A historical building repair achievement intelligent evaluation management system and method

Abstract

The present application relates to the technical field of building repair achievement evaluation management, and in particular to a historical building repair achievement intelligent evaluation management system and method, the system comprising a repair data acquisition module for acquiring multi-modal repair data, a digital twin construction module for fusing data to generate a repair completion digital twin, a multi-dimensional repair feature extraction module for quantitative analysis to generate a structured multi-dimensional repair feature vector, which includes a geometric deviation quantitative unit, a material performance evaluation unit and a value intervention quantitative unit, a fusion evaluation engine for inputting the feature vector into a machine learning model to obtain a comprehensive repair quality grade, and a visual interactive and management terminal for presenting the digital twin and evaluation details. The present application realizes multi-dimensional quantitative evaluation and intelligent comprehensive determination of repair achievements, and intuitively displays in a visual manner, thereby providing a scientific basis for the acceptance and archiving of historical building repair achievements.

Description

Technical Field

[0001] This invention relates to the field of building renovation outcome evaluation and management technology, specifically to an intelligent evaluation and management system and method for historical building renovation outcomes. Background Technology

[0002] Current techniques for assessing the restoration results of historical buildings mainly rely on manual on-site inspections, comparison of two-dimensional drawings, and empirical judgments. The assessment process lacks effective integration and quantitative analysis of multi-source heterogeneous data. Traditional methods struggle to integrate and utilize spatial geometric information, visual image information, and material performance information, resulting in assessment results that are highly subjective and have poor repeatability.

[0003] Furthermore, existing technologies lack quantitative methods for assessing the extent to which renovation projects interfere with the core historical value of a building, making it impossible to scientifically measure the impact of renovation activities on original components, traditional craftsmanship, and other valuable elements. Assessment results are mostly presented in paper reports or simple two-dimensional charts, lacking intuitive three-dimensional visualization tools, making it difficult for managers to quickly grasp the overall state of the building after renovation and the detailed condition of each part.

[0004] Therefore, there is an urgent need for a renovation outcome evaluation management system and method that can achieve multimodal data fusion, multidimensional quantitative analysis, intelligent comprehensive evaluation, and has visual interactive functions. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent evaluation and management system and method for the restoration of historical buildings, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A smart evaluation and management system for the results of historical building restoration includes:

[0008] The renovation data acquisition module is used to acquire multimodal renovation data of the target historical building after renovation is completed; the multimodal renovation data includes at least spatial geometric data, appearance image data and material performance data;

[0009] The digital twin construction module, connected to the renovation data acquisition module, is used to fuse multimodal renovation data and construct a completed digital twin that represents the true state of the target historical building after renovation.

[0010] A multi-repair feature extraction module, connected to the digital twin construction module, is used to perform multi-dimensional quantitative analysis on the completed digital twin of repairs, generating structured multi-repair feature vectors. The multi-repair feature extraction module includes:

[0011] The geometric deviation quantization unit is used to extract the geometric features of the repaired digital twin and compare it with the preset repair design digital model to generate geometric deviation feature values.

[0012] The material performance evaluation unit is used to extract the associated material performance data from the digital twin of the repaired project and compare it with the standard values ​​in the preset historical value assessment benchmark library to generate material performance compliance feature values.

[0013] The value intervention quantification unit is used to identify and quantify the degree of intervention of the core value elements reflected in the completed digital twin of the renovation based on a preset value element identification model, and generate the value intervention index feature value.

[0014] The integrated evaluation engine connects the geometric deviation quantification unit, the material performance evaluation unit, and the value intervention quantification unit respectively. It is used to combine the geometric deviation feature value, the material performance compliance feature value, and the value intervention index feature value to form a structured multi-maintenance feature vector. The structured multi-maintenance feature vector is then input into a pre-trained comprehensive maintenance quality evaluation machine learning model for processing to obtain the comprehensive maintenance quality level of the target historical building.

[0015] The visual interaction and management terminal connects to the integrated evaluation engine to present the completed digital twin of the repair in a visual manner, and displays the comprehensive repair quality level and its corresponding multi-repair feature vector details for managers to view and archive.

[0016] A smart evaluation and management method for the results of historical building restoration, the method is executed by a system and includes the following steps:

[0017] Step S1: Obtain multimodal renovation data of the target historical building after renovation through the renovation data acquisition module; the multimodal renovation data includes at least spatial geometric data, appearance image data, and material performance data;

[0018] Step S2: The multimodal renovation data is fused through the digital twin construction module to construct a completed digital twin that represents the actual state of the target historical building after renovation.

[0019] Step S3: Perform multi-dimensional quantitative analysis on the completed digital twin of the repaired structure using the multi-repair feature extraction module to generate a structured multi-repair feature vector; this step specifically includes:

[0020] Step S3-1: Extract the geometric features of the repaired digital twin through the geometric deviation quantization unit, perform spatial registration and alignment with the preset repair design digital model, calculate the spatial distance deviation and angle deviation between the corresponding feature surfaces, and after statistical analysis to remove outliers, fuse and generate the overall average geometric deviation, local maximum geometric deviation and overall deviation distribution uniformity as geometric deviation feature values.

[0021] Step S3-2: Through the material performance evaluation unit, traverse all building components in the repaired digital twin that are bound to material performance data, extract the material type and measured material performance value of each building component, retrieve the corresponding historical standard performance range from the preset historical value assessment benchmark library, compare the measured value with the standard range item by item to calculate the single-point conformity, and combine the component value importance weight for weighted fusion to obtain the material performance conformity feature value;

[0022] Step S3-3: The value intervention quantification unit calls the preset value element identification model to traverse and analyze the digital twin of the completed renovation, identify the core value elements, compare the historical archive data before renovation to quantify the treatment method and degree of each core value element in this renovation, and accumulate the value intervention index feature value according to the preset intervention coefficient and the value weight of the element, and normalize the data to obtain the value intervention index feature value.

[0023] Step S4: After dimensionless normalization of the geometric deviation feature value, material performance compliance feature value, and value intervention index feature value through the fusion evaluation engine, they are spliced ​​and combined in a preset order to form a structured multi-repair feature vector. The structured multi-repair feature vector is then input into a pre-trained comprehensive repair quality evaluation machine learning model. The model outputs the probability distribution of the target historical building belonging to each preset comprehensive repair quality level through forward calculation. The level with the highest probability is selected as the comprehensive repair quality level. At the same time, the feature importance analysis function of the model is used to calculate the contribution weight of each feature value to the judgment result as explanatory information.

[0024] Step S5: Receive and parse the completed digital twin of the repair project, the comprehensive repair quality level, details of multiple repair feature vectors, and explanations of contribution weights through the visual interaction and management terminal. Construct a 3D visualization scene and render and display the 3D model after repair in real time. Overlay the evaluation information graphically around the model and assign color-highlighted markers to different component parts of the model according to the numerical range of each feature value. Respond to user selection operations for components to query and display detailed inspection data and evaluation details of the component. Finally, integrate and generate a standard format repair result evaluation report according to user instructions and provide export and archiving functions.

[0025] As can be seen from the technical solution provided by the present invention above, the intelligent evaluation and management system and method for the restoration results of historical buildings provided by the present invention have the following beneficial effects:

[0026] By constructing a digital twin of the completed renovation, we have achieved multimodal data fusion and high-precision digital representation of the post-renovation building status, providing a unified and complete data foundation for subsequent evaluation.

[0027] By setting up multiple repair feature extraction modules, the repair results are quantitatively analyzed from three core dimensions: geometric deviation, material performance compliance, and degree of value intervention. This comprehensively covers the core elements of historical building repair assessment and accurately reflects the basic principles of not changing the original state of cultural relics, minimal intervention, and authenticity protection.

[0028] By using machine learning models in the fusion assessment engine to intelligently process structured multi-repair feature vectors, the comprehensive repair quality level can be objectively and accurately assessed, and the contribution weight of each feature to the assessment result can be output, which provides interpretability to the assessment results and improves the scientificity and credibility of the assessment.

[0029] The visual interactive and management terminal can intuitively present the completed digital twin of the repair and assessment details in a three-dimensional manner. It supports interactive queries and abnormal component warnings, and can generate standardized assessment reports with one click. It provides managers with an intuitive and efficient tool and realizes full life-cycle traceable management of the repair results assessment. Attached Figure Description

[0030] Figure 1 This is a schematic diagram of the structure of an intelligent evaluation and management system for the restoration results of historical buildings according to the present invention.

[0031] Figure 2 This is a schematic diagram of the steps in the intelligent evaluation and management method for the restoration results of historical buildings according to the present invention. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0033] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific embodiments.

[0034] like Figure 1-2 As shown, this embodiment of the invention provides an intelligent evaluation and management system for the results of historical building renovation, including:

[0035] The renovation data acquisition module is used to acquire multimodal renovation data of the target historical building after renovation is completed; the multimodal renovation data includes at least spatial geometric data, appearance image data and material performance data;

[0036] The digital twin construction module, connected to the renovation data acquisition module, is used to fuse multimodal renovation data and construct a completed digital twin that represents the true state of the target historical building after renovation.

[0037] A multi-repair feature extraction module, connected to the digital twin construction module, is used to perform multi-dimensional quantitative analysis on the completed digital twin of repairs, generating structured multi-repair feature vectors. The multi-repair feature extraction module includes:

[0038] The geometric deviation quantization unit is used to extract the geometric features of the repaired digital twin and compare it with the preset repair design digital model to generate geometric deviation feature values.

[0039] The material performance evaluation unit is used to extract the associated material performance data from the digital twin of the repaired project and compare it with the standard values ​​in the preset historical value assessment benchmark library to generate material performance compliance feature values.

[0040] The value intervention quantification unit is used to identify and quantify the degree of intervention of the core value elements reflected in the completed digital twin of the renovation based on a preset value element identification model, and generate the value intervention index feature value.

