A housing safety detection method and system

Abstract

The present invention relates to the technical field of housing safety monitoring, and in particular to a housing safety detection method and system. The method comprises the following steps: obtaining housing structure data, and collecting housing structure details according to the housing structure data to obtain housing structure detail data; using drone aerial photography technology to collect image data of the house to be detected to obtain housing image data, and constructing a digital twin model according to the house image data and the housing structure detail data to obtain a housing digital twin model; identifying housing structure anomalies according to the housing digital twin model to obtain housing structure anomaly data; processing housing safety risk of housing structure anomaly data to obtain housing safety risk data. The present invention can timely discover and handle these safety hazards through dangerous house management, prevent the occurrence of accidents such as dangerous house collapse and collapse, and effectively protect the lives and property of residents.

Description

Technical Field

[0001] The present invention relates to the technical field of house safety monitoring, and in particular to a house safety detection method and system. Background Art

[0002] With the acceleration of urbanization, the density of urban buildings continues to increase, and the types and complexity of buildings are also increasing. At the same time, many houses have gradually reduced their structural safety due to long-term use, disrepair or erosion by the natural environment, posing potential safety hazards. Especially in the context of frequent natural disasters such as earthquakes and typhoons, the earthquake resistance and disaster resistance of buildings are put to the test. Due to the complexity of building structures and the diversity of use environments, these problems are often difficult to discover and deal with in the early stages.

[0003] Traditional housing safety inspection methods usually rely on manual inspections, local sampling and static monitoring, which have many limitations in practical applications. First, manual inspections require professional inspectors to check the key parts of the building one by one. This method is not only time-consuming and labor-intensive, but also easily limited by subjective experience and personal ability, resulting in inconsistent test results. Secondly, although the local sampling method can provide detailed information on some key components, due to the limited detection range, it is easy to ignore potential problems in hidden areas or non-critical components, thus failing to accurately reflect the overall health of the house. In addition, static monitoring obtains data through fixed sensors or periodic detection, lacks dynamics and real-time performance, and is difficult to capture subtle changes in the structural state of the building during use, especially when disturbed by the external environment (such as strong winds or vibrations), which will miss key abnormal signals. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes a housing safety detection method and system to solve at least one of the above technical problems.

[0005] The present application provides a housing safety detection method, the method comprising:

[0006] S1. Acquire house structure data, and collect house structure details according to the house structure data to obtain house structure detail data;

[0007] S2. Use drone aerial photography technology to collect image data of the house to be inspected to obtain house image data, and construct a digital twin model based on the house image data and house structure detail data to obtain a house digital twin model;

[0008] S3. Identify house structure anomalies based on the house digital twin model to obtain house structure anomaly data;

[0009] S4. Perform housing safety risk processing on the housing structure abnormality data to obtain housing safety risk data.

[0010] In the present invention, by acquiring the basic structural data of the house and its detailed collection, the structural state of the house, including the parameters of each key component (such as beams, columns, foundations, etc.) can be fully and systematically grasped. Remote sensing data acquisition makes it possible to monitor the house from the air or from a distance, and the changes in the exterior and surrounding environment of the house can be fully obtained without contact inspection. Combined with digital twin technology, the physical and functional characteristics of the house can be reconstructed in a virtual environment, and then the performance of the house under different conditions can be simulated and predicted to identify potential risks. House structural anomaly identification can promptly detect abnormal problems (such as cracks, settlement, tilt, etc.) that occur in the house during use, and accurately locate the problem through real-time updating of the digital twin model. The slight changes in the house structure are automatically detected by intelligent algorithms to promptly discover potential risks. By processing the safety risk of abnormal data of the house structure, the safety risk of the house can be quantified and an operational risk processing report can be generated. According to the analysis results of the model, the safety risk of the house can be evaluated, warned and the deployment of response measures can be carried out in a timely manner, which can effectively prevent the occurrence of house safety accidents and ensure the safety of residents and property.

[0011] Optionally, S1 includes:

[0012] S11, obtaining building infrastructure data;

[0013] S12, identifying key structural components according to the building infrastructure data to obtain key structural component data;

[0014] S13, arrange monitoring points according to the data of key structural components to collect real-time data and obtain real-time IoT data of the house;

[0015] S14, integrating the house infrastructure data and the house real-time IoT data to obtain house structure data;

[0016] S15, extracting house structure features according to the house structure data to obtain house structure feature data;

[0017] S16, performing spatial annotation according to the house structure data to obtain house structure spatial data;

[0018] S17, performing characteristic coupling association on the house structure characteristic data and the house structure space data to obtain the house structure detail data.

[0019] In the present invention, detailed and accurate structural information is provided for subsequent housing safety analysis by acquiring housing infrastructure data (such as architectural drawings, design parameters, historical data, etc.). By analyzing the housing infrastructure data, key components (such as load-bearing walls, beams, columns, foundations, etc.) that are critical to safety in the house can be accurately identified, which helps to concentrate resources and monitoring priorities and improve detection efficiency and accuracy. According to the identified key component data, monitoring points are reasonably arranged for real-time data collection to ensure comprehensive monitoring of the housing safety status. Through scientific monitoring point arrangement, structural anomalies of the house can be discovered in time, which is convenient for risk management and emergency response. Integrate structural data and real-time Internet of Things data (such as sensor data and environmental data) to form comprehensive housing structure data, which provides multi-dimensional input data for safety analysis and risk assessment, and increases the depth and comprehensiveness of the analysis. Extract relevant features (such as structural strength, elastic modulus, seismic response, etc.) from the housing structure data, provide accurate structural attribute data for risk identification and prediction, help to identify potential structural problems in advance, and provide support for accurate modeling. Spatial annotation of the structural data of a house can clearly mark the location, shape, size, etc. of each structural component, thereby providing more detailed spatial information for risk analysis and prediction. By coupling and associating the structural characteristic data of a house with spatial data, the inherent connection between structural characteristics and spatial distribution can be revealed, potential structural weaknesses or dangerous areas can be identified, and real-time monitoring and early warning can be facilitated.

[0020] Optionally, S13 includes:

[0021] S131, arranging tilt monitoring points according to data of key structural components to collect real-time data and obtain tilt monitoring data;

[0022] S132, arranging crack monitoring points according to data of key structural components to collect real-time data and obtain crack monitoring data;

[0023] S133, arranging vibration monitoring points according to data of key structural components to collect real-time data and obtain vibration monitoring data;

[0024] S134, arranging settlement monitoring points according to the data of key structural components to collect real-time data and obtain settlement monitoring data;

[0025] S135. Integrate the tilt monitoring data, crack monitoring data, vibration monitoring data, and settlement monitoring data to obtain real-time IoT data of the house.

[0026] In the present invention, by arranging the inclination monitoring points on the key components of the structure, the inclination of the house can be accurately monitored, and the deformation or inclination problem of the house structure can be discovered in time. By arranging the crack monitoring points on the key components of the house, the changes of the tiny cracks in the house can be monitored in real time, especially on the load-bearing structural components, which helps to find the expansion trend of the cracks in the early stage and prevent the cracks from developing to the extent that endangers the safety of the house structure. By arranging the vibration monitoring points on the key components of the structure, the vibration changes of the house under different environmental conditions (such as earthquakes, wind or construction, etc.) can be monitored, and the dynamic response and seismic resistance of the house can be analyzed in real time, which helps to find the uneven force of the house or potential structural damage as early as possible. By arranging the settlement monitoring points, it is possible to detect in real time whether the important components of the foundation or structure of the house have settled. In the environment of uneven foundation settlement, soft soil foundation, etc., the settlement trend can be identified in advance, so as to take timely remedial measures to avoid cracks, inclination or damage in the house.

[0027] Optionally, S131 includes:

[0028] Carry out load distribution calculation based on the building foundation structure data and key structural component data to obtain structural load distribution data;

[0029] Material characteristics are processed according to the building foundation structure data, key structural component data and structural load distribution data to obtain material load distribution data;

[0030] Calculate the sensitivity of structural tilt change according to the material load distribution data to obtain the sensitivity data of structural tilt change;

[0031] Construct a structural diagram according to the data of key structural components to obtain structural component diagram data;

[0032] Numerical mapping is performed based on the structural tilt change sensitivity data and the structural component diagram data to obtain component tilt sensitivity matrix data;

[0033] Performing component load response processing on component tilt sensitivity matrix data to obtain tilt sensitivity matrix data;

[0034] Screen the monitoring points according to the tilt sensitivity matrix data to obtain the tilt monitoring point data;

[0035] The tilt monitoring points are arranged according to the tilt monitoring point data to collect real-time data and obtain the tilt monitoring data.

[0036] In the present invention, by calculating the load distribution of the basic structure and key components of the house, the load conditions of each component can be accurately predicted, so as to identify the weak points of the structure or the areas subject to excessive stress in advance. By processing the material properties, the material behavior of each component of the house under different loads can be accurately analyzed, and potential problems such as fatigue, deformation, and stress concentration of the material can be identified, thereby effectively predicting the structural performance and life of the house. By performing sensitivity calculation on the material load distribution, the structural tilt change caused by the load can be accurately evaluated. The sensitive parts of the structure, especially the areas where tilt or uneven settlement may occur, are identified in advance, providing forward-looking data support for house safety management. The construction of the structural diagram can associate the physical structure of the house with its function, and clearly show the relationship between the key components in a graphical way. By associating the sensitivity data of the tilt change of the structure with the structural component diagram data through numerical mapping, complex structural performance problems can be converted into numerical results that are easy to analyze. By processing the component load response, the response characteristics of the components under different loads can be analyzed. Identify which components are more sensitive to tilt changes, so as to discover potential dangerous parts in advance and formulate reasonable monitoring strategies and reinforcement measures. By analyzing the tilt sensitivity matrix data, the most critical monitoring points can be screened out, thereby achieving accurate monitoring and ensuring that the areas that are most likely to change and need the most attention are monitored in a timely manner under limited resources to avoid the generation of invalid data. By arranging the tilt monitoring points based on the selected key tilt monitoring points, the tilt of the house can be monitored in real time, especially in important parts and areas where tilt may occur, thus providing an efficient monitoring method for the safety assessment of the house.

