A method for evaluating the risk of external wall falling off

Abstract

The application discloses a kind of outer wall fall-off risk assessment methods, belong to building outer facade safety assessment field, comprising the following steps: S1, generation standardization multi-source risk assessment dataset;S2, generation outer wall damage characteristic parameter set;S3, generation multidimensional vulnerability factor set;S4, through MLP-ANN, risk index fitting is completed, fusion falls off damage critical index, and the outer wall fall-off risk index is calculated;S5, adopts K-means clustering algorithm to divide outer wall fall-off risk grade and completes dynamic verification, and the final outer wall fall-off risk grade result is output;S6, through useless decision analysis optimization maintenance intervention priority, generation outer wall fall-off risk assessment report and customized maintenance scheme, risk assessment core parameter is synchronously dynamically updated.The outer wall fall-off risk assessment method is used, realizes the whole process quantification, dynamic evaluation and fine operation and maintenance management of outer wall fall-off risk.

Description

Technical Field

[0001] This invention relates to the field of building facade safety assessment technology, and in particular to a method for assessing the risk of exterior wall detachment. Background Technology

[0002] With the continuous expansion of the existing urban building stock in my country, many buildings constructed in the 1980s and 1990s suffer from aging of the finishing layer, failure of the adhesive layer, and material deterioration. Coupled with long-term erosion from extreme wind loads, heavy rainfall, temperature fluctuations, earthquakes, and other climatic and disaster factors, exterior wall detachment accidents are frequent, becoming a major hidden danger affecting urban public safety. Traditional exterior wall safety assessment methods can no longer meet the current needs of refined, intelligent, and full life-cycle management. There is an urgent need to develop an exterior wall detachment risk assessment method that can achieve full-dimensional damage perception, multi-factor coupled assessment, dynamic risk updates, and risk-oriented operation and maintenance decisions.

[0003] Currently, existing technologies in the field of risk assessment for exterior wall detachment can be mainly divided into the following categories: Traditional manual inspection and simplified testing techniques, which rely on manual ground observation, high-altitude inspection combined with hollow-sound hammer tapping, depend on the experience of inspectors to judge the appearance defects and internal hollowness of the exterior walls. This was the primary method for early exterior wall safety assessments. However, traditional manual inspection relies on the subjective experience of inspectors, resulting in inconsistent and highly subjective assessments. High-altitude operations pose safety risks and are affected by obstructed views, leading to numerous blind spots and a high rate of missed detections on the facades of high-rise buildings. Hollow-sound hammer tapping is a single-point contact method, inefficient, and can only qualitatively determine the degree of hollowness, failing to quantify defect size, depth, and area, thus hindering sophisticated risk assessments.

[0004] Unmanned Aerial Vehicle (UAV) photogrammetry and machine vision inspection technology uses UAVs to acquire high-definition images of building facades. Combined with target detection models such as Convolutional Neural Networks (CNN) and YOLO (You Only Look Once, a single-stage target detection) series, it achieves automated identification and quantification of surface defects (such as cracks, peeling, and efflorescence), significantly improving inspection efficiency and the automation level of surface defect detection. However, existing UAV visual inspection can only identify surface defects on exterior walls and cannot detect hidden internal defects such as adhesive layer hollowing, peeling, and adhesion failure, resulting in a blind spot: "visible damage can be identified, but invisible hidden dangers cannot be perceived." Furthermore, existing YOLO series models are prone to false positives and false negatives when dealing with complex backgrounds on facades (such as decorative lines, pipelines, and wall textures), and their detection accuracy for small target defects such as micro-cracks and minor efflorescence is insufficient. Moreover, existing technologies do not integrate visual inspection data with other dimensional data, relying solely on surface defects for risk assessment, resulting in a single dimension and insufficient reliability of the assessment results. Summary of the Invention

[0005] The purpose of this invention is to provide a method for assessing the risk of exterior wall detachment, thereby solving the aforementioned technical problems.