[0041] The integrated evaluation engine connects the geometric deviation quantification unit, the material performance evaluation unit, and the value intervention quantification unit respectively. It is used to combine the geometric deviation feature value, the material performance compliance feature value, and the value intervention index feature value to form a structured multi-maintenance feature vector. The structured multi-maintenance feature vector is then input into a pre-trained comprehensive maintenance quality evaluation machine learning model for processing to obtain the comprehensive maintenance quality level of the target historical building.

[0042] The visual interaction and management terminal connects to the integrated evaluation engine to present the completed digital twin of the repair in a visual manner, and displays the comprehensive repair quality level and its corresponding multi-repair feature vector details for managers to view and archive.

[0043] In this embodiment, the renovation data acquisition module is the core data foundation for the intelligent evaluation and management system for historical building renovation results to achieve full-process quantitative evaluation. It provides a complete, accurate, and spatiotemporally aligned basic data source for subsequent digital twin construction and renovation quality evaluation through standardized collection and preprocessing of multi-dimensional status information after the completion of the target historical building's renovation. The renovation data acquisition module specifically includes:

[0044] 3D laser scanning unit:

[0045] Scanning scheme adaptation and on-site data acquisition: Based on the structural form, spatial scale and key repair areas of the target historical building, a hierarchical scanning scheme is developed, and corresponding scanning resolutions and sampling densities are set for the overall outline of the building, key structural nodes and core decorative components. The target historical building after repair is scanned in its entirety using a 3D laser scanning device to obtain spatial point cloud data covering the entire area of ​​the building. This data will serve as the core spatial geometric data of the system.

[0046] Initial verification of point cloud data: The integrity of the point cloud data acquired in a single scan is verified, scanning blind spots and missing data areas are identified, and supplementary scanning is performed on the missing parts to ensure that the acquired point cloud data can completely cover all repaired parts and core components of the target historical building without any structural data omissions.

[0047] Initial alignment of point cloud coordinates: Based on the unified reference control points set up on site, the point cloud data obtained by multi-station scanning is initially aligned to unify all scanning data under the same engineering coordinate system, laying the foundation for subsequent spatiotemporal registration of multi-source data.

[0048] High-definition image acquisition unit:

[0049] Data Acquisition Path Planning and Image Capture: Combining the station distribution of 3D laser scanning with the facade and detailed features of the target historical building, a multi-view surround acquisition path is planned to capture high-definition images of the building's exterior facade, interior space, decorative patterns, and component details. For key repair areas and core value elements, a macro shooting mode is used to acquire ultra-high resolution detailed images, fully recording the building's appearance after repair and forming the appearance image data required by the system.

[0050] Image quality screening and optimization: The captured image data is screened for sharpness, exposure and completeness, and invalid images that are blurry, overexposed or underexposed are removed. The valid images are subjected to distortion correction and color normalization to eliminate the optical distortion of the shooting equipment and the color deviation caused by different lighting environments, so as to ensure the consistency and authenticity of the apparent image data.

[0051] Initial image pose calibration: Based on the feature points and reference control points in the image, the camera pose of each valid image is initially calibrated, and the spatial position and viewing direction of the image are recorded to provide pose basis for subsequent texture mapping and multi-source data fusion.

[0052] Material performance testing unit:

[0053] Inspection point layout and scheme formulation: Based on the renovation plan, material type and component value grade of the target historical building, determine the key inspection parts and inspection points; for different types of renovation materials such as wood components, brick and stone components, plaster layer, and paint painting, formulate corresponding inspection schemes and clarify the inspection indicators and inspection methods.

[0054] On-site and laboratory testing: For the main building structure and immovable components, on-site non-destructive testing methods are used to obtain performance data such as material strength, density, moisture content, and residual defects; for sampleable repair materials, sampling laboratory testing is performed to obtain accurate data such as material chemical composition, mechanical properties, and durability indicators, forming a complete material performance data system;

[0055] Standardization and organization of test data: Standardize and organize all material performance data obtained from testing, clarify the component name, spatial location, material type, test index and test method corresponding to each data, establish a one-to-one correspondence between data and building entity parts, and provide a foundation for subsequent binding of performance information and digital model;

[0056] Data synchronization and preprocessing unit:

[0057] Multi-source data spatiotemporal registration: This process receives spatial point cloud data from the 3D laser scanning unit, appearance image data from the high-definition image acquisition unit, and material performance data from the material performance testing unit. Using a unified engineering coordinate system and reference control points as a benchmark, it performs spatiotemporal registration on these three types of data. Through feature point matching and coordinate transformation, it achieves precise alignment of spatial geometric data, appearance image data, and material performance data in spatial location, ensuring that multi-source data corresponding to the same building component have a consistent spatial coordinate benchmark.

[0058] Data denoising and outlier removal: Denoising is performed on the registered multi-source data. For spatial point cloud data, statistical filtering is used to remove noise data such as drift points and outliers generated during the scanning process. For appearance image data, filtering algorithms are used to eliminate salt-and-pepper noise and Gaussian noise in the image. For material performance data, outliers generated during the detection process are removed using the Grubbs criterion to ensure the validity of all input data.

[0059] Data normalization processing: Normalize the denoised multi-source data. For different types and magnitudes of data, formulate corresponding normalization rules to transform all data into a unified numerical range, eliminate the differences in dimensions and magnitudes between different data types, and form standardized multimodal repair data.

[0060] Data encapsulation and transmission: The multimodal repair data that has been registered, denoised and normalized is encapsulated in a structured manner, and a data index is established according to the hierarchy and spatial location of building components. The encapsulated standardized data is then synchronously transmitted to the digital twin construction module to provide a complete basic data source for the subsequent construction of the digital twin.

[0061] Furthermore, the multi-source data spatiotemporal registration technology, based on rigid body coordinate transformation theory, achieves precise alignment of data from different sources within the same spatial coordinate system. By minimizing the spatial distance deviation between corresponding feature points, the optimal coordinate transformation parameters are solved to complete the multi-source data registration operation. The core transformation formula is:

[0062] ,in, This is the three-dimensional coordinate vector of the feature points in the original data. The three-dimensional coordinate vector of the registered feature points. This is a three-dimensional rotation matrix used to implement spatial rotation transformations of feature points. It is a three-dimensional translation vector used to realize the spatial translation transformation of feature points;

[0063] By continuously optimizing the rotation matrix and translation vector through the iterative nearest point algorithm until the root mean square error between corresponding feature points reaches a preset threshold, accurate spatiotemporal registration of multi-source data is completed, ensuring the consistency of different types of data in spatial location.

[0064] Point cloud data statistical filtering denoising technology is based on statistical principles to identify and remove outlier noise points in point cloud data. It determines whether a point is an outlier by calculating the average spatial distance between each point and its neighbors. The core calculation formula is as follows:

[0065] ,in, For the first The point and its The average spatial distance between neighboring points, where k is the preset number of neighboring points. For the first The three-dimensional coordinates of the points For the first The first point The three-dimensional coordinates of the nearest points The Euclidean distance between the two points;

[0066] Calculate the global mean and standard deviation of the average spatial distance of all points, identify points whose average spatial distance exceeds the global mean plus n times the standard deviation as outlier noise points and remove them, thus completing the denoising process of the point cloud data and ensuring the accuracy of the spatial geometric data.

[0067] Multimodal data normalization technology, based on linear transformation, maps multi-source data of different dimensions and orders of magnitude to a unified numerical range, eliminating the differences in magnitude between data points and laying the foundation for subsequent data fusion and analysis. The core calculation formula is:

[0068] ,in, These are the normalized data values. The original data values, This represents the minimum value in the corresponding data sequence. It represents the maximum value in the corresponding data sequence;

[0069] This linear transformation maps all types of raw data to a numerical range of 0 to 1, eliminating the impact of dimensional differences while preserving the original distribution characteristics of the data, thus achieving standardized processing of multimodal data.

[0070] In this embodiment, the digital twin construction module is the core data fusion hub of the intelligent evaluation and management system for historical building renovation results. It receives standardized multimodal renovation data output by the renovation data acquisition module and, through multi-stage in-depth processing and information fusion, constructs a digital twin carrier that can accurately represent the true state of the target historical building after renovation. It is a key bridge connecting front-end data acquisition and back-end quantitative evaluation. The digital twin construction module specifically includes:

[0071] Geometric model reconstruction unit:

[0072] Deep preprocessing of point cloud data: The spatial geometric data, i.e., spatial point cloud data, from the multimodal renovation data is received. First, point cloud filtering is performed to remove residual noise points, redundant points, and outliers from the point cloud data, retaining effective point cloud data that can truly reflect the geometric form of the building. Then, the point cloud data collected from multiple stations is accurately stitched together. Based on a unified engineering coordinate system, feature point matching and registration algorithms are used to stitch multiple segments of point cloud data into a complete point cloud model covering the entire building area.

[0073] Triangulation: Triangulation is performed on the complete point cloud data after stitching and filtering. Based on the Poisson surface reconstruction algorithm, the discrete 3D point cloud data is converted into a continuous triangular mesh model, constructing a basic geometric model that can accurately reflect the geometric form of the renovated building. During the meshing process, mesh densification is performed on key building components and detailed features, while mesh lightweighting optimization is performed on large, flat areas that are not critical to the building. This controls the data volume of the model while ensuring its geometric accuracy, thereby improving the efficiency of subsequent processing and rendering.

[0074] Model accuracy verification and optimization: The accuracy of the generated basic geometric model is verified by comparing the model with the original point cloud data, calculating the spatial distance deviation between the model surface and the corresponding point cloud data, and determining whether the model accuracy meets the preset threshold requirements; for areas where the deviation exceeds the threshold, the point cloud registration and surface reconstruction operations are re-executed to complete the model optimization, ensuring that the basic geometric model can accurately restore the real geometric form of the renovated building.