[0037] Optionally, the component load response processing includes:

[0038] Using the component load response prediction model to process the component tilt sensitivity matrix data, the tilt sensitivity matrix data is obtained;

[0039] The steps for constructing the component load response prediction model include the following steps:

[0040] Obtaining historical component tilt sensitivity matrix data and historical component tilt sensitivity index data;

[0041] Perform load feature extraction and tilt sensitivity feature extraction based on historical component tilt sensitivity matrix data to obtain component load feature data and load tilt sensitivity feature data respectively;

[0042] Perform multi-dimensional feature fusion on component load feature data and load tilt sensitivity feature data to obtain fused feature data;

[0043] Response function modeling is performed according to component load characteristic data and load tilt sensitivity characteristic data to obtain an initial response function model;

[0044] Initialize the predicted value of the historical component tilt sensitivity index data to obtain the initial predicted result data;

[0045] Output the initial result according to the fused feature data and the initial response function model to obtain the initial result data;

[0046] Divide the data set according to the initial result data to obtain test data and verification data;

[0047] Perform splitting point splitting according to the test data to obtain splitting point data;

[0048] Construct a tree structure based on the split point data to obtain initial decision tree data;

[0049] The initial decision tree data is iteratively updated using the initial prediction result data and verification data to obtain a component load response prediction model;

[0050] The multi-dimensional feature fusion includes:

[0051] Characteristic encoding is performed on component load characteristic data and load tilt sensitivity characteristic data to obtain component load characteristic encoding data and load tilt sensitivity characteristic encoding data;

[0052] Combine the component load feature coding data and the load tilt sensitivity feature coding data to obtain preliminary feature fusion data;

[0053] Performing feature fusion decoding on the preliminary feature fusion data to obtain first feature fusion data;

[0054] A graph is constructed based on component load characteristic data and load tilt sensitivity characteristic data to obtain component load sensitivity joint graph data;

[0055] Perform graph convolution on the component load sensitivity joint graph data to obtain joint graph feature data;

[0056] Perform causal multi-head self-attention calculation on the joint graph feature data to obtain the second feature fusion data;

[0057] Perform masked multi-head self-attention calculation based on the first feature fusion data and the second feature fusion data to obtain preliminary fusion feature data;

[0058] The global context information is integrated on the preliminary fused feature data to obtain fused feature data.

[0059] In the present invention, through the accurate component load response prediction model, the response of each component of the structure under different loads can be accurately evaluated, and the hidden dangers that cause structural instability or damage can be discovered in advance. The load response prediction model can automatically identify the weak links of the structure, thereby providing accurate monitoring points. Through the multi-dimensional feature fusion of component load characteristics and tilt sensitivity characteristics, the potential correlation between multiple factors such as load, tilt, and material properties can be comprehensively considered. Compared with traditional methods, it can more comprehensively and accurately reflect the health status of the structure, avoiding misjudgment caused by the influence of a single factor. Through graph construction and convolution processing, the details of structural changes can be captured at multiple levels, especially for complex house structures, higher-dimensional feature data can be processed, and more accurate prediction results can be provided. By fusing component load feature data and tilt sensitivity feature data, cross-domain data fusion can be achieved, load response prediction can be optimized, and accurate monitoring points can be provided based on potential deep data relationships. In the feature fusion stage, through the integration of global context information, the various dimensional features of the house structure can be comprehensively analyzed. Integration can not only improve the accuracy of prediction, but also ensure that the model has stronger adaptability and generalization ability when dealing with complex structures.

[0060] Optionally, S15 includes:

[0061] Perform spatial coordinate mapping according to the house structure data to obtain the house spatial coordinate data;

[0062] Performing component feature encoding on the house space coordinate data to obtain house component feature encoding data;

[0063] Component spatial relationship modeling is performed according to the housing component feature coding data and the housing spatial coordinate data to obtain housing spatial relationship data;

[0064] The house spatial relationship data and the house component feature coding data are spatially fused to obtain the house structure spatial data.

[0065] In the present invention, by converting the house structure data into accurate spatial coordinate data, highly accurate modeling of the overall structure of the house can be achieved. Spatial coordinate mapping provides an accurate coordinate system for analysis, so that each component, monitoring point, etc. in the structure can be accurately located in three-dimensional space. Characteristic coding of house components can not only effectively extract the attribute information of each component of the house, but also provide structured coded data for data analysis. By combining the house structure characteristics with the spatial coordinate data, the obtained component characteristic coding data can provide a rich source of information for analysis. By modeling the component spatial relationship according to the house component characteristic coding data and the spatial coordinate data, the overall spatial layout information can be provided for the house structure, so that the relationship between each component can be dynamically displayed, and the interaction between the components of the house and their positional relationship in space can be analyzed. Fusion of house spatial relationship data with house component characteristic coding data can provide more comprehensive and three-dimensional data support for the house structure, especially in spatial analysis, physical space characteristics and structural characteristics can be integrated, thereby comprehensively revealing the structural behavior and performance of the building. Dynamic structural monitoring and evaluation can be achieved through real-time collection of house spatial coordinate data, component characteristic coding data and spatial relationship data. When the structure of a house changes, the rapid fusion of data can provide real-time feedback, helping to identify potential risks in a timely manner.

[0066] Optionally, S2 includes:

[0067] S21, using drone aerial photography technology to collect image data of the house to be inspected, to obtain house image data;

[0068] S22, performing three-dimensional reconstruction according to the house image data to obtain a three-dimensional reconstruction model of the house;

[0069] S23, performing point cloud processing on the three-dimensional reconstructed model of the house and meshing it to obtain a meshed model of the house;

[0070] S24, performing real-time fitting of the house grid model according to the tilt monitoring data, the crack monitoring data, and the settlement monitoring data to obtain a house fitting model;

[0071] S25. Use the house structure detail data to construct a digital twin model of the house fitting model to obtain a digital twin model of the house.

[0072] In the present invention, the image data of the house is obtained by drone aerial photography technology, and the external and surrounding environment information of the building can be captured from multiple angles in an all-round manner. After the image data is three-dimensionally reconstructed, the obtained three-dimensional reconstruction model of the house has higher spatial resolution and accuracy than the traditional manual detection method, avoiding the visual blind spots or detection omissions generated in manual inspections. The generation of the house grid model and the fitting process of real-time Internet of Things data can enable the model to be continuously updated and calibrated during actual use. Through real-time fitting, the deformation, settlement and other changes of the building can be dynamically reflected, further improving the accuracy and timeliness of the model. The construction of the digital twin model enables the physical entity of the house to be accurately reproduced in the virtual space, and the health status, various indicators and potential risks of the building can be intuitively displayed and analyzed through the model. Through three-dimensional reconstruction and real-time data update, the digital twin model of the house can promptly discover structural defects and potential risk points, promptly warn, and repair or reinforce before the problem worsens.

[0073] Optionally, S3 includes:

[0074] S31. Extract abnormal features according to the digital twin model of the house to obtain abnormal feature data of the house structure;

[0075] S32, performing abnormal classification on the abnormal characteristic data of the house structure to obtain abnormal classification data of the house structure;

[0076] S33. Perform spatial risk mapping on the digital twin model of the house based on the classified data of house structure anomalies to obtain the house structure anomaly data.

[0077] In the present invention, by extracting the abnormal features of the house structure according to the digital twin model of the house, the abnormal phenomena existing in the building, such as cracks, settlement, tilt, etc., can be quickly identified, the deviation of human subjective judgment can be eliminated, and the accuracy and consistency of detection can be improved. By performing abnormal classification on the abnormal feature data of the house structure, different types of abnormal problems can be distinguished, such as structural cracks, tilt, settlement, etc., so as to provide targeted data support for risk assessment and processing. Spatial risk mapping is performed on the abnormal classification data of the house structure, and various abnormal problems can be associated with specific spatial positions, so as to realize the spatial visualization of the abnormal structure of the house, so that the monitoring personnel can more intuitively identify the weaknesses and risk areas of the building, and take corresponding measures in time. Through the combination of abnormal classification of the house structure and spatial risk mapping, detailed abnormal data of the house structure can be generated, so as to realize accurate risk assessment, provide scientific basis for the subsequent maintenance and repair of the building, and help decision makers to formulate more effective repair plans. At the same time, through the real-time update of risk mapping, early warning can be issued in time when an abnormality occurs in the building.

[0078] Optionally, S4 includes:

[0079] S41, obtaining monitoring point location data;

[0080] S42, performing abnormal evolution simulation on the monitoring point location data and the abnormal housing structure data according to the abnormal housing structure classification data to obtain the abnormal housing structure data;

[0081] S43. Perform a safety risk assessment based on the abnormal data of the building structure to obtain building safety risk data.

[0082] In the present invention, by combining the classification data of house structure abnormalities for dynamic simulation, the time series of house structure changes can be obtained, the development trend of various abnormalities can be accurately predicted, and a scientific basis can be provided for maintenance and reinforcement. The simulation results are helpful to assess whether certain abnormal phenomena will further deteriorate, and take corresponding measures according to the evolution trend. Structural safety risk assessment can provide a detailed risk report through data analysis, covering the risk level, risk degree, risk development trend, etc. of abnormal areas. Compared with traditional manual assessment methods, this method is more objective and accurate, and can update risk assessment results in real time.