[0006] To achieve the above objectives, the present invention provides a method for assessing the risk of exterior wall detachment, comprising the following steps: S1. Collect terahertz non-destructive testing data, UAV photogrammetry visual data, building foundation attribute data, and climate-maintenance data, and complete standardized preprocessing to generate a standardized multi-source risk assessment dataset; S2. Based on the standardized multi-source risk assessment dataset generated by S1, extract terahertz hidden damage features, apparent damage features, material and process degradation features, building age decay features, and maintenance decay features to generate a set of external wall damage feature parameters. S3. Based on the set of external wall damage characteristic parameters generated by S2, calculate the technical vulnerability of the facade components, the terahertz strengthening degradation factor, the climate coupling vulnerability coefficient, and the maintenance and management correction coefficient, and generate a multi-dimensional vulnerability factor set. S4. Based on the multi-dimensional vulnerability factor set generated in S3, the feature importance was screened using random forest, the risk index was fitted using MLP-ANN, and the critical index of detachment damage was fused to calculate the risk index of detachment of the external wall. S5. Based on the external wall detachment risk index calculated in S4, the K-means clustering algorithm is used to classify the external wall detachment risk level and complete dynamic verification, and the final external wall detachment risk level result is output. S6. Based on the final risk level of exterior wall detachment output from S5, optimize the priority of maintenance intervention through useless decision analysis, generate an exterior wall detachment risk assessment report and customized maintenance plan, and update the core parameters of risk assessment in a synchronous and dynamic manner.

[0007] Therefore, the beneficial effects of the above-mentioned method for assessing the risk of exterior wall detachment in this invention are as follows: 1. Multi-source heterogeneous data fusion perception innovation: For the first time, terahertz non-destructive testing technology is integrated with UAV photogrammetry and manual visual inspection to achieve full-dimensional damage perception of external wall facade surface defects and internal hollowing / peeling / material deterioration, making up for the blind spot of traditional visual inspection that cannot identify hidden internal defects; 2. Multi-dimensional coupling innovation throughout the entire life cycle: Construct a risk assessment model that couples six dimensions: building age, materials and processes, climate conditions, maintenance and management, technical vulnerability and deterioration characteristics, and integrates terahertz damage characteristics to correct deterioration factors, breaking through the limitations of traditional assessments that only consider apparent damage or structural parameters. 3. Hybrid Intelligent Algorithm Innovation: An improved hybrid algorithm of YOLOv11 + Random Forest (RF) + Multilayer Perceptron Neural Network (MLP-ANN) is proposed. By combining the feature importance screening of Random Forest with the nonlinear fitting of MLP-ANN, the damage critical index (DCI) and risk clustering accuracy are optimized to achieve dynamic classification of risk levels. 4. Innovation in full lifecycle maintenance decision-making: Based on the theory of useless decision analysis (IDA), the maintenance intervention index is optimized, and customized maintenance plans are generated by combining risk assessment results to achieve closed-loop management from risk assessment to operation and maintenance decision-making.

[0008] In summary, this invention compensates for the blind spots of traditional detection through terahertz nondestructive testing, covers all life-cycle influencing factors through six-dimensional factor coupling, improves assessment accuracy through hybrid intelligent algorithms, and achieves precise operation and maintenance through useless decision analysis. Compared with traditional methods: 1) the risk assessment accuracy is increased to over 95%, and the internal hidden damage identification rate is 100%; 2) the assessment efficiency is increased by 80%, eliminating the need for manual high-altitude operations and automating the entire process; 3) it enables dynamic risk updates and customized maintenance, reducing the incidence of exterior wall detachment accidents by 90%.

[0009] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0010] Figure 1 This is a flowchart of a method for assessing the risk of exterior wall detachment as described in this invention. Detailed Implementation

[0011] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the embodiments of the present invention and are not intended to limit the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of this application. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout.

[0012] It should be noted that the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, such as a process, method, system, product, or server that includes a series of steps or units, not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product, or device.