[0075] Texture mapping blending unit:

[0076] Feature point matching and camera pose calculation: The system receives the basic geometric model output from the geometric model reconstruction unit, as well as the appearance image data from the multimodal restoration data. First, feature point matching is performed between the appearance image data and the basic geometric model to extract the feature points in the image and their corresponding 3D feature points in the basic geometric model, establishing a mapping relationship between the 2D image pixel coordinates and the 3D model spatial coordinates. Based on the matched feature point pairs, camera pose calculation is performed to accurately determine the camera position, shooting angle, and intrinsic parameter matrix corresponding to each image capture, providing accurate pose information for subsequent texture mapping.

[0077] Texture mapping and coordinate transformation: Based on the calculated camera pose parameters, a texture coordinate mapping algorithm is used to map the pixel information in the multi-view high-definition images one by one to the corresponding triangular mesh faces of the basic geometric model. Each triangular mesh face is assigned a corresponding texture coordinate, establishing a one-to-one correspondence between the model mesh and the image texture. During the mapping process, for overlapping areas of multi-view images such as building facades and component details, texture weight allocation is performed. According to the image shooting distance, viewing angle verticality, and sharpness, corresponding fusion weights are assigned to the texture information of different images to ensure the consistency and sharpness of texture mapping.

[0078] Texture fusion and optimization: The texture data that has completed the initial mapping is fused and optimized. Color normalization and illumination uniformization are performed for color and illumination differences between different images to eliminate color jumps and brightness differences at texture seams. At the same time, the texture image is deblurred and denoised to improve the clarity and realism of the texture, and finally a textured geometric model with realistic appearance is generated.

[0079] Texture model verification: Verify the generated textured geometric model, check the integrity of texture mapping, smoothness of seams and color consistency. For areas with missing, misaligned or distorted textures, re-execute feature point matching and texture mapping operations to ensure that the textured geometric model can realistically reproduce the appearance and details of the renovated building.

[0080] Performance information association unit:

[0081] Component hierarchy division and spatial coordinate matching: The system receives the textured geometric model output by the texture mapping fusion unit, as well as material performance data from the multimodal repair data. First, the textured geometric model is divided into component hierarchies. According to the building structure system, the overall model is broken down into independent building components and parts. A unique identifier is assigned to each building component, and the spatial coordinate range corresponding to each component is recorded. Simultaneously, the material performance data is structurally analyzed, extracting the detection location, spatial coordinates, component name, and performance indicators corresponding to each set of material performance data, establishing the correspondence between material performance data and spatial location.

[0082] Performance data is bound to model components: Based on spatial coordinates, the parsed material performance data is bound one by one to the corresponding building components or parts in the textured geometric model; for building components that cover multiple spatial point detection data, all detection data are associated with the corresponding grid area of ​​the component according to spatial coordinates, so as to achieve a precise correspondence between material performance data and model spatial location; after binding, a corresponding attribute information set is established for each building component, and information such as material type, detection index, measured value, and detection method are all included in the component attribute set to achieve integrated storage of model geometric information and performance information;

[0083] Integrated digital twin encapsulation: After binding all material performance data with model components, the model data containing geometric shape, surface texture, and material performance information is structurally encapsulated to construct an integrated digital twin of the repaired project, encompassing geometry, texture, and material performance information. Simultaneously, a unified data index is established for the twin, enabling rapid retrieval and access from the overall model to the component level and then to specific performance data. The encapsulated digital twin of the repaired project is then synchronously transmitted to a multi-repair feature extraction module, providing a core carrier for subsequent quantitative analysis.

[0084] Furthermore, the Poisson surface reconstruction technique is based on implicit function fitting theory. By solving the Poisson equation, discrete 3D point cloud data is converted into a continuous watertight triangular mesh model. The core formula is:

[0085] ,in, For the Laplace operator, These are indicator functions used to characterize the internal and external spatial regions of the model. For gradient operators, The gradient vector field corresponding to the point cloud data is constructed from the normal information of the point cloud;

[0086] By solving the Poisson equation, an implicit surface function that can best fit the point cloud data is obtained. Then, through the isosurface extraction algorithm, a continuous triangular mesh surface is extracted from the implicit function to generate a high-precision basic geometric model. This technology can effectively handle noise and holes in discrete point cloud data. The generated mesh model has good water tightness and smoothness and can accurately restore the complex geometric shape and detailed features of historical buildings.

[0087] Texture mapping camera pose calculation is based on a pinhole camera perspective projection model, establishing a mapping relationship between 2D image pixel coordinates and 3D model spatial coordinates. The core projection formula is:

[0088] ,in, As a scale factor, and These are the two-dimensional coordinates of the image pixels. For the camera intrinsic parameter matrix, It is a three-dimensional rotation matrix. It is a three-dimensional translation vector. , , These are the coordinates of a point in the 3D model space.

[0089] Based on multiple sets of matched 2D and 3D feature point pairs, the camera's intrinsic and extrinsic parameters are iteratively optimized using the bundle adjustment method to minimize the reprojection error, thus obtaining the optimal camera pose parameters. The optimization objective formula for the reprojection error is:

[0090] ,in, To match the number of feature point pairs, The coordinates of two-dimensional feature points on the image. For three-dimensional space points The reprojected pixel coordinates are calculated using the projection formula. The reprojection error is the square of the Euclidean distance between the two points.

[0091] By iteratively optimizing the reprojection error to reach a preset threshold, accurate camera pose parameters are obtained, providing a reliable pose basis for texture mapping and ensuring that the image texture can be accurately attached to the surface of the three-dimensional geometric model.

[0092] Multi-source information spatial association technology, based on spatial bounding box matching algorithms, achieves precise binding between material performance data and model components. Its core lies in the construction and intersection testing of axis-aligned bounding boxes, as shown in the formula:

[0093] ,in, For the axis-aligned bounding box corresponding to the building component. and For components in Minimum and maximum coordinate values ​​in the axial direction, shaft and The corresponding parameters for the axes are similar;

[0094] A corresponding axis-aligned bounding box is constructed for each building component, and a corresponding spatial point bounding box is constructed for the detection location of each set of material performance data. By conducting bounding box intersection tests, the building component to which the material performance data belongs is determined, achieving accurate spatial matching between performance data and model components. This technology can quickly complete the association and binding of massive performance data with three-dimensional model components, ensuring that material performance information can be accurately mapped to specific parts of the building, and achieving integrated fusion of multi-source information.

[0095] In this embodiment, the multi-renovation feature extraction module is the core quantitative analysis hub of the intelligent evaluation and management system for historical building renovation results. It is the key link in transforming historical building renovation results from a three-dimensional digital twin to standardized, measurable quantitative indicators. Furthermore, it serves as the quantitative execution carrier anchoring the three core principles of historical building renovation: preserving the original state of the cultural relic, minimal intervention, and authenticity protection. The module uses the completed digital twin of the renovation as the sole analysis object. Through multi-dimensional, multi-level, and comprehensive quantitative analysis and calculation, it generates a structured multi-renovation feature vector that can be directly used for intelligent evaluation. This provides comprehensive core data support for the subsequent integrated evaluation engine's comprehensive quality level determination. The multi-renovation feature extraction module specifically includes:

[0096] Geometric deviation quantization unit:

[0097] Tiered Lightweighting and Semantic Decomposition of Twin Mesh: After receiving the completed digital twin of the repaired structure, the geometric deviation quantization unit first performs tiered lightweighting on the triangular mesh model of the twin. During the process, the twin is first semantically decomposed into components. According to the building structure system and the importance of protection, the model is divided into three categories: main structural components, decorative feature components, and detailed node components. For main structural components and detailed node components, the original mesh accuracy is retained, and no lightweighting operation is performed to ensure that the geometric features of the core structure and key nodes are completely preserved. For non-critical large-area flat components, moderate lightweighting is performed. The number of meshes is simplified through the edge folding algorithm, and a mesh simplification threshold is set to ensure that the geometric deviation between the simplified model and the original model does not exceed the preset accuracy limit. After the lightweighting process is completed, the set of geometric feature surfaces to be measured, covering all repaired parts, is extracted from the tiered model. The set of geometric feature surfaces to be measured is simultaneously set with hierarchical labels according to the component classification to provide a basis for subsequent weight allocation.

[0098] Dual-model coarse registration and coordinate system one: After extracting the set of geometric feature surfaces to be measured, the pre-stored digital model of the repair design is retrieved, and the set of design geometric feature surfaces corresponding one-to-one with the set of geometric feature surfaces to be measured is extracted from the design model; firstly, the dual-model coarse registration operation is performed. Through the scale-invariant feature transformation algorithm, the key feature points in the two sets of feature surfaces are extracted, and feature point matching pairs are established. Based on the matching pairs, the initial three-dimensional rotation matrix and three-dimensional translation vector are solved, and the two sets of feature surfaces are transformed to the same engineering coordinate system, completing the initial coordinate system one; after the coarse registration is completed, the registration accuracy is checked, and the average reprojection error of the matching feature point pairs is calculated. If the error exceeds the preset threshold, the feature point extraction and matching operation is re-executed until the coarse registration accuracy meets the requirements;

[0099] Fine registration and optimal alignment of the two models: After coarse registration, fine registration of the two models is performed. The iterative nearest point algorithm is used to finely register and align the two sets of feature surfaces. The algorithm continuously adjusts the 3D rotation matrix and 3D translation vector through iterative optimization to minimize the Euclidean distance between corresponding points of the geometric feature surface set to be measured and the design geometric feature surface set until the algorithm converges to the preset accuracy threshold, thus completing the accurate spatial alignment of the two models. After fine registration, full model registration accuracy verification is performed. No less than one thousand sets of corresponding feature points are randomly selected from the model, and the spatial distance deviation after registration is calculated to ensure that the deviation values ​​of more than 95% of the feature points are within the preset accuracy limit. If the requirements are not met, the fine registration operation is re-executed until the registration accuracy is fully met.