[0083] Optionally, the present application further provides a housing safety detection system for executing the housing safety detection method as described above, the housing safety detection system comprising:

[0084] The housing structure detail collection module is used to obtain housing structure data, and collect housing structure details according to the housing structure data to obtain housing structure detail data;

[0085] A digital twin model building module is used to collect image data of the house to be inspected using drone aerial photography technology to obtain house image data, and to build a digital twin model based on the house image data and house structure detail data to obtain a house digital twin model;

[0086] A housing structure anomaly recognition module is used to identify housing structure anomalies based on the housing digital twin model and obtain housing structure anomaly data;

[0087] The housing safety risk processing module is used to perform housing safety risk processing on housing structure abnormality data to obtain housing safety risk data.

[0088] The purpose of the present invention is to provide accurate house structure details information through high-resolution structural data collection, greatly improve the accuracy of subsequent structural anomaly identification, reduce omissions and errors in traditional detection, and improve the reliability of overall house safety management. By combining remote sensing technology with house structure detail data, a high-precision digital twin model is successfully constructed. The model can not only fully reflect the spatial structure of the building, but also dynamically monitor its state changes in real time, providing strong support for anomaly identification and risk assessment. The digital twin model enables the structural state of the house to be accurately simulated, thereby realizing intelligent identification of house structural anomalies. Through model analysis, potential structural problems (such as cracks, deformation, settlement, etc.) can be discovered at an early stage, and abnormal alarms can be issued in time. By performing safety risk processing on abnormal data of house structure, the present invention can realize dynamic evaluation of house safety. After abnormal data identification, the system can not only classify risks, but also predict risks based on specific structural problems and their development trends, thereby making forward-looking judgments on future safety hazards of the house. Through dynamic evaluation and safety risk analysis of house structural anomalies, the present invention can help decision makers optimize resource allocation. By accurately assessing the levels of various risks, we can prioritize high-risk areas during the maintenance process and achieve accurate maintenance and repairs, which not only saves maintenance costs but also improves the overall safety of the house. BRIEF DESCRIPTION OF THE DRAWINGS

[0089] Other features, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting implementations made with reference to the following drawings:

[0090] Figure 1 A flowchart showing a method for detecting housing safety according to an embodiment of the present invention is provided;

[0091] Figure 2 A flowchart showing a method for collecting building structure details according to an embodiment of the present invention is provided;

[0092] Figure 3 A flowchart showing a method for constructing a digital twin model according to an embodiment of the present invention is shown;

[0093] Figure 4 A flowchart showing a method for identifying abnormalities in building structure according to an embodiment is shown;

[0094] Figure 5 A flowchart showing a method for handling housing safety risks according to an embodiment of the present invention is provided;

[0095] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0096] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0097] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0098] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0099] Using laser scanners and drones, we scanned the building comprehensively and collected basic structural data of the house, including wall thickness, column beam diameter, concrete strength, etc. We collected real-time structural health monitoring data (such as tilt, vibration, cracks, etc.) through smart sensors. We obtained house data: wall thickness: 300 mm, column beam diameter: 500 mm, concrete strength: C40. Real-time data: tilt angle: 0.3° (initial), vibration frequency: 4.2 Hz, crack width: 0.8 mm.

[0100] The drone is used to obtain the external image of the building and the surrounding environment data to generate a 3D point cloud model. The point cloud data is integrated with the house details to build a digital twin model of the house. Point cloud data: Points (x, y, z, reflection intensity): Point 1: 12.5, 15.2, 8.3, 0.85, Point 2: 12.7, 15.3, 8.4, 0.87, .... Digital twin modeling is performed to generate an overall three-dimensional model of the building, including the spatial information of all key components. The digital twin model verification result shows that the size error of the comparison model is ≤ 0.5%.

[0101] Abnormal feature extraction is performed based on the digital twin model. The extracted features include: tilt angle change rate: 0.3° → 0.5° (67% increase in 5 days), crack expansion rate: 0.8 mm → 1.5 mm (87% increase in 5 days), vibration frequency offset: 4.2 Hz → 6.5 Hz (50% change), abnormal classification: Tilt: belongs to the "uneven structural settlement" risk category. Crack: belongs to the "component fatigue crack" category. Vibration: belongs to the "dynamic external force impact" category. The abnormal data is mapped to the digital twin model, and the abnormal area is marked: Structural column (No. 2 on the south side): abnormal settlement. Middle of the south wall: crack expansion. Center of the roof: abnormal vibration.

[0102] Dynamic simulation of abnormal evolution through digital twin model: If the settlement rate remains at 0.1 mm / day, the column tilt angle is expected to exceed the limit to 2° after 2 months. Crack extension rate prediction: The crack width will reach 3.2 mm after 30 days, posing a serious safety risk. Vibration frequency deviation leads to cumulative effect of structural fatigue. Comprehensive risk assessment score (by weight): Tilt risk: 40% (high), Crack risk: 35% (high), Vibration risk: 25% (medium)

[0103] Overall risk level: High risk.

[0104] See also Figures 1 to 5 The present application provides a housing safety detection method, the method comprising:

[0105] S1. Acquire house structure data, and collect house structure details according to the house structure data to obtain house structure detail data;

[0106] In one embodiment, architectural drawings and design materials of the house are collected, including floor plans, cross-sections, and construction drawings. The overall structure of the house is scanned using a laser scanning device (such as a 3D laser scanner) to obtain 3D point cloud data. Information on the building materials of the house, such as concrete strength grade, steel bar configuration data, etc., is collected through construction records or existing sensors. The point cloud data is compared with the architectural design drawings, and the beams, columns, slabs and other structures of the house are accurately located through data matching and geometric feature extraction methods. High-resolution imaging equipment (such as industrial cameras or drones) is used to take high-definition photos of the details to ensure that information such as cracks and deformations can be recorded. Combined with sensor data (such as stress and strain sensors), the state changes of key components are monitored in real time.

[0107] S2. Use drone aerial photography technology to collect image data of the house to be inspected to obtain house image data, and construct a digital twin model based on the house image data and house structure detail data to obtain a house digital twin model;

[0108] In one embodiment, a remote sensing device (such as a drone) is used to obtain an external panoramic image and spatial positioning data of the house. Multispectral imaging technology is used to collect material degradation information on the surface of the house, such as color changes, reflectivity anomalies, etc. The geometric data of the house structure is fused with the remote sensing appearance data using the collected remote sensing data and house structure detail data. The house structure is reconstructed on a three-dimensional modeling platform, including the size, spatial position, and material information of beams, columns, and plates. When establishing a dynamic simulation model, historical environmental factors of the house (such as wind loads, earthquakes, etc.) can be considered.

[0109] S3. Identify house structure anomalies based on the house digital twin model to obtain house structure anomaly data;

[0110] In one embodiment, the digital twin model is compared with the standard structural design specifications to identify anomalies in geometry, connection methods, etc., such as misalignment of beams and columns, uneven thickness of plates, etc. The crack imaging data is used in combination with the depth analysis model to determine the length, width and extension direction of the crack. The twisted and deformed parts in the point cloud model are checked and marked as potential abnormal areas. The detected anomalies are classified, such as crack problems, insufficient bearing capacity, aging of materials, etc. Each abnormal data is associated with a specific house component, such as "abnormal bearing capacity of the second floor beam component on the south side."

[0111] In one embodiment, the digital twin model is compared point by point with the standard building specifications (such as design load, bearing capacity) to check the deviation of size, strength and node connection mode. The crack data is classified and processed: Crack width analysis: The crack width value is extracted according to the high-definition image (such as through gray value change edge detection) to determine whether it exceeds the specification limit. Crack direction analysis: According to the angle between the crack direction and the main force direction, it is determined whether the crack is a dangerous crack. Analyze the abnormal area in the thermal distribution data and combine the material property data to determine whether there is material degradation (such as concrete carbonization or steel bar corrosion). Generate an abnormal data set based on the comparison results, including the type of specific abnormality (such as insufficient bearing capacity, connection detachment), location (such as beam-column connection node) and severity.

[0112] Output abnormal distribution diagram and attach abnormal cause analysis.

[0113] S4. Perform housing safety risk processing on the housing structure abnormality data to obtain housing safety risk data.

[0114] In one embodiment, based on the abnormal data, the impact of each structural abnormality on the overall safety of the house is evaluated, for example, whether cracks will lead to a decrease in the overall bearing capacity, and whether deformation will affect the seismic performance. A weighted analysis of risk factors is performed in combination with environmental data (such as surrounding geological conditions, wind speed records, etc.) to determine the safety level of the house. Based on the evaluation results, a specific risk level report is output, such as "medium risk: reinforcement is required" or "high risk: immediate evacuation of personnel is required." Repair suggestions are generated for risky components, including repair plans, lists of required materials, and construction suggestions.

[0115] In one embodiment, the risk score of each abnormal point is calculated in combination with the abnormal data. For example: Risk score = abnormal severity coefficient × impact range coefficient × weight factor. The severity coefficient is based on the physical impact of the abnormality (such as crack width), the impact range coefficient is based on the abnormal coverage area, and the weight factor considers environmental conditions (such as the seismic resistance level of the earthquake zone). Cumulative risk: Through the weighted superposition method, the overall risk level of the house is evaluated and the risk level is divided into low, medium, and high. Output the risk level of specific components and generate corresponding maintenance recommendations, such as "medium risk: reinforcement is required" or "high risk: urgent repair is required". Provide treatment plan optimization, for example: for crack abnormalities, it is recommended to use epoxy resin injection repair; for components with insufficient bearing capacity, it is recommended to add steel plate wrapping or carbon fiber reinforcement measures. Generate a complete risk report, including a risk distribution map, specific repair recommendations and a material budget list.

[0116] Optionally, S1 includes:

[0117] S11, obtaining building infrastructure data;

[0118] In one embodiment, basic data is obtained from architectural drawings and construction files of the house, including design dimensions of beams, columns, walls, etc., material strength grades, and construction process instructions.