[0013] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0014] like Figure 1 As shown, a method for assessing the risk of exterior wall detachment includes the following steps: S1. Collect terahertz non-destructive testing data, UAV photogrammetry visual data, building foundation attribute data, and climate-maintenance data, and complete standardized preprocessing to generate a standardized multi-source risk assessment dataset; S2. Based on the standardized multi-source risk assessment dataset generated by S1, extract terahertz hidden damage features, apparent damage features, material and process degradation features, building age decay features, and maintenance decay features to generate a set of external wall damage feature parameters. S3. Based on the set of external wall damage characteristic parameters generated by S2, calculate the technical vulnerability of the facade components, the terahertz strengthening degradation factor, the climate coupling vulnerability coefficient, and the maintenance and management correction coefficient, and generate a multi-dimensional vulnerability factor set. S4. Based on the multi-dimensional vulnerability factor set generated in S3, the feature importance was screened using random forest, the risk index was fitted using MLP-ANN, and the critical index of detachment damage was fused to calculate the risk index of detachment of the external wall. S5. Based on the external wall detachment risk index calculated in S4, the K-means clustering algorithm is used to classify the external wall detachment risk level and complete dynamic verification, and the final external wall detachment risk level result is output. S6. Based on the final risk level of exterior wall detachment output from S5, optimize the priority of maintenance intervention through useless decision analysis, generate an exterior wall detachment risk assessment report and customized maintenance plan, and update the core parameters of risk assessment in a synchronous and dynamic manner.

[0015] In step S1, the terahertz nondestructive testing data is first processed. and drone photogrammetry visual data Gaussian filtering is performed, and then the filtered terahertz nondestructive testing data, UAV photogrammetric visual data, and building foundation attribute data are integrated. Climate-Maintenance Data This yields a standardized multi-source risk assessment dataset.

[0016] Specifically, the data acquisition steps for terahertz nondestructive testing are as follows: The first step involves using a continuous terahertz imaging system to perform a full-coverage scan of the eaves, balconies, wall finishes, bay windows, and other easily detachable components of the building's exterior walls. The scanning resolution is set to 0.1 mm, and the scanning depth covers the exterior wall finish to the adhesive layer (0-50 mm). The second step involves acquiring terahertz time-domain spectral, reflectance, transmittance, and internal defect imaging data, focusing on identifying hidden damage such as hollow areas, peeling, adhesive failure, and material loosening within the exterior walls. The location, size, and depth of the damage are recorded in three dimensions. The third step generates raw terahertz non-destructive testing data. .

[0017] The steps for acquiring visual data via UAV photogrammetry are as follows: The first step involved using a quadcopter drone equipped with a 20-megapixel RGB camera to acquire high-resolution images of the exterior walls in a dual-grid flight mode (80% forward overlap and 70% lateral overlap). The drone flew at a horizontal distance of 1.5 meters from the exterior walls to ensure complete image coverage of the facade area. The second step involved using Agisoft Metashape software to perform 3D reconstruction of the drone images, generating a high-density point cloud, textured mesh model, and digital surface model (DSM) of the exterior walls. The third step involved extracting the location, area, and morphological data of surface defects (cracks, efflorescence, peeling, and exposed reinforcement) to generate raw drone photogrammetric visual data. .

[0018] The steps for collecting basic building attribute data are as follows: Step 1: Retrieve building completion archives and collect basic data such as building age, exterior facade material type, construction process parameters, and component geometric dimensions (area, thickness, height-to-thickness ratio); Step 2: Manually verify and confirm the types of exterior wall components: parapet walls, eaves, balconies, wall finishes, doors and windows, and external equipment; Step 3: Generate a basic building attribute dataset. ; The steps for collecting climate-maintenance data are as follows: Step 1: Retrieve monitoring data from the local meteorological station for the past 10 years, collecting data on annual average rainfall, maximum wind speed, peak ground acceleration, temperature fluctuation range, and chloride ion concentration. Step 2: Retrieve building maintenance records, collecting data on maintenance frequency, maintenance process, last maintenance date, and maintenance defect repair rate. Step 3: Generate a climate-maintenance dataset. .