[0100] Full Feature Surface Deviation Calculation and Initial Dataset Construction: After achieving precise alignment of the two models, refined deviation calculations are performed on each corresponding feature surface. For each set of corresponding feature surfaces, two core deviation indices are calculated: the first is spatial distance deviation, calculated by using the shortest distance method from a point to a surface to determine the vertical distance from each sampling point on the feature surface to be tested to the corresponding design feature surface, resulting in a distance deviation sequence for that feature surface; the second is angular deviation, calculated by determining the angle between the normal vector of the feature surface to be tested and the normal vector of the corresponding design feature surface, resulting in an angular deviation value for that feature surface. The distance deviation sequences and angular deviation values ​​of all feature surfaces are then fully summarized to form the initial deviation dataset. The initial deviation dataset is synchronously associated with the component classification labels of the corresponding feature surfaces, providing a foundation for subsequent weighted calculations.

[0101] Outlier Removal and Effective Dataset Optimization: Statistical analysis and outlier handling are performed on the initial deviation dataset. First, the Grubbs criterion is used to identify outliers in the dataset, setting a 95% confidence interval. Values ​​exceeding this interval are considered outliers. Outliers mainly originate from scanning noise and abrupt changes caused by local non-construction factors; these are removed after identification. After outlier removal, dataset integrity verification is performed to ensure that the amount of effective data after removal is not less than 98% of the original dataset. If the data loss exceeds the limit, outlier identification and filtering are repeated to avoid excessive removal leading to data distortion. Upon successful verification, an effective deviation dataset that accurately reflects the geometric deviations caused by the repair work is obtained.

[0102] Multi-indicator hierarchical weighted fusion and feature value generation: Based on the effective deviation dataset, three core quantitative indicators are calculated respectively. The first is the overall average geometric deviation, which is obtained by taking the arithmetic mean of all distance deviation values ​​in the effective deviation dataset. It is used to characterize the average deviation between the overall geometric shape of the building and the design model. The second is the local maximum geometric deviation of key nodes, which is obtained by extracting the maximum value of the deviation data corresponding to the main structural components and detailed node components. It is used to characterize the maximum deviation between the core protected parts of the building and the key stress nodes. The third is the uniformity of the overall deviation distribution, which is obtained by calculating the standard deviation of all deviation values ​​in the effective deviation dataset. It is used to characterize the dispersion of the deviation values ​​across the entire building range. The smaller the standard deviation, the more uniform the deviation distribution.

[0103] After calculating the three core indicators, they are weighted and fused according to preset hierarchical weighting coefficients to generate geometric deviation feature values. The weighting coefficients are dynamically adapted based on the protection level of the target historical building; the higher the protection level of the building, the higher the weighting coefficient corresponding to the maximum local geometric deviation of the key node. The core formula for weighted fusion is:

[0104] ,in, The geometric deviation characteristic value, The overall average geometric deviation, This represents the maximum local geometric deviation at the critical node. For the overall uniformity of deviation distribution, The preset weighting coefficients corresponding to the overall average geometric deviation. The preset weighting coefficients are the local maximum geometric deviations at critical nodes. The preset weighting coefficients are the uniformity of the overall deviation distribution, and satisfy the following conditions: ;

[0105] After generating geometric deviation feature values, a numerical rationality check is performed to ensure that the feature values ​​are within the preset reasonable value range. If they exceed the range, a recalculation process is triggered. After the check is passed, the feature values ​​and detailed calculation data are synchronously transmitted to the feature integration unit.

[0106] Material performance evaluation unit:

[0107] Full component data traversal and structured analysis: After receiving the completed digital twin of the repaired building, the material performance evaluation unit performs a full traversal of the twin, enumerating all building components bound with material performance data one by one. It extracts the material type, testing location, testing method, and measured values ​​of various material properties at each location for each building component without omission. After extraction, all data is structured and analyzed, categorized according to the protection level, material type, and spatial location of the building components. An independent material performance data file is established for each component, along with a spatial index between components and data, ensuring that each set of measured data accurately corresponds to a specific building component and spatial location. After data processing, a data integrity check is performed to ensure that all components in key repair areas have corresponding material performance test data without missing data. If missing data is found, a data completion reminder is triggered. After successful verification, the process proceeds to the next stage.

[0108] Layered Retrieval and Matching of Historical Standard Benchmark Database: Based on the historical age, architectural style, regional architectural style, and material and technological characteristics of the target historical building, a layered retrieval and precise matching are conducted from a pre-set historical value assessment benchmark database. The historical value assessment benchmark database adopts a three-level index structure: the first level is the building's historical age index, the second level is the regional architectural style index, and the third level is the material type index. Through the three-level index retrieval, the historical standard performance range or benchmark value corresponding to each material type is accurately determined. For different material types, corresponding performance index systems are matched: wooden components are matched with indicators such as strength, moisture content, and degree of decay; brick and stone components are matched with indicators such as strength, density, and degree of weathering; painted components are matched with indicators such as pigment composition and adhesive material performance; and metal components are matched with indicators such as degree of corrosion and mechanical properties. After matching, the corresponding historical standard performance range is determined for each performance index of each component, providing a unified and compliant comparison benchmark for subsequent compliance calculations.

[0109] Single-point compliance calculation and confidence level correction: Each extracted measured value of material performance is compared item by item with the corresponding historical standard performance range to calculate the single-point compliance of each performance index. The single-point compliance value is positively correlated with the degree of fit between the measured value and the standard range. When the measured value falls completely within the standard range, the single-point compliance value is set to its maximum value of 1. The further the measured value deviates from the standard range, the lower the single-point compliance value. When it exceeds the critical range, it is set to 0. The core formula for calculating the single-point compliance is:

[0110] ,in, For the first The first component Single-point compliance of performance indicators For the first The first component Measured values ​​of the performance indicators For the first The lower limit of the historical standard performance range corresponding to each performance indicator. For the first The upper limit of the historical standard performance range corresponding to each performance indicator;

[0111] After completing the basic single-point compliance calculation, the single-point compliance is corrected based on the confidence level of the material performance testing methods. The confidence level of the testing methods is determined according to the accuracy of the testing method, with laboratory sampling testing having the highest confidence level, followed by on-site non-destructive testing, and surface observation testing having the lowest confidence level. The corrected single-point compliance calculation formula is as follows:

[0112] ,in, For the revised first The first component Single-point compliance of performance indicators For the first The first component The confidence coefficient of the corresponding performance index for the testing method ranges from 0 to 1.

[0113] Component Value Importance Classification and Weight Allocation: Based on the importance of architectural components in the overall value composition of historical buildings, all architectural components are classified into three levels of value importance: Level 1 consists of core value components, including original components that carry the core historical value of the building, iconic decorative components, and key structural load-bearing components; Level 2 consists of important components, including main building enclosure components, general decorative components, and secondary structural components; Level 3 consists of general components, including non-protective ancillary components added during repairs and non-core supporting components.

[0114] Based on the importance level of the components, corresponding weight coefficients are assigned to different building components, with core value components having the highest weight coefficient, followed by important components, and general components having the lowest. The setting of weight coefficients strictly follows the principle of prioritizing the protection of cultural relics with core value, ensuring that the material performance of core value components plays a dominant role in the final evaluation results.

[0115] Component-level compliance fusion and outlier marking: For each building component, the arithmetic mean of the single-point compliance of all performance indicators after correction is calculated to obtain the comprehensive material performance compliance value of the component; after the comprehensive value calculation of all components is completed, outlier identification is performed, and components whose comprehensive value is lower than the preset threshold are marked as outliers. The location, material type, measured data and compliance value of the component are recorded simultaneously to provide a basis for subsequent problem localization; after the outlier marking is completed, the comprehensive compliance values ​​of all components and the corresponding weight coefficients are summarized and entered into the overall fusion calculation stage;

[0116] Weighted Fusion and Feature Value Generation for the Entire Building: A weighted average is calculated by taking the comprehensive material performance compliance values ​​of all building components and their corresponding weighting coefficients to obtain the overall material performance score of the entire building. This score is then used as the material performance compliance feature value. The core calculation formula is:

[0117] ,in, This represents the characteristic value of the material's performance conformity. The total number of building components. For the first The weighting coefficient of the value importance of each component For the first The overall value of the material performance compliance of each component;

[0118] After generating the material performance compliance characteristic value, a numerical rationality check is performed to ensure that the characteristic value is within a reasonable range of 0 to 1. At the same time, the component data marked with anomalies is checked a second time. After confirming that there are no errors, the characteristic value, detailed calculation data and abnormal component information are synchronously transmitted to the feature integration unit.

[0119] Value intervention quantification unit:

[0120] Full traversal of the digital twin and identification of core value elements: After receiving the completed digital twin of the renovation project, the value intervention quantification unit calls the preset value element identification model to perform a full traversal and pixel-level analysis of the twin. The value element identification model is an instance segmentation model based on deep learning, trained through a massive dataset of labeled core value elements of historical buildings, which can accurately identify various core value elements contained in the twin. Core value elements include original historical components, traces of traditional craftsmanship, key decorative patterns, architectural details with characteristics of the era, and structural nodes that preserve historical information.

[0121] After identification, each core value element is spatially located, its boundaries are defined, and its identity is marked to establish a complete list of core value elements. The list simultaneously records the type, spatial location, boundary range, and volume data of each element. After the list is established, the identification results are verified by comparing them with the pre-renovation survey report and historical archive data to ensure that no core value elements are missed or incorrectly identified. After the verification is passed, the process proceeds to the next stage.