[0119] S12, identifying key structural components according to the building infrastructure data to obtain key structural component data;

[0120] In one embodiment, geometric features (such as size and connection method) are used in point cloud data to locate key components of the house, such as beams, columns, shear walls, etc. The differences between the components recorded in the basic data and the actual scan data are compared, and components that may be abnormal are marked. Key components are divided into load-bearing structures (such as beams and columns) and non-load-bearing structures (such as infill walls), and classification labels are assigned. Specific parameters of each key component, such as the span of the beam and the cross-sectional size of the column, are extracted and compared with the design standard. A key component data table is output, which contains the location, size, and material properties of each component.

[0121] S13, arrange monitoring points according to the data of key structural components to collect real-time data and obtain real-time IoT data of the house;

[0122] In one embodiment, the location of monitoring points is selected according to the importance of key components (such as bearing capacity, stress concentration area), such as beam ends, column nodes and the middle of the wall. The importance of each component is scored using a weighted analysis method (based on stress level, deformation sensitivity), and sensors are arranged preferentially for high-scoring components. Sensors, including strain gauges, displacement sensors and temperature sensors, are arranged at selected monitoring points to capture real-time structural data. The installation position and angle of each sensor need to be optimized through field testing to ensure data accuracy. The sensor is connected to the Internet of Things platform through a wireless transmission module (such as LoRa or 5G) to upload monitoring data in real time.

[0123] S14, integrating the house infrastructure data and the house real-time IoT data to obtain house structure data;

[0124] In one embodiment, static data (such as size and material) in the structure database is combined with real-time data (such as strain and temperature) from the Internet of Things. The static and dynamic data are aligned in time axis through a time synchronization algorithm to ensure the relevance of the two. The fused house structure data is output, in which the records of each component include design parameters, real-time monitoring values, and historical status.

[0125] S15, extracting house structure features according to the house structure data to obtain house structure feature data;

[0126] In one embodiment, geometric features (such as length, width, height, and cross-sectional shape), mechanical features (such as stress distribution, deformation), and environmental features (such as temperature gradient) are extracted for each component. The global features are decomposed into regional features and component features using a hierarchical analysis method to facilitate local analysis. The extracted features are compared with the design specifications to identify parameters that exceed the standards (such as the deflection of a beam exceeding the design limit). A feature report is generated to record the features and potential anomalies of each component.

[0127] S16, performing spatial annotation according to the house structure data to obtain house structure spatial data;

[0128] In one embodiment, each key component is spatially annotated in the 3D point cloud model, and the relationship between its center point, boundary and adjacent components is recorded. Key geometric parameters (such as connection angle and center of gravity position) are annotated for complex connection components (such as beam-column nodes). The structural spatial data is output, with the 3D coordinate system as the reference, and each component is accompanied by geometric shape and position parameters.

[0129] S17, performing characteristic coupling association on the house structure characteristic data and the house structure space data to obtain the house structure detail data.

[0130] In one embodiment, the characteristic data of the building structure (such as size, stress, and deformation) are associated with the spatial data. For example, the deformation data of a beam is combined with its position in space to locate the specific deformation area. The coupling method represents the connection relationship between components through an adjacency matrix to establish a global data network. The detailed data of the building structure is output, and each record contains characteristic parameters, spatial position, and coupling characteristics (such as the relationship between the deformation of a beam and the force of its adjacent column).

[0131] Optionally, S13 includes:

[0132] S131, arranging tilt monitoring points according to data of key structural components to collect real-time data and obtain tilt monitoring data;

[0133] In one embodiment, the most suitable location for installing the inclination monitoring point is determined based on the data of key structural components, especially the components with greater stress (such as the facade, foundation, tower, etc. of a high-rise building). Tilt sensors are arranged at the parts (such as the contact points between the column and the ground, the support nodes, etc.) where the possibility of tilting is relatively high according to the empirical data, ensuring that the sensors can cover the entire stress area. According to the service life of the components, the bearing capacity calculation and the structural design specifications, the inclination angle of each component is evaluated. A high-precision inclination sensor (such as a level, a laser rangefinder, etc.) is installed to ensure that it can monitor small inclination changes in real time. The automatic calibration function of the installed equipment is used to ensure that the initial data is free of deviation. The inclination sensor monitors the slight deformation of the structure in real time, collects inclination data regularly, and transmits the data to the Internet of Things platform via a wireless network. During data collection, the device synchronizes the data through the built-in timestamp to ensure that the collected data has time consistency.

[0134] S132, arranging crack monitoring points according to data of key structural components to collect real-time data and obtain crack monitoring data;

[0135] In one embodiment, according to the structural design drawings of the house, based on empirical formulas or deep learning algorithms, the locations where cracks occur are identified, such as structural joints (connections between columns and beams, intersections of walls, etc.) and locations with large force changes. Crack sensors are arranged on key components, mainly including crack width sensors (for detecting changes in cracks) and crack sensors (for detecting the occurrence and growth of cracks). According to the structural characteristics, the location where the cracks occur (such as areas of uneven foundation settlement) is evaluated, and the appropriate sensor type is selected according to the possible crack directions (vertical, horizontal, and oblique). High-precision crack sensors are installed. The sensors generally include strain sensors and displacement sensors, which monitor the occurrence and expansion of cracks by sensing the opening changes of the cracks. Ensure that the installation position of the sensor is stable and the sensor readings accurately reflect the actual state of the cracks. The crack sensor monitors the changes in crack width in real time and transmits the data to the monitoring system in real time through a wireless communication network. The data format includes information such as crack location (based on the coordinate system), crack width, and expansion direction.

[0136] S133, arranging vibration monitoring points according to data of key structural components to collect real-time data and obtain vibration monitoring data;

[0137] In one embodiment, vibration monitoring points are selected based on key components of the structure and areas where vibration occurs (such as large machinery and areas with large traffic dynamic loads). Focus on the supports between floors, foundations, bridge structures or vibration-sensitive parts of high-rise buildings. Install acceleration sensors, velocity sensors, etc. to detect the response of the building at different frequencies. The equipment needs to have sufficient sensitivity to capture low-frequency and high-frequency vibration signals, especially in high-rise buildings, small vibrations caused by factors such as wind, earthquakes, and traffic loads. The vibration sensor regularly collects vibration data of the building structure under different loads. The data includes vibration frequency, acceleration, velocity, etc. The data is transmitted to the central database via a real-time wireless network, and the data is synchronized by timestamp to ensure data consistency.

[0138] S134, arranging settlement monitoring points according to the data of key structural components to collect real-time data and obtain settlement monitoring data;

[0139] In one embodiment, based on the foundation design and geological conditions of the house, determine the locations where there is a risk of settlement, such as places where the groundwater level fluctuates greatly, the foundation soil is loose, or the load is uneven. Settlement monitoring points are installed at these key locations, arranged around the building, in the foundation area, or in places where there is obvious stress concentration. Use settlement sensors (such as fiber optic sensors, underground settlement monitors) to monitor the settlement of the foundation or house. Accurate positioning is required during equipment installation to ensure that the equipment is synchronized with the ground to avoid the accumulation of errors. The settlement sensor collects the settlement in real time and uploads it to the data center regularly. Perform real-time analysis of the settlement data, evaluate the settlement speed and settlement amplitude, and determine whether the safety threshold has been reached.

[0140] S135. Integrate the tilt monitoring data, crack monitoring data, vibration monitoring data, and settlement monitoring data to obtain real-time IoT data of the house.

[0141] In one embodiment, various types of monitoring data are processed synchronously to ensure the temporal consistency of data from different monitoring points. When integrating data, all monitoring data are aligned by timestamp to eliminate time delay errors. Different types of monitoring data (such as inclination, cracks, vibration, and settlement) are standardized to ensure that various types of data can be analyzed uniformly. The integrated data is stored in a cloud database, and a unified data report is generated, which includes various types of monitoring data for each component (such as crack width, inclination change, vibration amplitude, and settlement). The data platform displays real-time monitoring results through a graphical interface and generates real-time monitoring curves (such as the relationship curve between inclination angle and time, and the change trend of crack width and time). Based on the integrated data, analysis is performed to determine whether there are any abnormal changes. For example, if the crack width exceeds the safety standard value, the system will issue an alarm; if the inclination angle exceeds the predetermined threshold, the warning program will be immediately started. The system will automatically generate a risk assessment report and provide monitoring suggestions and repair measures.

[0142] Optionally, S131 includes:

[0143] Carry out load distribution calculation based on the building foundation structure data and key structural component data to obtain structural load distribution data;

[0144] In one embodiment, the deadweight distribution of each component is calculated based on the building infrastructure data and the geometric parameters and material properties (such as density and elastic modulus) of key components. According to the design specifications, the external loads (such as personnel, equipment, wind force, etc.) in the building use scenario are distributed and modeled, and the distribution of live loads is based on the form of area-uniformly distributed loads. The results of dead loads and live loads are superimposed to form structural load distribution data, and the distribution positions of each load in the house (such as beams, columns, plates, etc.) are marked.

[0145] Material characteristics are processed according to the building foundation structure data, key structural component data and structural load distribution data to obtain material load distribution data;

[0146] In one embodiment, the structural load is combined with the material properties to calculate the mechanical response characteristics (such as stress and strain distribution) of each component under the load, and output the material load distribution data. The material properties (such as elastic modulus, Poisson's ratio, tensile strength, etc.) of each component are extracted from the building foundation structure data. Based on the load distribution data, the stress distribution of each component is calculated using the finite element analysis method. The strain distribution of the component is calculated using the stress-strain relationship to obtain the material load distribution data.