[0019] In step S2, terahertz concealed damage characteristic value The calculation formula is as follows: ; In the formula, Indicates the internal void ratio; Indicates the peeling depth; This indicates the percentage of area where adhesion failed. Indicates the porosity of the material; The steps for extracting apparent damage features are as follows: Step 1: Based on the UAV photogrammetric visual data output from S1, an improved YOLOv11 target detection model is used to identify apparent damage to the exterior walls (cracks, spalling, efflorescence, exposed rebar). The improved YOLOv11 target detection model includes the following layers arranged sequentially: a 3D texture input adaptation layer, a C3k2 grouped residual feature extraction backbone, a SPPF fast spatial pyramid pooling layer, a C2PSA parallel spatial attention enhancement layer, a multi-scale damage detection neck layer, a terahertz feature fusion auxiliary layer, an apparent damage classification and regression detection head, and an exterior wall damage weighted-focal joint loss layer. The 3D texture input adaptation layer is used to normalize the R values ​​in the input S1 preprocessed UAV photogrammetric visual data. GB imagery and digital surface model are used for 6×6 convolution feature extraction, pixel-wise normalization, and 3D texture mesh alignment to output an aligned multimodal facade initial feature map. A C3k2 grouped residual feature extraction backbone is used to process the input multimodal facade initial feature map through grouped convolution, residual connections, and depth downsampling to enhance the facade material texture and damage edge features, outputting a multi-scale facade depth feature map. A SPPF fast spatial pyramid pooling layer is used to process the input multi-scale facade depth feature map through multi-scale receptive field fusion, dilated convolution downsampling, and local feature pooling to eliminate interference from damage scale differences such as fine cracks, blocky peeling, and planar efflorescence, outputting... The system integrates global and local multi-scale pooling feature maps. A C2PSA parallel spatial attention enhancement layer processes the input multi-scale pooling feature map through parallel spatial attention weighting, channel feature calibration, and damage region feature enhancement. This suppresses background interference from doors, windows, pipelines, decorative lines, and wall textures, and strengthens the feature responses of four types of damage: cracks, spalling, efflorescence, and exposed reinforcement. The output is an attention-enhanced facade damage feature map. A multi-scale damage detection neck layer processes the attention-enhanced facade damage feature map through upsampling, cross-scale feature stitching, and branch splitting. This splits the map into a small target detection branch (suitable for slender micro-cracks), a medium target detection branch (suitable for planar efflorescence and small spalling), and a large target detection branch. The target detection branch (adapted for large-area spalling and exposed rebar areas) outputs a three-scale branch damage feature map. The terahertz feature fusion auxiliary layer performs scale-wise feature concatenation, 1×1 convolutional channel compression, and weighted feature fusion on the input three-scale branch damage feature map and terahertz hidden damage features, deeply coupling the terahertz hidden damage features with the apparent visual damage features, and outputs a multi-scale damage detection feature map fused with terahertz features. The apparent damage classification and regression detection head performs classification of external wall apparent damage (cracks, spalling, efflorescence, exposed rebar), bounding box regression, and detection confidence prediction on the input multi-scale damage detection feature map fused with terahertz hidden damage features, outputting the external wall apparent damage category. Damage area 1. Width of apparent cracks in exterior walls Detection confidence alkali efflorescence damage level Degree of exposed rebar corrosion Number of damages The damage bounding box coordinate parameters; the external wall damage weighted-focal joint loss layer is used to perform multi-task fusion calculation of the input damage classification prediction results, bounding box regression results, and manually labeled real damage labels, including focal loss (to solve the problem of imbalanced samples for cracks and exposed rebar), weighted EIOU loss (to optimize the regression accuracy of irregular damage bounding boxes), and classification loss, and outputs the total model loss. This is applied to the backpropagation gradient calculation and network parameter iterative update of the improved YOLOv11 model. Step 2: Calculate the apparent damage characteristic values : ; In the formula, Indicates the total area of ​​the exterior walls; Material degradation characteristics include material degradation features and process defect characteristics ; in, ; ; Building age degradation characteristics and maintenance degradation characteristics include building age degradation coefficient. and maintenance management attenuation coefficient ; in, ; ; In the formula, Indicates the actual service life of the building; Indicates the maintenance defect repair rate; Indicates the current time; Indicates the time of the last maintenance; By integrating terahertz hidden damage characteristics, apparent damage characteristics, material and process degradation characteristics, building age degradation characteristics, and maintenance degradation characteristics, a set of external wall damage characteristic parameters is generated. .

