspection data map, perform dynamic behavior analysis and anomaly trend prediction to obtain the inspection data feature model; Step S3: Based on the obtained inspection data feature model, and combined with the preset inspection standard data and historical inspection records, perform multi-dimensional data comparison and analysis to generate inspection anomaly identification and classification results; Step S4: Based on the inspection anomaly identification and classification results, and combined with the spatial marking function of the augmented reality device, spatial positioning and visual annotation of the abnormal area are performed to obtain inspection anomaly handling guidelines; Step S5: Based on the inspection anomaly handling guidelines and combined with remote collaboration functions for expert guidance and real-time feedback, generate an inspection closed-loop data report.

2. The method for collecting and comparing inspection data based on augmented reality according to claim 1, characterized in that, The process of acquiring multi-source inspection data includes: Depth images of the target inspection equipment are acquired by a depth sensing terminal within an augmented reality device. Temperature distribution data of the target inspection equipment is collected by the thermal imaging sensor terminal in the augmented reality device; acoustic characteristic data of the target inspection equipment is collected by the acoustic sensor terminal in the augmented reality device. The spatial location data of the target inspection equipment is determined by the inertial measurement unit and spatial positioning algorithm in the augmented reality device; and the collected data is correlated to a unified spatial coordinate system to form multi-source inspection data.

3. The method for collecting and comparing inspection data based on augmented reality according to claim 2, characterized in that, The process of constructing a 3D inspection data map includes: The depth image is denoised and incomplete, and the processed depth image is transformed to generate corresponding 3D point cloud data; the temperature distribution data and acoustic feature data are subjected to scene-adaptive denoising; and the 3D point cloud data and the processed temperature distribution data and acoustic feature data are fused to generate a 3D inspection data map.

4. The method for collecting and comparing inspection data based on augmented reality according to claim 3, characterized in that, The 3D inspection data map includes the equipment's geometry, spatial location, operating parameters, and environmental conditions.

5. The method for collecting and comparing inspection data based on augmented reality according to claim 3, characterized in that, The implementation process of scene adaptive noise reduction includes: Establish a feature library containing common noise types; extract time-frequency features from input temperature distribution data and acoustic feature data; obtain the similarity between these features and each noise type in the feature library; identify the current main noise type and mixed noise composition based on the similarity; and select the optimal denoising algorithm from a pre-built denoising database for denoising processing based on these similarities.

6. The method for collecting and comparing inspection data based on augmented reality according to claim 4, characterized in that, The process of constructing the inspection data feature model includes: The three-dimensional inspection data map is sliced ​​and the evolution pattern is extracted to obtain the equipment state change sequence. Then, time series analysis and periodic pattern recognition are performed to obtain the equipment operation cycle map. Based on the equipment operation cycle map, state transition probability calculation and Markov chain modeling are performed, and future state prediction and uncertainty quantification are carried out to obtain the state prediction probability distribution. Based on the state prediction probability distribution and predefined spatial association rules, collaborative analysis and mutual influence assessment of device group behavior are performed, and anomaly propagation path analysis and risk area identification are conducted to obtain a risk propagation map. Based on the risk propagation graph and a predefined historical anomaly pattern library, similarity matching and case reasoning are performed to obtain an anomaly type feature library. Based on this feature library, multi-source anomaly pattern fusion and classifier training are performed to obtain an anomaly recognition model. By integrating the anomaly identification model with spatiotemporal features and optimizing the parameters, a spatiotemporally sensitive feature representation framework is obtained. Based on this framework, multi-dimensional features are integrated and the model is encapsulated to obtain the inspection data feature model.

7. The method for collecting and comparing inspection data based on augmented reality according to claim 6, characterized in that, The process of obtaining inspection anomaly identification and classification results includes: A dynamic threshold adjustment mechanism is established, and the baseline threshold is adjusted in combination with multi-source inspection data to obtain an adaptive anomaly threshold set. Based on the adaptive anomaly threshold set, the current multi-source inspection data is subjected to multi-level screening and priority sorting, and feature correlation analysis and root cause inference are performed to obtain an anomaly correlation map. An anomaly type identification and classification model was applied based on an anomaly association graph to obtain preliminary classification results, and a classification label with confidence weighting was constructed based on it. Based on classification labels, the severity of anomalies is assessed and the scope of impact is analyzed to obtain a risk level matrix. Anomaly information is then aggregated and structured to generate inspection anomaly identification and classification results.

8. The method for collecting and comparing inspection data based on augmented reality according to claim 7, characterized in that, The inspection anomaly identification and classification results include anomaly type, location, severity, root cause, and processing priority.

9. The method for collecting and comparing inspection data based on augmented reality according to claim 8, characterized in that, The process of obtaining the inspection anomaly handling guidelines includes: Based on the inspection anomaly identification and classification results, virtual anchor points are generated and their positions are calibrated in augmented reality space to obtain the calibrated anomaly region boundary. Based on the calibrated boundaries of the abnormal regions, visual marking design and transparency optimization are performed to obtain a hierarchical visual marking scheme. Based on this scheme, anomaly type symbols are designed and color-coded to obtain an anomaly identification system. Based on the anomaly identification system, and combined with the interactive functions of augmented reality devices, interactive response rules are designed and gesture recognition is configured to obtain an anomaly area interaction framework, and multi-level information is displayed based on it. Based on the multi-level information display results, the processing flow is visualized to obtain a visual operation guide, and a safe processing specification is obtained by embedding safe operation constraint rules and configuring risk warning strategies. Based on the safety handling specifications, the system associates anomaly handling knowledge bases and extracts expert experience, generates handling solutions and prioritizes them, and generates inspection anomaly handling guidelines.

10. The method for collecting and comparing inspection data based on augmented reality according to claim 9, characterized in that, The process of obtaining the inspection closed-loop data report includes: Based on the inspection anomaly handling guidelines, and combined with remote collaboration sessions, a collaborative space sharing framework was established for on-site view sharing and augmented reality marker synchronization. Based on the collaborative space sharing framework, expert remote marker generation and 3D spatial guidance projection are performed to obtain a remote guidance marker set. The remote guidance marker set is then used to interact with on-site operators in real time and confirm operations to obtain operation execution feedback. Based on the operation execution feedback, the processing process is recorded and key steps are screenshotted to obtain a processing process archive. Based on the processing process archive, the processing effect is evaluated and a before-and-after comparison analysis is performed to obtain processing effect verification data. Based on the verification data of the treatment effect, abnormal state repair confirmation and re-inspection data collection are carried out to obtain closed-loop verification results. The effectiveness of the treatment measures and experience are evaluated and summarized through the closed-loop verification results to obtain treatment experience knowledge items. Based on processing experience and knowledge items, reports are automatically generated and key information is extracted. Data visualization processing and multi-format export are then performed to generate closed-loop inspection data reports.

Images