[0147] Calculate the sensitivity of structural tilt change according to the material load distribution data to obtain the sensitivity data of structural tilt change;

[0148] In one embodiment, the tilt change sensitivity of the housing structure under load is calculated to evaluate the stability of the structure. Based on the material load distribution data, the deformation and displacement of the components under load are calculated, and the overall tilt change of the structure is estimated using the rigid body rotation theory. By analyzing the amplitude of the tilt change under different load conditions (such as a combination of dead load and live load), the tilt sensitivity index of the structure (such as the degree of response to load changes) is evaluated. The tilt change sensitivity of each key component is recorded to form the structure tilt change sensitivity data.

[0149] Construct a structural diagram according to the data of key structural components to obtain structural component diagram data;

[0150] In one embodiment, a graph model is constructed to represent the relationship between key structural components (such as connection, force transmission, etc.), providing a topological structure basis for the construction of the sensitivity matrix. Each key structural component (such as beams, columns, walls, etc.) is defined as a node of the graph, and the node attributes include the geometric parameters, material properties and load distribution data of the component. According to the physical connection relationship of the components (such as the intersection of beams and columns, the intersection of walls, etc.), the edges between the nodes are defined, and the weights of the edges can be assigned according to the force transmission relationship (such as torque size, connection stiffness, etc.). The connection relationship between the components is represented by the adjacency matrix, and the obtained structural component graph data is a graph model, which contains the attributes (nodes) and connection relationships (edges) of the components.

[0151] Numerical mapping is performed based on the structural tilt change sensitivity data and the structural component diagram data to obtain component tilt sensitivity matrix data;

[0152] In one embodiment, the tilt change sensitivity data and the structural component graph data are used to map the sensitivity information of the components to the graph model of the entire house to form a tilt sensitivity matrix. The tilt change sensitivity of each component is used as the attribute value of the node and assigned to the structural component graph. The comprehensive sensitivity of the node and its neighboring nodes is calculated through the graph model to generate a tilt sensitivity matrix.

[0153] Performing component load response processing on component tilt sensitivity matrix data to obtain tilt sensitivity matrix data;

[0154] In one embodiment, the tilt sensitivity matrix is ​​processed according to the load response prediction model to predict the tilt change of each component under different load conditions. The component load response model (the model is constructed by an empirical formula, such as a regression function, or a deep learning algorithm) is used to input the tilt sensitivity matrix data to predict the tilt response of each component under different load conditions. The prediction results are updated to the sensitivity matrix to generate the final tilt sensitivity matrix.

[0155] Screen the monitoring points according to the tilt sensitivity matrix data to obtain the tilt monitoring point data;

[0156] In one embodiment, according to the tilt sensitivity matrix data, the structural components that need to be monitored are screened out, and the monitoring point locations are determined. The monitoring point locations are screened according to the high-risk components (such as components with higher sensitivity values) in the sensitivity matrix. The tilt monitoring point data containing the highly sensitive components are output.

[0157] The tilt monitoring points are arranged according to the tilt monitoring point data to collect real-time data and obtain the tilt monitoring data.

[0158] In one embodiment, a tilt sensor is arranged on a highly sensitive component to collect tilt data in real time. The tilt sensor is installed at the tilt monitoring point, and debugged and calibrated. The data is collected in real time and uploaded to the Internet of Things platform to obtain tilt monitoring data.

[0159] Optionally, the component load response processing includes:

[0160] Using the component load response prediction model to process the component tilt sensitivity matrix data, the tilt sensitivity matrix data is obtained;

[0161] In one embodiment,

[0162] The steps for constructing the component load response prediction model include the following steps:

[0163] Obtaining historical component tilt sensitivity matrix data and historical component tilt sensitivity index data;

[0164] In one embodiment, the tilt sensitivity matrix data of each component is obtained from the historical monitoring data, the data includes the relationship between the load response and the tilt change of the structural component, and the data structure is a multi-dimensional matrix, which records the tilt change of different monitoring points under different load conditions. The tilt sensitivity index of the historical component is obtained, and the index is usually obtained by analyzing the historical data, which reflects the tilt degree and sensitivity of the component under a specific load.

[0165] Perform load feature extraction and tilt sensitivity feature extraction based on historical component tilt sensitivity matrix data to obtain component load feature data and load tilt sensitivity feature data respectively;

[0166] In one embodiment, the load response characteristics of each component are extracted based on historical data. The load response characteristics include the stress, deformation, response frequency, etc. of each component under different load conditions. For each component, based on its position and load state, the distribution characteristics of the load and the law of load intensity change are calculated. The tilt sensitivity characteristics of the component are extracted, mainly including the tilt response data of the component under different load conditions. The characteristics are obtained by statistically analyzing the tilt sensitivity matrix data of the component, such as the maximum tilt angle, tilt rate, etc.

[0167] Perform multi-dimensional feature fusion on component load feature data and load tilt sensitivity feature data to obtain fused feature data;

[0168] In one embodiment, the load feature data is fused with the tilt sensitivity feature data, such as weighted fusion or splicing fusion, that is, the load data and the tilt sensitivity data are directly combined to form a new feature vector, or a weighted average is used. In the feature fusion process, each feature is given a different weight, and the weight can be determined by correlation analysis of training data to ensure that the fused feature can accurately reflect the load response characteristics of the component.

[0169] Response function modeling is performed according to component load characteristic data and load tilt sensitivity characteristic data to obtain an initial response function model;

[0170] In one embodiment, an initial response function is constructed based on the load characteristics and the tilt sensitivity characteristics. The response function is used to describe the tilt response relationship of the component under different load conditions. The function is a nonlinear function and is modeled by regression analysis. The response function is ,in Represents the fused feature data. The output of the response function is the tilt response value of the component. Use historical data to train the response function, and use the least squares method or other optimization methods to fit the best response function coefficients. Verify the accuracy of the model, calculate the training error, and perform preliminary optimization on the response function model.

[0171] In one embodiment, the tilt response is determined by the load and sensitivity characteristics, and the basic form of the response function is:

[0172] ;

[0173] in is the tilt angle, which represents the tilt response of the component under load and is the output variable of the response function. is the load influence coefficient, indicating the load The contribution weight to the tilt response, is the load size, which refers to the external force applied to the component (such as uniform load or concentrated load), and is the main external factor affecting the tilt response. is the influence coefficient of the material elastic modulus, which indicates the material elastic modulus The weight of the tilt response, is the elastic modulus, which is a mechanical property parameter of the material and reflects the rigidity of the material when subjected to stress. is the aspect ratio influence coefficient, which represents the weight of the component’s geometric aspect ratio on the tilt response. is the component height, which refers to the vertical dimension of the component. is the component width, which refers to the horizontal cross-sectional dimension of the component. is the connection stiffness influence coefficient, which indicates the connection stiffness of the components The weight of the tilt response, is the connection stiffness coefficient, is the interaction coefficient between load and geometric characteristics, indicating the load The interaction term with aspect ratio weighs the tilt response.

[0174] Initialize the predicted value of the historical component tilt sensitivity index data to obtain the initial predicted result data;

[0175] In one embodiment, the initial response function model is used to predict the historical component tilt sensitivity index to obtain a preliminary prediction result, which is used as the starting point for model optimization.

[0176] Output the initial result according to the fused feature data and the initial response function model to obtain the initial result data;

[0177] In one embodiment, the tilt response of each component under different loads is calculated by fusion feature data and an initial response function model to obtain an initial prediction result.

[0178] Divide the data set according to the initial result data to obtain test data and verification data;

[0179] In one embodiment, the data set is divided into a training set, a validation set, and a test set according to the initial result data. A division method of 70% for training, 15% for validation, and 15% for testing is generally adopted. The data distribution of the training set, validation set, and test set is ensured to be consistent to avoid excessive deviation in the data set division, which may lead to unbalanced model training.

[0180] Perform splitting point splitting according to the test data to obtain splitting point data;

[0181] In one embodiment, when constructing a prediction model, a split point is selected for data segmentation by a greedy algorithm or based on information gain (ID3 algorithm). The split point is selected based on the critical value of certain characteristic values, such as finding the optimal split point on the load feature or the tilt sensitivity feature. The goal of split point selection is to maximize the uniformity of the split data group on the response function, that is, to minimize the prediction error of each subset. A recursive splitting algorithm (such as a binary method based on load features and tilt sensitivity features) is used to partition the data and find the best split point.

[0182] Construct a tree structure based on the split point data to obtain initial decision tree data;

[0183] In one embodiment, a preliminary decision tree is constructed using split point data. Each node of the decision tree represents a decision condition for a feature value, and each leaf node corresponds to a predicted value. The hierarchical structure of the tree is established by recursively dividing the data layer by layer. The root node of the tree is the entire data set, the split point is the internal node, and the predicted result is the leaf node.

[0184] The initial decision tree data is iteratively updated using the initial prediction result data and verification data to obtain a component load response prediction model;

[0185] In one embodiment, the initial prediction result data and the verification data are input into the decision tree, and the tree structure is iteratively updated through the error back propagation algorithm to optimize the split points and nodes. During each iterative update, the tree structure is adjusted to minimize the prediction error and improve the accuracy of the model. The performance of the decision tree model is evaluated by cross-validation and other techniques to avoid overfitting. The model is fine-tuned according to the verification data, and finally a more accurate component load response prediction model is obtained.

[0186] The multi-dimensional feature fusion includes:

[0187] Characteristic encoding is performed on component load characteristic data and load tilt sensitivity characteristic data to obtain component load characteristic encoding data and load tilt sensitivity characteristic encoding data;

[0188] In one embodiment, the original component load feature data and load tilt sensitivity feature data are converted into a coding form that can be input into a neural network or other machine learning model. The purpose of feature coding is to convert the original high-dimensional or discrete features into a dense continuous value representation. For the component load feature data, a standard numerical coding method is used. For example, the continuous load feature data (such as stress, strain, moment, etc.) is directly normalized to the interval of 0 to 1, or a regularization method is applied to make the data in the same dimension. For the load tilt sensitivity feature data, a similar method is used to first standardize the different sensitivity values, and then use category encoding (such as One-Hot Encoding) to process the discrete data. The result will be component load feature encoding data and load tilt sensitivity feature encoding data, respectively, and these two types of data will be further processed for feature fusion.