[0122] Data matching and retrieval of pre- and post-renovation status: For each identified core value element, retrieve the complete historical status information recorded in the pre-renovation historical archive data for that element; the historical archive data includes pre-renovation 3D laser point cloud data, multi-view high-definition image data, on-site survey report, historical survey drawings, and value assessment archives; accurately spatially match the current status of the element in the completed digital twin with the pre-renovation historical status to ensure that the two are in the same spatial coordinate system and that the boundary ranges are completely corresponding, providing a unified comparison benchmark for quantifying the degree of subsequent intervention;

[0123] Treatment method identification and intervention degree quantification: For each core value element, the treatment method received by the element during the repair process is analyzed by comparing the state before and after the repair. The treatment methods are divided into four categories: original preservation, repair and reinforcement, partial replacement, and overall restoration. According to different treatment methods, corresponding preset intervention coefficients are assigned. The intervention coefficient for original preservation is the lowest, and the intervention coefficient for overall restoration is the highest. The intervention coefficient is positively correlated with the degree of intervention of the treatment method on the core value element.

[0124] Simultaneously, calculate the actual treated area or volume of the core value element, its proportion to the overall volume, and combine this with the intervention coefficient to calculate the individual intervention level value for each core value element; the core calculation formula is:

[0125] ,in, Let k be the single intervention level value of the core value element at position k. For the first The preset intervention coefficient corresponding to the treatment method of the core value element. For the first The proportion of the actual processing area or volume of the core value element to its overall volume;

[0126] Calculation of core value element weight coefficient: Based on the pre-constructed value hierarchy assessment system, the weight coefficient of each core value element in the overall value composition of the target historical building is calculated; the value hierarchy assessment system comprehensively scores from three core dimensions, namely value scarcity, historical representativeness, and spatial location significance; the score range for each dimension is 0 to 1, and the comprehensive value score of the core value element is obtained by weighted summation of the scores of the three dimensions. The higher the comprehensive value score, the greater the corresponding weight coefficient.

[0127] The core formula for calculating the value weighting coefficient is:

[0128] ,in, For the first The value weight coefficient of the core value element For the first The value scarcity score of the core value element For the first Historical representativeness score of core value elements For the first Spatial location salience score of core value elements, The weighting coefficient corresponding to the scarcity of value. The weighting coefficients corresponding to historical representativeness. Let be the weighting coefficient corresponding to the spatial location significance, and satisfy . m represents the total number of identified core value elements;

[0129] Overall Intervention Level Weighted Accumulation and Anomaly Marking: The individual intervention level value of each core value element is weighted and accumulated with its corresponding value weight coefficient to obtain the overall intervention accumulation value of the core value elements of the building for this renovation project; after the accumulation calculation is completed, anomaly identification is performed, and core value elements whose individual intervention level values ​​exceed the preset threshold are marked as anomalies. The location, type, processing method and intervention level value of the element are recorded simultaneously to provide accurate basis for subsequent renovation compliance judgment;

[0130] Normalization and Feature Value Generation: Based on a preset value intervention tolerance threshold range, the overall cumulative intervention value is normalized, mapping the value to a standardized range of 0 to 1. The normalization process uses a reverse mapping logic: the lower the overall cumulative intervention value, the less the repair behavior interferes with the core value, and the closer the final value intervention index feature value is to 1; conversely, the higher the value, the closer it is to 0. The core calculation formula is:

[0131] ,in, The characteristic value of the value intervention index. The minimum value of the preset cumulative value of value intervention. The maximum value of the preset value intervention accumulation;

[0132] The tolerance threshold range for value intervention is dynamically adapted based on the protection level of the target historical building. The higher the protection level of the building, the stricter the tolerance threshold range and the lower the maximum allowable cumulative intervention value. After generating the value intervention index feature value, a numerical rationality check is performed to ensure that the feature value is within a reasonable range of 0 to 1. At the same time, the core value element data of the anomaly marker is reviewed a second time. After confirming that there are no errors, the feature value, detailed calculation data and anomaly element information are synchronously transmitted to the feature integration unit.

[0133] Feature integration and standardized verification unit:

[0134] Data validity verification: After receiving the geometric deviation characteristic value, material performance compliance characteristic value, and value intervention index characteristic value, a full validity verification of the three characteristic values ​​is first performed. The verification includes verification of the reasonableness of the numerical range, verification of the consistency of the calculation logic, and verification of the integrity of the data source. If a characteristic value is found to be outside the reasonable range, or there is a contradiction in the calculation logic, or the original data is missing, the recalculation process of the corresponding quantification unit is immediately triggered until all three characteristic values ​​pass the validity verification.

[0135] Dimensionless normalization: After verification, the three characteristic values ​​are subjected to dimensionless normalization to eliminate the dimensional and order-of-magnitude differences between different characteristic values, and to uniformly map the three characteristic values ​​to the standardized range of 0 to 1; among them, the geometric deviation characteristic value is subjected to inverse normalization, and the smaller the deviation, the closer the normalized value is to 1. The material performance compliance characteristic value and the value intervention index characteristic value are already in the range of 0 to 1, and only consistency calibration is required.

[0136] Structured Feature Vector Concatenation: Following a preset feature vector structure order, the normalized geometric deviation feature components, material performance compliance feature components, and value intervention index feature components are sequentially concatenated to generate a one-dimensional structured multi-repair feature vector. The structure and order of the feature vector strictly match the input requirements of the machine learning model in the subsequent fusion evaluation engine, ensuring that it can be directly input into the model for calculation. After concatenation, the feature vector is format-verified. Once confirmed to be correct, the feature vector, along with the detailed calculation data of the three feature values, is synchronously output to the fusion evaluation engine, while local data backup and archiving are also completed.

[0137] In this embodiment, the fusion evaluation engine is the core intelligent decision-making hub of the intelligent evaluation and management system for historical building renovation results. It is a key link in realizing the transformation from multi-dimensional quantitative renovation characteristics to standardized comprehensive quality level judgment, and it is also the core data processing and decision-making bridge connecting the front-end multi-dimensional renovation feature extraction module and the back-end visual interaction and management terminal. Through standardized data processing procedures and pre-trained machine learning models, it transforms scattered multi-dimensional feature indicators into intuitive and practical comprehensive renovation quality levels, while outputting interpretable feature contribution information, providing core decision-making basis for the final acceptance, compliance judgment and archiving management of historical building renovation results.

[0138] The fusion evaluation engine specifically includes:

[0139] Data receiving and preprocessing unit:

[0140] Multi-source data reception; the unit synchronously receives geometric deviation feature values ​​from the geometric deviation quantification unit, material performance compliance feature values ​​from the material performance evaluation unit, and value intervention index feature values ​​from the value intervention quantification unit. At the same time, it synchronously receives detailed calculation process data and original measured support data corresponding to the three feature values, ensuring that all input data is traceable and verifiable.

[0141] Full data validity verification; the unit performs full-dimensional validity verification on the three received feature values, including numerical integrity verification, numerical range reasonableness verification, and outlier identification verification; numerical integrity verification is used to check whether there are missing feature values, numerical range reasonableness verification is used to confirm whether the feature values ​​are within the preset reasonable numerical range, and outlier identification verification is used to identify abnormal values ​​that exceed the normal fluctuation range; if a problem is found in the data during the verification process, the unit will immediately trigger the recalculation process of the multi-repair feature extraction module until all data passes the validity verification.

[0142] Dimensionless normalization processing: The unit performs dimensionless normalization processing on the three verified feature values ​​to eliminate dimensional and order-of-magnitude differences between different feature values, uniformly mapping all feature values ​​to a standardized numerical range of 0 to 1. Among them, the geometric deviation feature value is subjected to inverse normalization processing, the smaller the deviation value, the closer the normalized value is to 1; the material performance compliance feature value and the value intervention index feature value are subjected to forward normalization processing, the higher the value, the closer the normalized value is to 1; after processing, the unit outputs the normalized geometric deviation feature component, material performance compliance feature component, and value intervention index feature component.

[0143] Feature vector structured building blocks:

[0144] The feature components are ordered in a fixed order; the unit arranges the three normalized feature components in an orderly manner according to the preset feature vector structure order. The order of arrangement is completely consistent with the sample feature order in the model training stage, ensuring the structural consistency of the model input and avoiding model inference errors caused by disordered feature order.

[0145] Structured vector concatenation: The unit performs one-dimensional concatenation and combination of the three feature components that have been sorted in a fixed order to generate a structured multi-dimensional repair feature vector. The vector dimension is fixed to three dimensions, which is completely matched with the input dimension of the model training samples, ensuring that the model can correctly recognize and process the input feature vector.

[0146] Vector format compliance verification: The unit performs a full format verification on the generated structured multi-maintenance feature vectors. The verification includes vector dimension verification, numerical range verification, and data format verification to confirm that the vectors fully comply with the input specifications of the machine learning model. After the verification passes, the unit transmits the structured multi-maintenance feature vectors to the machine learning model inference unit. If the verification fails, the feature component sorting and concatenation operation is re-executed until the vectors fully comply with the requirements.