[0020] Step S3 specifically includes the following steps: S31. Obtain the weight coefficient of apparent damage severity based on the recognition results of the improved YOLOv11 target detection model. : ; In the formula, This indicates the proportion of the damaged area to the total area of ​​the exterior wall; These represent mild efflorescence, moderate efflorescence, severe efflorescence, and critical efflorescence, respectively. These represent no corrosion, slight corrosion, moderate corrosion, and severe exposed corrosion, respectively. Simultaneously, the weighting coefficient for the severity of terahertz hidden damage is calculated. : ; Weighting coefficients for apparent damage severity Weighting coefficients for the severity of terahertz hidden damage Calculate the apparent-terahertz damage severity fusion weight. : ; S32. Calculate the technical vulnerability of facade components, terahertz strengthening degradation factor, climate-coupled vulnerability coefficient, and maintenance and management correction coefficient; The formula for calculating the technical vulnerability of facade components is as follows: ; In the formula, Indicates the vulnerability value of the technology; Indicates the component number Vulnerability assignment for each technical feature; Terahertz Enhancement Degradation Factor The calculation formula is as follows: ; In the formula, Indicates the first Abnormal gravity coefficient for damage-like conditions; Indicates the first Percentage of area affected by this type of injury; express The corresponding percentage of the damaged area; Climate Coupling Vulnerability Coefficient The calculation formula is as follows: ; In the formula, This represents the normalized value of rainfall; This represents the normalized value of wind speed; This represents the normalized value of seismic acceleration. This represents the normalized value of the temperature alternation amplitude; Maintenance and management correction factor The calculation formula is as follows: ; In the formula, Indicates the maintenance attenuation coefficient; S33, Technical Vulnerability of Integrated Facade Components Terahertz Enhancement Degradation Factor Climate-coupled vulnerability coefficient and maintenance management correction factor Generate a multi-dimensional vulnerability factor set .

[0021] Step S4 specifically includes the following steps: S41, Set up multi-dimensional vulnerability factors Input a random forest model, with dropout risk as the output target, and train the model to calculate the feature importance weights of each factor. After sorting the feature importance weights in descending order, the feature importance weights are then removed. Redundant features, retain the core feature set In this embodiment, the ranking results are as follows: Terahertz enhancement degradation feature (0.543) > Technical vulnerability (0.378) > Climate coupling vulnerability coefficient (0.19) > Maintenance and management correction coefficient (0.06). It can be seen that the maintenance correction coefficient is removed, while the terahertz enhancement degradation feature, technical vulnerability and climate coupling vulnerability coefficient are retained. S42, Core Feature Set The input is a Sigmoid activation function, and the optimization algorithm is an MLP-ANN model trained using trainlm. The output is an initial risk index. : ; In the formula, This represents the Sigmoid activation function; Indicates the first Core features; Indicates the first Item weight; Indicates neuron bias; S43. Integrating terahertz hidden damage characteristics, building age, and maintenance management factors, a critical index for detachment damage is constructed. : ; In the formula, , , , and All represent weights; This represents the normalized value of the damaged area; S44, Initial Risk Index of Integration and the critical index of shedding damage The final risk index for exterior wall detachment was obtained. : .

[0022] Step S5 specifically includes the following steps: S51. Using the K-means clustering algorithm to determine the risk index of exterior wall detachment. Perform clustering and assign initial risk levels: ; S52. Select 10% of the evaluation sample for manual on-site verification and calculate the clustering accuracy. ;like Then adjust the cluster weights until... Output the final risk level .

[0023] Step S6 specifically includes the following steps: S61, Constructing and Maintaining a Useless Index : ; In the formula, This indicates the economic cost of maintenance; Indicates the need to protect environmental impact; This represents the maximum economic cost of maintenance among all maintenance intervention options to be evaluated; This represents the maximum environmental impact of maintenance among all maintenance intervention options to be evaluated. S62, in conjunction with the final risk level and maintenance of useless index Customized maintenance priority: .