[0189] Combine the component load feature coding data and the load tilt sensitivity feature coding data to obtain preliminary feature fusion data;

[0190] In one embodiment, a unified feature vector is generated by combining the component load feature and the load tilt sensitivity feature to provide global information for subsequent processing. The component load feature encoding data and the load tilt sensitivity feature encoding data are concatenated or weighted added to form a higher dimensional feature vector.

[0191] Performing feature fusion decoding on the preliminary feature fusion data to obtain first feature fusion data;

[0192] In one embodiment, the purpose of feature fusion decoding is to recover or extract high-level abstract features from the preliminary fusion data, which are helpful to capture more patterns and associations. The preliminary feature fusion data is processed using a decoder (such as a multi-layer perceptron MLP, a back propagation network, or a fully connected neural network) (such as performing nonlinear mapping to learn a high-order representation of the feature) to generate a new feature representation, which is called the first feature fusion data.

[0193] A graph is constructed based on component load characteristic data and load tilt sensitivity characteristic data to obtain component load sensitivity joint graph data;

[0194] In one embodiment, by constructing a graph model, component load characteristics and load tilt sensitivity characteristics data are associated to capture the relationship and mutual influence between different components. The component load characteristic data and the load tilt sensitivity characteristic data are regarded as nodes of the graph. The edges between the nodes represent the impact of the load on the component tilt or other relationships, which are constructed based on physical models, correlations of historical data or domain knowledge. A weighted graph is constructed, in which each node is characterized by the load and tilt sensitivity data of the component, and the weight of the edge represents the strength of the relationship between the nodes. Generate component load sensitivity joint graph data, which is a graph with node features and edge weights for further graph convolution operations.

[0195] Perform graph convolution on the component load sensitivity joint graph data to obtain joint graph feature data;

[0196] In one embodiment, graph convolution aims to learn high-order features of nodes in the graph by propagating features of adjacent nodes, thereby capturing dependencies and interactions between components. For each node, information of neighboring nodes is aggregated through convolution operations, such as through graph convolutional networks (GCNs), that is, information is propagated through features of the node neighborhood: , is the joint graph feature data, is the activation function, is the convolution kernel weight, is the adjacency matrix of the graph, is the feature of the node, is the bias term. Through graph convolution, we get the joint graph feature data, which contains the high-level feature information of the node and its neighborhood.

[0197] Perform causal multi-head self-attention calculation on the joint graph feature data to obtain the second feature fusion data;

[0198] In one embodiment, a causal mask matrix is ​​introduced on the basis of multi-head self-attention calculation to limit the attention range and adopt a similar mask multi-head self-attention calculation mechanism. If we can only focus on our own and previous data points, the causal mask matrix is:

[0199] ;

[0200] Indicates that the current point can focus on the previous data point, and the attention score is calculated normally; Indicates that future points are unreachable and the attention score is set to negative infinity. When calculating the attention score, the mask matrix is ​​added to the attention weight. The mask ensures the causal direction.

[0201] In one embodiment, the attention mechanism is helped to capture the causal direction by introducing additional time step features or causal encodings to the input data. A time step feature or event sequence number is added to each input vector as an additional dimension. Short-term dependencies (local causality) and long-term dependencies (global causality) are calculated separately on different attention heads. Local causality is captured by focusing only on adjacent historical data points through masks. Different scaling mechanisms are used to make the attention mechanism focus on long-term causal chains.

[0202] Perform masked multi-head self-attention calculation based on the first feature fusion data and the second feature fusion data to obtain preliminary fusion feature data;

[0203] In one embodiment, a masking mechanism is used to hide certain information in the self-attention calculation (e.g., to prevent future information leakage), so that the model can better model local and sequential dependencies. When performing multi-head self-attention calculations, the masking mechanism will mask certain information to prevent the model from using irrelevant node information during processing. The calculation of masked self-attention achieves local and sequential dependency modeling by applying mask values ​​to the attention matrix.

[0204] ;

[0205] in, To initially fuse feature data, is the normalized equation, is the query matrix, is the key matrix, is the transpose symbol, is the mask value, is the feature dimension, is a value matrix.

[0206] The global context information is integrated on the preliminary fused feature data to obtain fused feature data.

[0207] In one embodiment, the preliminary fused feature data is further integrated using a global context integration method (such as a fully connected network, attention weighted merging, etc.) to ensure that the model can understand the dependencies between all features from a global perspective. The preliminary fused feature data is weighted and aggregated, or fully connected through a multi-layer perceptron to output fused feature data that incorporates all local information and global context.

[0208] Optionally, S15 includes:

[0209] Perform spatial coordinate mapping according to the house structure data to obtain the house spatial coordinate data;

[0210] In one embodiment, the purpose of spatial coordinate mapping is to determine the position of each component or structural element of the house in three-dimensional space, and provide basic data for component relationship modeling and feature extraction. The house structure data already contains the physical properties of the house components (such as component type, size, material, etc.) and their relative positions in the house. First, the spatial position of each component is calibrated according to the design drawings or scan data of the house (such as lidar scans, 3D modeling data, etc.). Usually, the spatial coordinates of the house use a Cartesian coordinate system (x, y, z) to represent the position of each component in the house. Through the spatial coordinate conversion method, combined with the global positioning system (GPS) data or other known reference points, the local coordinates of the component can be mapped to the global coordinate system. For each house component, its spatial coordinates (x, y, z) can be directly extracted from the scan data or design data, or obtained by a spatial calculation method (such as coordinate conversion, interpolation, etc.).

[0211] Performing component feature encoding on the house space coordinate data to obtain house component feature encoding data;

[0212] In one embodiment, the spatial coordinate data (such as position) of each component of the house needs to be combined with other structural features (such as material, purpose, etc.) to fully reflect the spatial characteristics of the house. Therefore, feature coding can convert this information into a format that can be processed by the model. The spatial coordinate data of each house component is subjected to component feature extraction, which involves extracting physical properties (such as material, size, weight, etc.) from the house structure data. The properties can be directly obtained from the design drawing or on-site detection data. Next, feature coding is performed on each component. For example, the characteristics of each component (such as material, functional type, etc.) are converted into numerical form using methods such as numerical coding, One-Hot coding, and embedded coding. For continuous data (such as size, weight, strength, etc.), standardization or normalization can be performed directly; for categorical data (such as component type), One-Hot coding or label coding methods can be used. For spatial coordinate data (x, y, z), it can be mapped to a unified coordinate scale through a standardization method, so that the spatial position data of different components have consistent dimensions when feature coding. The obtained house component feature coding data is a multidimensional feature vector, which contains the spatial coordinates of each component and the coding information of other related physical features.

[0213] Component spatial relationship modeling is performed according to the housing component feature coding data and the housing spatial coordinate data to obtain housing spatial relationship data;

[0214] In one embodiment, the spatial coordinates and physical characteristics of the house components are modeled to capture the spatial relationships (such as adjacent relationships, load-bearing relationships, connection relationships, etc.) between the components. First, the spatial adjacency relationship between the components is considered, that is, which components are close to each other or connected to each other in space. The adjacency relationship can be defined by calculating the Euclidean distance or Manhattan distance between the components. For each pair of adjacent components, the relationship between them can be represented by a connection matrix. On this basis, combined with the house component feature coding data, their spatial relationship can be calculated for each pair of connected components. The relationship feature vector between the components is obtained by splicing, weighted summation or other combination operations of the component feature vectors.

[0215] The house spatial relationship data and the house component feature coding data are spatially fused to obtain the house structure spatial data.

[0216] In one embodiment, the house spatial relationship data and the house component feature coding data are fused. This can be accomplished by weighted summation, splicing or other fusion methods. For example, the splicing operation can directly connect the component feature vector and the spatial relationship vector. Through spatial data fusion, the spatial features of the component are combined with the relationship information of the adjacent components to obtain a more comprehensive house structure spatial data.

[0217] Optionally, S2 includes:

[0218] S21, using drone aerial photography technology to collect image data of the house to be inspected, to obtain house image data;

[0219] In one embodiment, a drone is used for high-altitude aerial photography to collect image data of the house to be inspected and its surrounding environment. Aerial photography by drones usually uses high-resolution cameras or multi-spectral cameras to ensure that the image is clear and covers the entire building. According to the specific location of the house, the flight route and shooting angle of the drone are set to ensure that the captured image can cover all key structures. Orthophoto photography and oblique photography are usually performed. After the shooting is completed, the image data will be transmitted to the computer system via a wireless network to form a set of high-resolution house image data.

[0220] S22, performing three-dimensional reconstruction according to the house image data to obtain a three-dimensional reconstruction model of the house;

[0221] In one embodiment, key feature points are extracted from multiple images taken by a drone. Feature points are unique graphics (such as corners, edges, etc.) that can be recognized at different viewing angles. Images from different viewing angles are registered using a feature point matching algorithm (such as SIFT, SURF, or ORB). Then, these image data are merged using stereo vision technology to obtain preliminary three-dimensional point cloud data. The image data is converted into a three-dimensional reconstruction model using techniques such as structured light, stereo vision, or multi-view geometry. The point cloud is spatially parsed using a three-dimensional geometry algorithm to restore the three-dimensional structure of the building. The preliminary three-dimensional model data of the house is obtained, which is a point cloud data structure that represents the shape and external structure of the house.