[0147] Machine learning model inference unit:

[0148] Model loading and initialization: The pre-trained machine learning model for comprehensive evaluation of repair quality is loaded into the unit, and the model's structural parameters, decision tree weights, and hyperparameter configurations are loaded simultaneously to complete the model initialization and runtime environment warm-up, ensuring that the model can stably perform forward inference calculations; This model is a multi-classification model trained using gradient boosting tree algorithm based on sample data from the historical repair case library, and has extremely strong nonlinear fitting and generalization capabilities, and can accurately capture the intrinsic correlation between multi-dimensional features and repair quality levels;

[0149] The model performs forward inference computation; the unit inputs the validated structured multi-repair feature vector into the initialized machine learning model and calls the model to perform forward computation; inside the model, through the integrated computation of multiple decision trees, the input feature vector is split and matched layer by layer, each decision tree outputs the corresponding prediction residual, and finally, through the weighted summation and integrated voting of multiple trees, the predicted score of the target historical building belonging to the four preset comprehensive repair quality levels is output;

[0150] Probability distribution normalization output; The unit performs Softmax normalization on the four levels of predicted scores output by the model, converting the predicted scores into corresponding probability values, ensuring that the sum of the probability values ​​of the four levels is 1, and finally outputs the probability distribution of the target historical building belonging to the four comprehensive renovation quality levels of excellent, good, qualified and unqualified.

[0151] Assessment Level Determination Unit:

[0152] Initial selection of the highest probability level: The probability values ​​corresponding to the four comprehensive repair quality levels output by the unit traversal model are selected. The level with the highest probability value is selected as the preliminary comprehensive repair quality level. The difference between the probability value of this level and the second highest probability level is recorded simultaneously.

[0153] Secondary threshold verification: The unit performs threshold verification on the preliminary judgment result. If the difference between the highest probability and the second highest probability is less than the preset threshold, the unit will trigger a secondary verification mechanism. The unit will combine the industry standard thresholds corresponding to the three feature values ​​for manual verification to avoid misjudgment due to the probability difference being too small.

[0154] The unit conducts a compliance review of the preliminary assessment results based on the cultural relic protection level of the target historical building. For historical buildings with high protection levels, if the characteristic values ​​related to the protection of core values ​​do not meet the requirements of the regulations, even if the probability value of the corresponding level is the highest, the unit will make a compliance correction to the assessment level in accordance with the cultural relic protection regulations to ensure that the final assessment results fully comply with the relevant industry standards for the repair and protection of historical buildings.

[0155] Final grade confirmation; After the unit completes all verification and review processes, the final comprehensive repair quality grade is confirmed, and the basis for the grade determination and the verification results are recorded simultaneously to ensure that the determination process is traceable and verifiable.

[0156] Feature importance parsing unit:

[0157] Feature contribution weight calculation: The unit utilizes the feature importance analysis function built into the gradient boosting tree model to calculate the contribution weight of geometric deviation feature value, material performance compliance feature value, and value intervention index feature value to the current level judgment result based on the total gain brought by each feature when splitting all decision trees during this inference process; the feature with higher total gain has a greater impact on the final level judgment result and the corresponding contribution weight is higher.

[0158] Weight normalization and ranking: The unit performs normalization processing on the contribution weights of the three feature dimensions obtained from the calculation to ensure that the sum of the three weight values ​​is 1, clearly reflecting the proportion of each feature dimension's influence on the final result; at the same time, the unit ranks the three feature dimensions in descending order of contribution weight, clarifying the core dominant factors and secondary factors affecting the evaluation results.

[0159] Explainability information generation: The unit integrates feature contribution weights, dimension ranking results, and actual values ​​of each feature dimension to generate complete explainable analysis information, clearly explaining the degree of influence of each repair dimension on the final quality level, providing a clear basis for managers to understand the assessment results and locate the weak links in the repair project;

[0160] Data association storage and output unit:

[0161] Multi-source data association and binding: The unit will associate and bind the final confirmed comprehensive repair quality level with the corresponding geometric deviation characteristic value, material performance compliance characteristic value, value intervention index characteristic value details, structured multi-repair characteristic vector, and characteristic contribution weight interpretability information to establish a complete project evaluation data archive and ensure that the relationship between all data is clear and traceable.

[0162] Structured data storage; the unit will store all the associated and bound evaluation data in a structured manner according to the preset database structure, establish a unique index identifier for each project, support subsequent data query, retrieval, traceability and statistical analysis, and at the same time achieve long-term secure archiving of evaluation data;

[0163] Standardized data output: The unit will simultaneously output the completed digital twin of the repair, the comprehensive repair quality level, the details of multiple repair feature vectors, and the explanation of the contribution weight of each feature value to the visualization interaction and management terminal, providing complete and standardized data support for subsequent 3D visualization rendering, evaluation information overlay display, interactive detail query, and evaluation report generation.

[0164] In this embodiment, the visual interaction and management terminal is the core carrier for the intelligent evaluation and management system for historical building renovation results to realize data implementation and human-computer interaction. Through 3D visualization rendering and multi-dimensional information linkage display, it intuitively presents the completed digital twin of the renovation and the comprehensive evaluation results to the management personnel, enabling the renovation results to be searchable, viewable, traceable, and archived. The visual interaction and management terminal specifically includes:

[0165] Data receiving and parsing unit:

[0166] Multi-source data synchronous reception: Real-time synchronous reception of the completed digital twin of the repair, comprehensive repair quality level, details of multiple repair feature vectors, and explanation of the contribution weight of each feature value output by the fusion evaluation engine; synchronous reception of associated original detection data, component attribute information and evaluation process data; establishment of correlation mapping relationship between various types of data; ensuring that all data are received completely without omission.

[0167] Data parsing and format conversion: The received multi-source heterogeneous data is structured and parsed, and the inherent structure of 3D model data, texture data, attribute data, and evaluation index data is deconstructed and converted into a standardized data structure that can be recognized by the visualization rendering engine. At the same time, the parsed data is checked for consistency, and data items with incorrect format or missing content are removed to ensure the stability of subsequent rendering and display processes.

[0168] Data hierarchical index construction: According to the four-level hierarchy of the building as a whole, structural zone, individual component, and detailed features, a hierarchical retrieval index is constructed for the parsed data, establishing a fast retrieval path from macro-level comprehensive evaluation results to micro-level single component detection data, providing efficient data support for subsequent interactive detailed queries;

[0169] 3D visualization rendering unit:

[0170] 3D scene construction and model loading: Based on the parsed digital twin of the renovated building, a suitable 3D visualization scene is constructed, the initial configuration of scene lighting, initial viewpoint and engineering coordinate system is completed, and an integrated digital twin model containing geometric shape, appearance texture and attribute information is loaded to accurately restore the real spatial form and appearance details of the renovated historical building.

[0171] Real-time rendering and performance optimization: A real-time rendering engine is used to render the 3D model in full frames. Through a Level of Detail (LOD) level detail optimization algorithm, mesh precision and texture resolution are dynamically adjusted based on the distance between the model and the viewpoint. The core formula is:

[0172] ,in, The appropriate LOD level for the current model. This represents the total number of LOD levels. The distance threshold that triggers level switching, The spatial distance between the observation point and the center point of the model;

[0173] This algorithm balances rendering quality and running efficiency, ensuring the smoothness of complex architectural models during operations such as rotation, translation, scaling, and sectioning.

[0174] Multi-mode interactive operation support: It supports managers to perform interactive operations such as rotation, translation, scaling, and sectioning on the 3D model through mouse or touch. It can realize global browsing of the entire building space, and can also view the repair status of the internal structure and hidden works of the building through sectioning operation. It supports observing the full-dimensional details of building repair from any angle and scale.

[0175] Evaluation information overlay display unit:

[0176] The overall assessment information will be graphically displayed, including the comprehensive repair quality level, details of multiple repair feature vectors, and contribution weight information of each feature value. These will be overlaid on a fixed area around the 3D visualization scene or on a draggable floating panel in graphical form such as bar charts, radar charts, and progress bars, to intuitively present the overall assessment results of the repair project and the performance of each dimension indicator.

[0177] Spatial distribution visualization mapping: Based on the numerical ranges of geometric deviation characteristic values, material performance compliance characteristic values, and value intervention index characteristic values, different component parts of the 3D model are assigned corresponding color highlights, establishing a mapping relationship between index values ​​and color gradients. The core formula is:

[0178] ,in, The rendering color value corresponding to the component. The base color corresponding to the lower limit of the numerical range. The base color corresponding to the upper limit of the numerical range. The feature values ​​of the corresponding dimension of the component. This is the lower limit of the numerical range of this eigenvalue. This represents the upper limit of the numerical range of the eigenvalue;

[0179] The color gradient visually reflects the distribution of geometric deviations, material performance compliance, and value intervention throughout the building space, helping managers to quickly locate and repair weak areas.

[0180] Anomaly warning prompt: For abnormal components with geometric deviations exceeding the threshold, material performance compliance not meeting the standard, or value intervention exceeding the standard, flashing highlight and location marking are performed in the 3D model. At the same time, the number, location and anomaly type of abnormal components are displayed in the floating panel to realize intuitive warning and rapid location of abnormal information.

[0181] Interactive details query unit:

[0182] Component selection and interactive response: Real-time capture of managers' selection operations on specific components or parts in the 3D model; based on ray collision detection algorithm, determine the target component corresponding to the selection operation, trigger the corresponding details query process, and quickly respond to the user's interactive commands;

[0183] Multi-level detailed linkage display: For the selected target component, query and display the full-dimensional detailed information of the component, including the component's basic attributes, original data of repair and inspection, measured values ​​of material performance, geometric deviation values ​​from the design model, details of intervention of core value elements, and the degree of contribution of the component to the overall evaluation level, realizing multi-level information linkage display from the overall evaluation results to the details of individual components;

[0184] Multi-component comparison query support: Supports the simultaneous selection of multiple similar components, synchronously displaying the repair data and evaluation index comparison information of each component, clearly presenting the differences in repair quality of different components, and providing comprehensive data support for managers to analyze the local construction quality of repair projects;

[0185] Assessment report generation and archiving unit:

[0186] Customizable report content integration: Based on the instructions of the management personnel, the system automatically integrates the currently displayed 3D view of the completed digital twin of the repair, the comprehensive repair quality level, details of multiple repair feature vectors, analysis information on the contribution weight of each feature value, details of abnormal components, and detailed information of a single component queried by the user, and completes the structured integration of the report content according to the preset standard report framework.