[0024] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for assessing the risk of exterior wall detachment, characterized in that: Includes the following steps: S1. Collect terahertz non-destructive testing data, UAV photogrammetry visual data, building foundation attribute data, and climate-maintenance data, and complete standardized preprocessing to generate a standardized multi-source risk assessment dataset; S2. Based on the standardized multi-source risk assessment dataset generated by S1, extract terahertz hidden damage features, apparent damage features, material and process degradation features, building age decay features, and maintenance decay features to generate a set of external wall damage feature parameters. S3. Based on the set of external wall damage characteristic parameters generated by S2, calculate the technical vulnerability of the facade components, the terahertz strengthening degradation factor, the climate coupling vulnerability coefficient, and the maintenance and management correction coefficient, and generate a multi-dimensional vulnerability factor set. S4. Based on the multi-dimensional vulnerability factor set generated in S3, random forest is used to screen the feature importance, MLP-ANN is used to fit the risk index, and the critical index of detachment damage is fused to calculate the risk index of detachment of the external wall. S5. Based on the external wall detachment risk index calculated in S4, the K-means clustering algorithm is used to classify the external wall detachment risk level and complete dynamic verification, and the final external wall detachment risk level result is output. S6. Based on the final external wall detachment risk level result output by S5, optimize the maintenance intervention priority through useless decision analysis, generate an external wall detachment risk assessment report and customized maintenance plan, and update the core parameters of the risk assessment in a synchronous and dynamic manner. In step S2, terahertz concealed damage characteristic value The calculation formula is as follows: ； In the formula, Indicates the internal void ratio; Indicates the peeling depth; Indicates the percentage of area where adhesion failed; Indicates the porosity of the material; The steps for extracting apparent damage features are as follows: Step 1: Based on the UAV photogrammetric visual data output by S1, an improved YOLOv11 target detection model is used to identify apparent damage to the exterior wall. The improved YOLOv11 target detection model includes the following layers arranged sequentially: a 3D texture input adaptation layer, a C3k2 grouped residual feature extraction backbone, a SPPF fast spatial pyramid pooling layer, a C2PSA parallel spatial attention enhancement layer, a multi-scale damage detection neck layer, a terahertz feature fusion auxiliary layer, an apparent damage classification and regression detection head, and an exterior wall damage weighted-focal joint loss layer. The supplementary layer is used to perform 6×6 convolution feature extraction, pixel-wise normalization, and 3D texture mesh alignment on the standardized RGB image and digital surface model in the input S1 preprocessed UAV photogrammetric visual data, outputting the aligned multimodal facade initial feature map; the C3k2 grouped residual feature extraction backbone is used to perform grouped convolution, residual connection, and depth downsampling feature extraction on the input multimodal facade initial feature map, outputting a multi-scale facade depth feature map; the SPPF fast spatial pyramid pooling layer is used to process the input multi-scale facade depth features. The system performs multi-scale receptive field fusion, dilated convolution downsampling, and local feature pooling to output a multi-scale pooled feature map that fuses global and local features. A C2PSA parallel spatial attention enhancement layer is used to perform parallel spatial attention weighting, channel feature calibration, and damage region feature enhancement on the input multi-scale pooled feature map, outputting an attention-enhanced facade damage feature map. A multi-scale damage detection neck layer is used to upsample, perform cross-scale feature concatenation, and branch splitting on the attention-enhanced facade damage feature map, splitting it into a small target detection branch, a medium target detection branch, and a... The system consists of two branches: a target detection branch and a large target detection branch, which output a three-scale branch damage feature map. A terahertz feature fusion auxiliary layer processes the input three-scale branch damage feature map and terahertz hidden damage features through scale-wise feature concatenation, 1×1 convolutional channel compression, and weighted feature fusion, outputting a multi-scale damage detection feature map fused with terahertz features. An apparent damage classification and regression detection head processes the input multi-scale damage detection feature map fused with terahertz hidden damage features through external wall apparent damage classification, bounding box regression, and detection confidence prediction, outputting the external wall apparent damage category. Damage area 1. Width of apparent cracks in exterior walls Detection confidence alkali efflorescence damage level Degree of exposed steel reinforcement corrosion Number of damages The damage bounding box coordinate parameters; the external wall damage weighted-focal joint loss layer is used to perform multi-task fusion calculation of focal loss, weighted EIOU loss and classification loss on the input damage classification prediction results, bounding box regression results and manually labeled real damage labels, and outputs the total model loss. This is applied to the backpropagation gradient calculation and network parameter iterative update of the improved YOLOv11 model. Step 2: Calculate the apparent damage characteristic values : ； In the formula, Indicates the total area of ​​the exterior walls; Material degradation characteristics include material degradation features and process defect characteristics ; in, ; ； Building age degradation characteristics and maintenance degradation characteristics include building age degradation coefficient. and maintenance management attenuation coefficient ; in, ; ； In the formula, Indicates the actual service life of the building; Indicates the maintenance defect repair rate; Indicates the current time; Indicates the time of the last maintenance; By integrating terahertz hidden damage characteristics, apparent damage characteristics, material and process degradation characteristics, building age degradation characteristics, and maintenance degradation characteristics, a set of external wall damage characteristic parameters is generated. .