[0222] S23, performing point cloud processing on the three-dimensional reconstructed model of the house and meshing it to obtain a meshed model of the house;

[0223] In one embodiment, the three-dimensional point cloud is preprocessed, including denoising, hole filling and data cleaning. Noise mainly comes from incomplete areas in the image or environmental factors (such as changes in lighting). Three-dimensional reconstruction is to accurately align point clouds with multiple data sources obtained through multiple aerial photography to ensure that point clouds from different data sources can be seamlessly connected. The point cloud data is gridded, such as using Delaunay triangulation or Poisson reconstruction algorithm to convert the point cloud into a three-dimensional mesh model. This step will make the three-dimensional model no longer scattered points, but constitute a continuous patch. A complete three-dimensional mesh model is obtained, including the facade and structure of the house.

[0224] S24, performing real-time fitting of the house grid model according to the tilt monitoring data, the crack monitoring data, and the settlement monitoring data to obtain a house fitting model;

[0225] In one embodiment, real-time data of each monitoring point of the house is obtained, including tilt monitoring data, crack monitoring data and settlement monitoring data. The data is collected in real time by embedded sensors (such as tilt sensors, crack sensors, and settlement sensors). The real-time monitoring data is aligned with the corresponding part of the three-dimensional grid model to ensure that the data of each sensor accurately corresponds to the spatial position of the grid model. The data of each monitoring point is numerically fitted. For tilt monitoring data, an elastic body model can be used to simulate the tilt behavior of the house; for crack monitoring data, a crack extension model can be used to analyze the expansion trend of cracks; and for settlement data, a foundation settlement model is used for analysis. These data are reflected on the grid model, and the geometric shape of the grid model is adjusted to make it as consistent as possible with the data fed back by the sensor. Through the feedback of real-time data, the grid model is dynamically updated to achieve real-time tracking of the house status and ensure that the model is fitted with changes over time. A fitting model of the house is obtained, which is more accurate than the previous grid model and reflects the real-time status of the house.

[0226] S25. Use the house structure detail data to construct a digital twin model of the house fitting model to obtain a digital twin model of the house.

[0227] In one embodiment, the three-dimensional point cloud data of the house is fused with the structural detail data to form a complete spatial data description. The key points (such as nodes, edges, etc.) in the point cloud data and the house structural detail data are identified using feature extraction technology to establish a corresponding relationship between the point cloud and the structural data. The two sets of data are aligned using the rigid body transformation formula:

[0228] ;

[0229] in is the point cloud coordinate after alignment, is the rotation matrix, is the original point cloud data, For the translation vector, R and T are solved by an optimization algorithm (such as the least squares method) to minimize the matching error between the point cloud data and the structural data. The geometric information in the point cloud data is combined with the physical properties (such as materials, connection relationships, etc.) in the structural detail data to generate a three-dimensional model containing geometric and attribute information.

[0230] Optionally, S3 includes:

[0231] S31. Extract abnormal features according to the digital twin model of the house to obtain abnormal feature data of the house structure;

[0232] In one embodiment, first, according to the characteristics of the house structure, key indicators that may be abnormal are defined. These indicators include, but are not limited to: component displacement, crack extension, settlement rate, vibration anomaly, etc. The actual measured values ​​of these abnormal indicators are obtained by extracting the real-time monitoring data in the digital twin model (such as sensor data from the Internet of Things device). These data may include the tilt angle of the component, crack width, settlement, vibration frequency, etc. The above raw data are subjected to statistical analysis (such as mean, standard deviation, etc.), time series analysis (such as fluctuation, trend), spectrum analysis (such as vibration frequency analysis), etc., to extract features related to abnormal house structure. For example, if the inclination of a component exceeds the normal threshold, it may indicate deformation of the structure, and then become an abnormal feature. For multi-source data (such as spatial data, sensor data, external environment data, etc.), data from different sources are integrated through feature fusion technology to extract comprehensive features that can reflect structural anomalies, such as weighted average, principal component analysis (PCA), autoencoder and other technologies.

[0233] S32, performing abnormal classification on the abnormal characteristic data of the house structure to obtain abnormal classification data of the house structure;

[0234] In one embodiment, a classification label is defined according to the type of abnormal structure of the house. The abnormal types include: tilt abnormality, crack abnormality, vibration abnormality, settlement abnormality, etc., which are defined by expert experience or historical data. The abnormal feature data of the house structure is standardized and normalized to eliminate the influence of different feature scales and ensure that the classification model can learn effectively. A classification method based on supervised learning is used, such as decision tree, support vector machine (SVM), random forest, etc. After selecting the model, the abnormal classification model is fitted by training historical data. During the training process, the abnormal feature data of the house is used as input and the abnormal type (label) is used as output for model training. The most discriminative features are selected by methods such as information gain and chi-square test to improve the classification accuracy. The training data is cross-validated and the model parameters are optimized to improve the accuracy of classification. The abnormal feature data of each component or structural part is predicted by the classification model to obtain the abnormal classification data of the house structure. The classification result of each component or structural element can be a label indicating which abnormal type it belongs to (such as tilt abnormality, crack abnormality, etc.).

[0235] S33. Perform spatial risk mapping on the digital twin model of the house based on the classified data of house structure anomalies to obtain the house structure anomaly data.

[0236] In one embodiment, the spatial coordinates of each component of the house are combined with the abnormal classification data. Through the coordinate system mapping method, the spatial position (such as x, y, z coordinates) of each component in the house (such as walls, beams, roofs, etc.) is mapped to a three-dimensional space or a two-dimensional plane. According to the spatial position of the component and its abnormal classification result, the abnormal data of the house structure is spatially aggregated. For each spatial unit (such as different floors, areas or structural components of the house), the abnormal types in the area are summarized. The abnormal distribution of each area can be presented in the form of heat maps, risk maps, 3D models, etc. For example, if there are more crack anomalies in a certain area, the risk mapping value of the area will be higher and marked in red. The risk level is calculated for each spatial unit or component. The risk level can be weighted based on factors such as the severity of the anomaly type and the frequency of the anomaly. For example, for components with multiple crack anomalies and larger crack widths, a higher risk score is given. The risk score is combined with the spatial coordinate data of the house to generate a visual abnormal risk map of the house structure. The map can be presented in the form of heat maps, three-dimensional maps, etc. to help identify high-risk areas.

[0237] Optionally, S4 includes:

[0238] S41, obtaining monitoring point location data;

[0239] In one embodiment, by obtaining the location data of monitoring points in the house structure, a spatial basis is provided for subsequent abnormal evolution simulation. Monitoring points are usually installed in key structural parts to collect real-time data, such as tilt, cracks, settlement, etc. According to the structural design and engineering requirements of the house, monitoring points will be installed in key locations where abnormalities such as deformation, cracks, settlement, etc. occur. Monitoring points include: Tilt monitoring points: used to monitor the tilt angle of the house structure. Crack monitoring points: used to detect the crack width and expansion trend of walls or beams. Settlement monitoring points: used to monitor foundation settlement or height differences between house components. Vibration monitoring points: used to monitor structural vibration. The location of each monitoring point should be accurately marked, and the coordinate data of the monitoring point should be obtained by means of global positioning system (GPS), laser scanning technology, three-dimensional laser scanning (LiDAR), etc., and a monitoring point location database should be established. These data include the three-dimensional coordinates of each monitoring point and are associated with the corresponding monitoring data type (such as tilt, cracks, etc.). The coordinate data of the monitoring point is matched with the three-dimensional digital twin model of the house to ensure that the monitoring point data is accurately reflected in the actual building structure.

[0240] S42, performing abnormal evolution simulation on the monitoring point location data and the abnormal housing structure data according to the abnormal housing structure classification data to obtain the abnormal housing structure data;

[0241] In one embodiment, the evolution of structural abnormalities of a house is simulated to predict its future development trend and changes, thereby assessing the risks brought by structural abnormalities. First, an abnormal evolution model is constructed using the abnormal classification data of the house structure and historical monitoring data. The model aims to simulate the dynamic evolution process of abnormal characteristics (such as crack extension, settlement rate, tilt change, etc.) over time and environmental changes. The evolution model is constructed based on empirical formulas, physical models or data-driven methods (such as time series prediction models). The model needs to consider the following key factors: Environmental impact: Environmental factors such as temperature, humidity, and earthquakes may cause accelerated degradation of the structure. Use load: The load borne by the house, such as household activities, external impacts, etc., may also have an impact on the structure. Material aging: Long-term use may cause aging of building materials, thereby affecting the stability of the structure. Abnormal evolution prediction: Based on the abnormal evolution model, combined with the real-time data of the monitoring point, the abnormal evolution of the house structure in the future period of time is predicted. For example, the crack width increases over time, and the settlement rate may be accelerated due to uneven settlement of the foundation. By combining the monitoring point location data with the house structure abnormality data, the future abnormal development trend of each monitoring point can be calculated. Through historical monitoring data, various abnormalities of the house structure (such as crack width, settlement, tilt angle, etc.) are modeled in time series to obtain the predicted values ​​of each monitoring point in the future. Time series prediction methods can include ARIMA, long short-term memory network (LSTM), etc. By taking environmental factors and house use load data as additional inputs, the abnormal evolution model is further adjusted to obtain more accurate prediction results.

[0242] S43. Perform a safety risk assessment based on the abnormal data of the building structure to obtain building safety risk data.