[0187] Standard format report generation: Based on the integrated content, a standard format assessment report that conforms to the industry standards for the renovation of historical buildings is automatically generated. The report includes complete chapters such as an overview of the renovation project, data collection instructions, digital twin construction, multi-dimensional assessment and analysis, comprehensive quality level determination, and suggestions for problem rectification, ensuring the professionalism and compliance of the report.

[0188] Multi-mode export and archiving management: Provides multi-format export function for reports, supports exporting the generated assessment report to a common file format, and supports the association and binding of the report with the corresponding digital twin of the completed repair, all original test data, and assessment process data, performs online encrypted archiving, and establishes a traceable repair results archive for managers to access, call and manage for a long time.

[0189] A smart evaluation and management method for the results of historical building restoration, the method is executed by a system and includes the following steps:

[0190] Step S1: Standardized collection of multimodal renovation data: For the target historical building that has been renovated, spatial geometric data, appearance image data and material performance data are acquired by 3D laser scanning, multi-view high-definition image shooting and on-site / laboratory material performance testing, respectively; then the three types of data are preprocessed by spatiotemporal registration, noise reduction and normalization to form standardized multimodal renovation data after registration and alignment.

[0191] Step S2: Integrated Construction of the As-built Digital Twin of the Repaired Building: Deeply integrate the multimodal repair data. First, generate a high-precision basic geometric model through point cloud filtering, stitching, and triangulation. Then, through feature point matching and texture mapping, attach the high-definition image to the model surface to generate a textured geometric model. Finally, bind the material performance data to the corresponding building components based on spatial coordinates to construct an integrated digital twin of the repaired building, which fully represents the real state of the building after repair.

[0192] Step S3: Quantitative Extraction and Structured Output of Multiple Repair Features: Using a digital twin as the core carrier, quantitative analysis is completed from three core dimensions to generate a structured multiple repair feature vector.

[0193] Geometric deviation quantification: The twin is registered and compared with the digital model of the renovation design. After removing outliers, the geometric deviation feature values ​​are generated by fusion.

[0194] Material performance evaluation: Traverse the measured data of component material performance, compare it with the standard values ​​in the historical value assessment benchmark library, and combine the component value weighted fusion to generate material performance compliance characteristic values;

[0195] Value intervention quantification: Identify the core value elements of a building, compare the state before and after renovation to quantify the degree of intervention, and combine the element value weights to accumulate and normalize to generate the value intervention index feature value;

[0196] Step S4: Intelligent Fusion Assessment Based on Machine Learning: The three types of feature values ​​are dimensionlessly normalized and concatenated in a preset order to form a structured multi-class repair feature vector. This vector is then input into a pre-trained multi-classification model for comprehensive repair quality evaluation, constructed based on the gradient boosting tree algorithm. Through forward computation of the model, the probability distribution of the building corresponding to four levels—excellent, good, qualified, and unqualified—is output. The level corresponding to the highest probability value is selected as the final comprehensive repair quality level. Simultaneously, the contribution weight of each feature value to the evaluation result is calculated, and interpretability analysis information is output.

[0197] Step S5: Visual Interaction and Evaluation Result Management: Real-time 3D visualization rendering of the digital twin, supporting interactive operations such as rotation, translation, scaling, and sectioning; graphical overlay display of information such as evaluation level, feature vector details, and contribution weight; color highlighting and anomaly warning for components based on indicator values; and enabling linked query of component-level details in response to user operations; finally, it can integrate data according to instructions to generate a standardized renovation result evaluation report that conforms to industry standards, and provide export and online encrypted archiving functions to achieve full lifecycle traceability management of renovation evaluation results.

[0198] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent evaluation and management system for the results of historical building restoration, characterized in that: include: The renovation data acquisition module is used to acquire multimodal renovation data of the target historical building after renovation is completed; the multimodal renovation data includes at least spatial geometric data, appearance image data and material performance data; The digital twin construction module, connected to the renovation data acquisition module, is used to fuse multimodal renovation data and construct a completed digital twin that represents the true state of the target historical building after renovation. The multi-repair feature extraction module, connected to the digital twin construction module, is used to perform multi-dimensional quantitative analysis on the completed digital twin of the repair project and generate structured multi-repair feature vectors. The multi-repair feature extraction module includes: The geometric deviation quantization unit is used to extract the geometric features of the repaired digital twin and compare it with the preset repair design digital model to generate geometric deviation feature values. The material performance evaluation unit is used to extract the associated material performance data from the digital twin of the repaired project and compare it with the standard values ​​in the preset historical value assessment benchmark library to generate material performance compliance feature values. The value intervention quantification unit is used to receive the completed digital twin of the renovation project, identify the various core value elements contained therein, calculate the degree of intervention for each core value element based on the preset intervention coefficient corresponding to the treatment method of each core value element, and combine the proportion of the actual treated area or volume of the core value element to its overall volume. It also determines the weight coefficient of each core value element according to the pre-constructed value hierarchy evaluation system, which includes a comprehensive score of at least three dimensions: value scarcity, historical representativeness, and spatial location significance. Finally, it weights and accumulates the degree of intervention for each core value element with its weight coefficient and normalizes the result to generate the value intervention index feature value. The integrated evaluation engine connects the geometric deviation quantification unit, the material performance evaluation unit, and the value intervention quantification unit respectively. It is used to combine the geometric deviation feature value, the material performance compliance feature value, and the value intervention index feature value to form a structured multi-maintenance feature vector. The structured multi-maintenance feature vector is then input into a pre-trained comprehensive maintenance quality evaluation machine learning model for processing to obtain the comprehensive maintenance quality level of the target historical building. The visual interaction and management terminal connects to the integrated evaluation engine to present the completed digital twin of the repair in a visual manner, and displays the comprehensive repair quality level and its corresponding multi-repair feature vector details for managers to view and archive.

2. The intelligent evaluation and management system for the results of historical building renovation as described in claim 1, characterized in that: The renovation data acquisition module is used to acquire multimodal renovation data of the target historical building after renovation is completed; specifically including: The three-dimensional laser scanning unit is used to perform three-dimensional laser scanning on the target historical building to obtain spatial point cloud data after the renovation is completed, which serves as spatial geometric data. The high-definition image acquisition unit is used to capture high-definition images of the target historical building from multiple perspectives and obtain surface image data. The material performance testing unit is used to conduct on-site non-destructive testing or sampling laboratory testing on key parts of the target historical building to obtain material performance data. The data synchronization and preprocessing unit is connected to the 3D laser scanning unit, the high-definition image acquisition unit, and the material performance testing unit, respectively. It is used to perform spatiotemporal registration, denoising, and normalization processing on the acquired spatial point cloud data, appearance image data, and material performance data to form multimodal repair data after registration and alignment, and then transmit it to the digital twin construction module.

3. The intelligent evaluation and management system for the results of historical building renovation as described in claim 1, characterized in that: The digital twin construction module is used to fuse multimodal renovation data to construct a completed digital twin representing the actual state of the target historical building after renovation; specifically including: The geometric model reconstruction unit is used to preprocess the spatial geometric data in the multimodal renovation data, including point cloud filtering, stitching and triangulation, to generate a high-precision basic geometric model that reflects the geometric form of the building after renovation. The texture mapping and fusion unit, connected to the geometric model reconstruction unit, is used to receive the appearance image data from the basic geometric model and multimodal repair data. Through feature point matching and camera pose calculation, the multi-view high-definition images are mapped onto the surface of the basic geometric model. After texture fusion and optimization, a textured geometric model with realistic appearance texture is generated. The performance information association unit, connected to the texture mapping fusion unit, is used to receive material performance data from the textured geometric model and multimodal repair data. Based on the spatial location coordinates, the material performance data is bound one by one to the corresponding building components or parts in the textured geometric model, forming a repair completion digital twin that integrates geometry, texture and material performance information.

4. The intelligent evaluation and management system for the results of historical building renovation as described in claim 1, characterized in that: The geometric deviation quantization unit is used to extract the geometric features of the repaired digital twin and compare it with the preset repair design digital model to generate geometric deviation feature values. Specifically, it includes: The geometric deviation quantization unit receives the completed digital twin of the renovation project and performs overall mesh lightweighting on the completed digital twin to extract the set of geometric feature surfaces to be measured, which reflects the actual form of the building after renovation. The set of geometric feature surfaces to be measured is spatially registered and aligned with the corresponding set of design geometric feature surfaces in the pre-stored digital model of the renovation design, so that the two are in the same spatial coordinate system. Based on the registered and aligned set of geometric feature surfaces to be measured and the set of design geometric feature surfaces, the spatial distance deviation and angular deviation between each corresponding feature surface are calculated to form an initial deviation dataset. Statistical analysis was performed on the initial deviation dataset to remove outlier deviation values ​​caused by scanning noise or local abnormal mutations, resulting in an optimized effective deviation dataset. Based on the effective deviation dataset, the overall average geometric deviation, the local maximum geometric deviation at key nodes, and the overall deviation distribution uniformity were calculated. The overall average geometric deviation, the local maximum geometric deviation, and the overall deviation distribution uniformity were then weighted and fused to generate geometric deviation feature values ​​that characterize the degree of consistency between the renovated building geometry and the design intent.