2. The method for assessing the risk of exterior wall detachment according to claim 1, characterized in that: In step S1, the terahertz nondestructive testing data is first processed. and drone photogrammetry visual data Gaussian filtering is performed, and then the filtered terahertz nondestructive testing data, UAV photogrammetric visual data, and building foundation attribute data are integrated. Climate-Maintenance Data This yields a standardized multi-source risk assessment dataset.

3. The method for assessing the risk of exterior wall detachment according to claim 2, characterized in that: Step S3 specifically includes the following steps: S31. Obtain the weight coefficient of apparent damage severity based on the recognition results of the improved YOLOv11 target detection model. : ； In the formula, This indicates the proportion of the damaged area to the total area of ​​the exterior wall; Simultaneously, the weighting coefficient for the severity of terahertz hidden damage is calculated. : ； Weighting coefficients for apparent damage severity Weighting coefficients for the severity of terahertz hidden damage Calculate the apparent-terahertz damage severity fusion weight. : ； S32. Calculate the technical vulnerability of facade components, terahertz strengthening degradation factor, climate-coupled vulnerability coefficient, and maintenance and management correction coefficient; The formula for calculating the technical vulnerability of facade components is as follows: ； In the formula, Indicates the vulnerability value of the technology; Indicates the component number Vulnerability assignment for each technical feature; Terahertz Enhancement Degradation Factor The calculation formula is as follows: ； In the formula, Indicates the first Abnormal gravity coefficient for damage-like conditions; Indicates the first Percentage of area affected by this type of injury; express The corresponding percentage of the damaged area; Climate Coupling Vulnerability Coefficient The calculation formula is as follows: ； In the formula, This represents the normalized value of rainfall; This represents the normalized value of wind speed; This represents the normalized value of seismic acceleration. This represents the normalized value of the temperature alternation amplitude; Maintenance and management correction factor The calculation formula is as follows: ； In the formula, Indicates the maintenance attenuation coefficient; S33, Technical Vulnerability of Integrated Facade Components Terahertz Enhancement Degradation Factor Climate-coupled vulnerability coefficient and maintenance management correction factor Generate a multi-dimensional vulnerability factor set .

4. The method for assessing the risk of exterior wall detachment according to claim 3, characterized in that: Step S4 Specifically, the following steps are included: S41, Set up multi-dimensional vulnerability factors Input a random forest model, with dropout risk as the output target, and train the model to calculate the feature importance weights of each factor. After sorting the feature importance weights in descending order, the feature importance weights are then removed. Redundant features, retain the core feature set ; S42, Core Feature Set The input is a Sigmoid activation function, and the optimization algorithm is an MLP-ANN model trained using trainlm. The output is an initial risk index. : ； In the formula, This represents the Sigmoid activation function; Indicates the first Core features; Indicates the first Item weight; Indicates neuron bias; S43. Integrating terahertz hidden damage characteristics, building age, and maintenance management factors, a critical index for detachment damage is constructed. : ； In the formula, , , , and All represent weights; This represents the normalized value of the damaged area; S44, Initial Risk Index of Integration and the critical index of shedding damage The final risk index for exterior wall detachment was obtained. : 。 5. The method for assessing the risk of exterior wall detachment according to claim 4, characterized in that: Step S5 specifically includes the following steps: S51. Using the K-means clustering algorithm to determine the risk index of exterior wall detachment. Perform clustering and assign initial risk levels: ； S52. Select 10% of the evaluation sample for manual on-site verification and calculate the clustering accuracy. ;like Then adjust the cluster weights until... Output the final risk level .

6. The method for assessing the risk of exterior wall detachment according to claim 5, characterized in that: Step S6 specifically includes the following steps: S61, Constructing and Maintaining a Useless Index : ； In the formula, This indicates the economic cost of maintenance; Indicates the need to protect environmental impact; This represents the maximum economic cost of maintenance among all maintenance intervention options to be evaluated; This represents the maximum environmental impact of maintenance among all maintenance intervention options to be evaluated. S62, in conjunction with the final risk level and maintenance of useless index Customized maintenance priority: 。

Images