[0243] In one embodiment, a safety risk assessment standard is defined based on the characteristics of the house structure and abnormal data. The risk assessment standard includes: abnormal severity: the type and severity of the abnormality (such as crack width greater than a certain threshold, settlement rate exceeding a predetermined value, etc.). Abnormal expansion speed: the development speed of the abnormality (such as crack expansion speed, tilt change rate, etc.). Structural bearing capacity: whether the current bearing capacity of the house structure can withstand the existing abnormality. According to the type and severity of the abnormal data, a risk quantification model is constructed through a weight calculation formula to calculate the risk level of each part of the house. The risk score can comprehensively consider multiple factors, such as the type, duration, and expansion trend of the abnormality. The risk model is calibrated through simulation experiments or historical accident data to ensure the accuracy and reliability of the risk assessment results. For example, based on historical data, the weight values ​​of abnormal types such as cracks and settlement are adjusted. The risk level of each monitoring point is combined with the spatial location data of the house to generate a spatial risk distribution map of the house. This can intuitively show which areas in the house are at higher risk, helping engineers and decision makers to make more accurate processing in actual operations.

[0244] Optionally, the present application further provides a housing safety detection system for executing the housing safety detection method as described above, the housing safety detection system comprising:

[0245] The housing structure detail collection module is used to obtain housing structure data, and collect housing structure details according to the housing structure data to obtain housing structure detail data;

[0246] A digital twin model building module is used to collect image data of the house to be inspected using drone aerial photography technology to obtain house image data, and to build a digital twin model based on the house image data and house structure detail data to obtain a house digital twin model;

[0247] A housing structure anomaly recognition module is used to identify housing structure anomalies based on the housing digital twin model and obtain housing structure anomaly data;

[0248] The housing safety risk processing module is used to perform housing safety risk processing on housing structure abnormality data to obtain housing safety risk data.

[0249] The purpose of the present invention is to provide accurate house structure details information through high-resolution structural data collection, greatly improve the accuracy of subsequent structural anomaly identification, reduce omissions and errors in traditional detection, and improve the reliability of overall house safety management. By combining remote sensing technology with house structure detail data, a high-precision digital twin model is successfully constructed. The model can not only fully reflect the spatial structure of the building, but also dynamically monitor its state changes in real time, providing strong support for anomaly identification and risk assessment. The digital twin model enables the structural state of the house to be accurately simulated, thereby realizing intelligent identification of house structural anomalies. Through model analysis, potential structural problems (such as cracks, deformation, settlement, etc.) can be discovered at an early stage, and abnormal alarms can be issued in time. By performing safety risk processing on abnormal data of house structure, the present invention can realize dynamic evaluation of house safety. After abnormal data identification, the system can not only classify risks, but also predict risks based on specific structural problems and their development trends, thereby making forward-looking judgments on future safety hazards of the house. Through dynamic evaluation and safety risk analysis of house structural anomalies, the present invention can help decision makers optimize resource allocation. By accurately assessing the levels of various risks, we can prioritize high-risk areas during the maintenance process and achieve accurate maintenance and repairs, which not only saves maintenance costs but also improves the overall safety of the house.

[0250] Therefore, from any point of view, the embodiments should be regarded as illustrative and non-restrictive, and the scope of the present invention is limited by the attached application documents rather than the above description, and it is intended that all changes falling within the meaning and scope of equivalent elements of the application documents are included in the present invention.

[0251] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A housing safety detection method, characterized in that: The method comprises: S1. Acquire house structure data, and collect house structure details according to the house structure data to obtain house structure detail data; S2. Use drone aerial photography technology to collect image data of the house to be inspected to obtain house image data, and construct a digital twin model based on the house image data and house structure detail data to obtain a house digital twin model; S3. Identify house structure anomalies based on the house digital twin model to obtain house structure anomaly data; S4, performing housing safety risk processing on the housing structure abnormality data to obtain housing safety risk data; S1 includes: S11, obtaining building infrastructure data; S12, identifying key structural components according to the building infrastructure data to obtain key structural component data; S13, arrange monitoring points according to the data of key structural components to collect real-time data and obtain real-time IoT data of the house; S14, integrating the house infrastructure data and the house real-time IoT data to obtain house structure data; S15, extracting house structure features according to the house structure data to obtain house structure feature data; S16, performing spatial annotation according to the house structure data to obtain house structure spatial data; S17, performing characteristic coupling association on the house structure characteristic data and the house structure space data to obtain the house structure detail data; S13 includes: S131, arranging tilt monitoring points according to data of key structural components to collect real-time data and obtain tilt monitoring data; S132, arranging crack monitoring points according to data of key structural components to collect real-time data and obtain crack monitoring data; S133, arranging vibration monitoring points according to data of key structural components to collect real-time data and obtain vibration monitoring data; S134, arranging settlement monitoring points according to the data of key structural components to collect real-time data and obtain settlement monitoring data; S135, integrating the tilt monitoring data, crack monitoring data, vibration monitoring data, and settlement monitoring data to obtain real-time IoT data of the house; S131 includes: Carry out load distribution calculation based on the building foundation structure data and key structural component data to obtain structural load distribution data; Material characteristics are processed according to the building foundation structure data, key structural component data and structural load distribution data to obtain material load distribution data; Calculate the sensitivity of structural tilt change according to the material load distribution data to obtain the sensitivity data of structural tilt change; Construct a structural diagram according to the data of key structural components to obtain structural component diagram data; Numerical mapping is performed based on the structural tilt change sensitivity data and the structural component diagram data to obtain component tilt sensitivity matrix data; Performing component load response processing on component tilt sensitivity matrix data to obtain tilt sensitivity matrix data; Screen the monitoring points according to the tilt sensitivity matrix data to obtain the tilt monitoring point data; The tilt monitoring points are arranged according to the tilt monitoring point data to collect real-time data and obtain the tilt monitoring data.

2. The method according to claim 1, characterized in that The component load response processing includes: The component load response prediction model is used to process the component tilt sensitivity matrix data to obtain the tilt sensitivity matrix data; The construction steps of the component load response prediction model include the following steps: Obtaining historical component tilt sensitivity matrix data and historical component tilt sensitivity index data; Perform load feature extraction and tilt sensitivity feature extraction based on historical component tilt sensitivity matrix data to obtain component load feature data and load tilt sensitivity feature data respectively; Perform multi-dimensional feature fusion on component load feature data and load tilt sensitivity feature data to obtain fused feature data; Response function modeling is performed according to component load characteristic data and load tilt sensitivity characteristic data to obtain an initial response function model; Initialize the predicted value of the historical component tilt sensitivity index data to obtain the initial predicted result data; Output the initial result according to the fused feature data and the initial response function model to obtain the initial result data; Divide the data set according to the initial result data to obtain test data and verification data; Perform splitting point splitting according to the test data to obtain splitting point data; Construct a tree structure based on the split point data to obtain initial decision tree data; The initial decision tree data is iteratively updated using the initial prediction result data and verification data to obtain a component load response prediction model; The multi-dimensional feature fusion includes: Characteristic encoding is performed on component load characteristic data and load tilt sensitivity characteristic data to obtain component load characteristic encoding data and load tilt sensitivity characteristic encoding data; Combine the component load feature coding data and the load tilt sensitivity feature coding data to obtain preliminary feature fusion data; Performing feature fusion decoding on the preliminary feature fusion data to obtain first feature fusion data; A graph is constructed based on component load characteristic data and load tilt sensitivity characteristic data to obtain component load sensitivity joint graph data; Perform graph convolution on the component load sensitivity joint graph data to obtain joint graph feature data; Perform causal multi-head self-attention calculation on the joint graph feature data to obtain the second feature fusion data; Perform masked multi-head self-attention calculation based on the first feature fusion data and the second feature fusion data to obtain preliminary fusion feature data; The global context information is integrated on the preliminary fused feature data to obtain fused feature data.

3. The method according to claim 1, characterized in that S15 includes: Perform spatial coordinate mapping according to the house structure data to obtain the house spatial coordinate data; Performing component feature encoding on the house space coordinate data to obtain house component feature encoding data; Component spatial relationship modeling is performed according to the housing component feature coding data and the housing spatial coordinate data to obtain housing spatial relationship data; The house spatial relationship data and the house component feature coding data are spatially fused to obtain the house structure spatial data.

4. The method according to claim 1, characterized in that S2 include: Use drone aerial photography technology to collect image data of the house to be inspected, and obtain house image data; Perform three-dimensional reconstruction according to the house image data to obtain a three-dimensional reconstruction model of the house; Performing point cloud processing and meshing on the three-dimensional reconstruction model of the house to obtain a mesh model of the house; The house grid model is fitted in real time according to the tilt monitoring data, crack monitoring data and settlement monitoring data to obtain a house fitting model; The digital twin model of the house is constructed by using the house structure detail data to fit the house model and obtain the house digital twin model.

5. The method according to claim 1, characterized in that S3 include: Extract abnormal features based on the digital twin model of the house to obtain abnormal feature data of the house structure; Abnormal classification is performed on the abnormal characteristic data of the house structure to obtain abnormal classification data of the house structure; The house structure abnormality classification data is used to perform spatial risk mapping on the house digital twin model to obtain the house structure abnormality data.

6. The method according to claim 5, characterized in that S4 includes: Obtain monitoring point location data; According to the abnormal classification data of the house structure, the abnormal evolution of the monitoring point location data and the abnormal data of the house structure is simulated to obtain the abnormal data of the house structure; A safety risk assessment is performed based on the abnormal data of the house structure to obtain the house safety risk data.

7. A housing safety detection system, characterized in that: Used to perform the housing safety detection method according to claim 1, the housing safety detection system comprises: The housing structure detail collection module is used to obtain housing structure data, and collect housing structure details according to the housing structure data to obtain housing structure detail data; A digital twin model building module is used to collect image data of the house to be inspected using drone aerial photography technology to obtain house image data, and to build a digital twin model based on the house image data and house structure detail data to obtain a house digital twin model; A housing structure anomaly recognition module is used to identify housing structure anomalies based on the housing digital twin model and obtain housing structure anomaly data; The housing safety risk processing module is used to perform housing safety risk processing on housing structure abnormality data to obtain housing safety risk data.

Images