5. The intelligent evaluation and management system for the results of historical building renovation according to claim 4, characterized in that: The material performance evaluation unit is used to extract the associated material performance data in the repaired digital twin and compare it with the standard values ​​in the preset historical value assessment benchmark library to generate material performance compliance feature values. Specifically, it includes: The material performance evaluation unit receives the repaired digital twin, traverses all building components in the repaired digital twin that are bound with material performance data, and extracts the material type, detection location, and measured material performance value at the corresponding location for each building component. Based on the historical period, architectural style, and material and technological characteristics of the target historical building, the historical standard performance range or benchmark value corresponding to each material type is retrieved and determined from the pre-set historical value assessment benchmark database. Each extracted material performance measured value is compared with the corresponding historical standard performance range or benchmark value. Based on the degree to which the measured value falls within the standard range or the degree of closeness to the benchmark value, the single-point compliance of each performance index is calculated, and the single-point compliance is corrected in combination with the confidence level of the material performance testing method. Based on the importance of architectural components in historical buildings, corresponding weighting coefficients are assigned to the single-point conformity of different architectural components. The weighted single-point compliance of all building components is integrated and statistically analyzed to calculate the overall material performance comprehensive score, which serves as the material performance compliance characteristic value representing the degree of compliance between the performance of the renovated building materials and the requirements of historical authenticity.

6. The intelligent evaluation and management system for the results of historical building renovation according to claim 5, characterized in that: The value intervention quantification unit specifically includes: The value intervention quantification unit receives the completed digital twin of the renovation project, calls the preset value element identification model to traverse and analyze the completed digital twin of the renovation project, and identifies the various core value elements contained therein; the core value elements include at least original historical components, traces of traditional craftsmanship, key decorative patterns and architectural details with characteristics of the times; For each identified core value element, the value intervention quantification unit compares the completed digital twin of the core value element with the historical status information recorded in the historical archive data before the renovation, and quantifies the treatment methods and extent that the core value element was subjected to during this renovation process; the treatment methods include at least original preservation, repair and reinforcement, partial replacement or overall restoration. The value intervention quantification unit calculates the individual intervention level value of each core value element based on the preset intervention coefficient corresponding to each treatment method and the proportion of the actual treatment area or volume of the core value element to its overall volume. Based on the pre-constructed value hierarchy assessment system, determine the weight coefficient of each core value element in the overall value composition of the target historical building; the value hierarchy assessment system includes a comprehensive score of at least three dimensions: value scarcity, historical representativeness, and spatial location prominence. The individual intervention level value of each core value element is weighted and accumulated with its corresponding weight coefficient. The accumulated result is then normalized according to the preset value intervention tolerance threshold range to generate a value intervention index feature value that characterizes the overall impact of this renovation project on the core value elements of the historical building.

7. The intelligent evaluation and management system for the results of historical building renovation according to claim 6, characterized in that: The fusion evaluation engine combines geometric deviation feature values, material performance compliance feature values, and value intervention index feature values ​​to form a structured multi-repair feature vector. This structured multi-repair feature vector is then input into a pre-trained machine learning model for comprehensive repair quality evaluation to obtain the overall repair quality level of the target historical building. Specifically, this includes: The fusion evaluation engine first receives geometric deviation feature values ​​from the geometric deviation quantification unit, material performance compliance feature values ​​from the material performance evaluation unit, and value intervention index feature values ​​from the value intervention quantification unit. It then performs dimensionless normalization on the three received feature values ​​to eliminate differences in dimensions and orders of magnitude, resulting in normalized geometric deviation feature components, material performance compliance feature components, and value intervention index feature components. Finally, it concatenates and combines these three normalized feature components according to a preset feature vector structure order to generate a structured multi-maintenance feature vector. The generated structured multi-repair feature vectors are input into a pre-trained machine learning model for comprehensive evaluation of repair quality. The machine learning model for comprehensive evaluation of repair quality is a multi-classification model trained using the gradient boosting tree algorithm based on sample data from a historical repair case library. The sample data includes structured multi-repair feature vectors corresponding to multiple evaluated historical building repair projects and their comprehensive repair quality level labels determined by experts. During model training, the loss function is minimized through iterative optimization, and cross-validation is used to determine the model hyperparameters. Finally, a decision model that can map the input feature vectors to the comprehensive repair quality level is obtained. The fusion evaluation engine calls the machine learning model to perform forward calculation on the current structured multi-repair feature vector. The machine learning model outputs the probability distribution of the target historical building belonging to several preset comprehensive repair quality levels through integrated voting or weighted summation of multiple decision trees. The preset comprehensive repair quality levels include at least four levels: excellent, good, qualified, and unqualified. The integrated assessment engine selects the level with the highest probability value as the comprehensive renovation quality level of the target historical building based on the probability distribution of the output. It also stores the details of this level and the corresponding geometric deviation characteristic value, material performance compliance characteristic value, and value intervention index characteristic value in association, so that they can be displayed and accessed by the visualization interaction and management terminal. Meanwhile, the integrated assessment engine also utilizes the feature importance analysis function of the machine learning model to calculate the contribution weight of geometric deviation feature values, material performance compliance feature values, and value intervention index feature values ​​to the current grade determination result. This information is output as explanatory information to help managers understand the degree of influence of each maintenance dimension on the final quality grade.

8. The intelligent evaluation and management system for the results of historical building renovation according to claim 1, characterized in that: The visualization interaction and management terminal is used to present the completed digital twin of the repair in a visual manner, and to display the comprehensive repair quality level and its corresponding multi-repair feature vector details for management personnel to view and archive; specifically including: The data receiving and parsing unit is connected to the fusion evaluation engine. It is used to receive the digital twin of the completed repair, the comprehensive repair quality level, the details of multiple repair feature vectors, and the explanation information of the contribution weight of each feature value output by the fusion evaluation engine. It also parses and converts the received data to generate the data structure required for visualization rendering. The 3D visualization rendering unit connects to the data receiving and parsing unit. It is used to construct a 3D visualization scene and perform real-time rendering based on the parsed digital twin of the completed renovation. It displays the 3D model of the target historical building after renovation on the interactive interface and supports users to rotate, translate, zoom and cut the building through mouse or touch, so as to observe the details of the building from any perspective. The assessment information overlay display unit is connected to the three-dimensional visualization rendering unit. It is used to graphically overlay the comprehensive repair quality level, details of multiple repair feature vectors and their contribution weight information on the periphery or floating panel of the three-dimensional visualization scene. According to the numerical range of each feature value, it assigns corresponding color highlight marks to different component parts of the three-dimensional model, intuitively reflecting the spatial distribution of geometric deviation, material performance compliance and value intervention degree. The interactive details query unit, connected to the evaluation information overlay display unit, is used to respond to the user's selection operation of a specific component or part in the 3D model, query and pop up to display the detailed repair and inspection data, measured values ​​of material properties, deviation values ​​from the design model, details of intervention of core value elements, and the contribution of the component to the overall evaluation level, so as to realize multi-level information linkage display from the whole to the part; The assessment report generation and archiving unit, connected to the interactive details query unit, is used to integrate the currently displayed digital twin view of the completed repair, the comprehensive repair quality level, details of multiple repair feature vectors, the contribution weight of each feature value, and the detailed information of the components queried by the user, to generate a standard format repair result assessment report based on user instructions. It also provides export and online archiving functions for managers to view and store for a long time.

9. A method for intelligent evaluation and management of the results of historical building restoration, wherein the method is executed by the system according to any one of claims 1-8, characterized in that: The method includes the following steps: Step S1: Obtain multimodal renovation data of the target historical building after renovation through the renovation data acquisition module; the multimodal renovation data includes at least spatial geometric data, appearance image data, and material performance data; Step S2: The multimodal renovation data is fused through the digital twin construction module to construct a completed digital twin that represents the actual state of the target historical building after renovation. Step S3: Perform multi-dimensional quantitative analysis on the completed digital twin of the repaired structure using the multi-repair feature extraction module to generate a structured multi-repair feature vector; this step specifically includes: Step S3-1: Extract the geometric features of the repaired digital twin through the geometric deviation quantization unit, perform spatial registration and alignment with the preset repair design digital model, calculate the spatial distance deviation and angle deviation between the corresponding feature surfaces, and after statistical analysis to remove outliers, fuse and generate the overall average geometric deviation, local maximum geometric deviation and overall deviation distribution uniformity as geometric deviation feature values. Step S3-2: Through the material performance evaluation unit, traverse all building components in the repaired digital twin that are bound to material performance data, extract the material type and measured material performance value of each building component, retrieve the corresponding historical standard performance range from the preset historical value assessment benchmark library, compare the measured value with the standard range item by item to calculate the single-point conformity, and combine the component value importance weight for weighted fusion to obtain the material performance conformity feature value; Step S3-3: The value intervention quantification unit calls the preset value element identification model to traverse and analyze the digital twin of the completed renovation, identify the core value elements, compare the historical archive data before renovation to quantify the treatment method and degree of each core value element in this renovation, and accumulate the value intervention index feature value according to the preset intervention coefficient and the value weight of the element, and normalize the data to obtain the value intervention index feature value. Step S4: After dimensionless normalization of the geometric deviation feature value, material performance compliance feature value, and value intervention index feature value through the fusion evaluation engine, they are spliced ​​and combined in a preset order to form a structured multi-repair feature vector. The structured multi-repair feature vector is then input into a pre-trained comprehensive repair quality evaluation machine learning model. The model outputs the probability distribution of the target historical building belonging to each preset comprehensive repair quality level through forward calculation. The level with the highest probability is selected as the comprehensive repair quality level. At the same time, the feature importance analysis function of the model is used to calculate the contribution weight of each feature value to the judgment result as explanatory information. Step S5: Receive and parse the completed digital twin of the repair project, the comprehensive repair quality level, details of multiple repair feature vectors, and explanations of contribution weights through the visual interaction and management terminal. Construct a 3D visualization scene and render and display the 3D model after repair in real time. Overlay the evaluation information graphically around the model and assign color-highlighted markers to different component parts of the model according to the numerical range of each feature value. Respond to user selection operations for components to query and display detailed inspection data and evaluation details of the component. Finally, integrate and generate a standard format repair result evaluation report according to user instructions and provide export and archiving functions.

Images