Method for detecting apparent quality of building wood structure

Abstract

The application discloses a kind of building wood structure apparent quality detection methods, it is related to ancient building detection technical field.The method first obtains the original point cloud data of scene by three-dimensional laser scanning, and generates documentary point cloud model reflecting the present situation of building after pre-processing;Subsequently, the model is deconstructed and cut to obtain component point cloud data, and reverse parameterized modeling is carried out with reference to it, to obtain standard component correction model;Then, random surface sampling and Euclidean distance ratio calculation are carried out on the two, to construct reference shape distribution curve and present situation shape distribution curve;Further, shape similarity index is calculated, and defect identification is carried out in combination with structure stability threshold and curve shape deviation characteristics.The application builds double time shape distribution curve, effectively avoids the registration error caused by complex texture in traditional method, realizes the nondestructive, quantitative detection of building wood structure apparent quality, and can provide reliable basis for subsequent maintenance decision.

Description

Technical Field

[0001] This application relates to the field of ancient building testing technology, and in particular to a method for testing the appearance quality of wooden building structures. Background Technology

[0002] Approximately 70% of China's existing ancient buildings are wooden structures. These structures are prone to surface defects such as cracks, deformation, and component detachment due to long-term natural aging, insect infestation, and climatic influences, directly impacting structural safety. Traditional inspection methods primarily rely on manual visual inspection and tapping, which is not only dependent on personnel experience and inefficient but also lacks objective quantification and may even cause secondary damage to the ancient buildings. While machine vision technology can reduce subjective errors, it is significantly affected by lighting, surface stains, and natural wood textures (such as grain and knots), exhibiting poor robustness and difficulty adapting to complex inspection scenarios.

[0003] Currently, the industry is gradually adopting a detection approach that uses multi-temporal point cloud registration and geometric difference calculation to quantitatively analyze defects. However, this point-by-point comparison-based change detection method is highly dependent on registration accuracy. Because the surface texture of ancient architectural wooden components is complex and often exhibits non-uniform local deformation, registration errors are easily caused. In particular, millimeter-level registration errors can easily mask minute crack signals, failing to meet the precise detection requirements for minor damage in ancient architectural wooden structures. Therefore, there is an urgent need for a detection method that can avoid the interference of traditional registration errors and effectively quantify differences in apparent geometric features. Summary of the Invention

[0004] This application provides a method for testing the appearance quality of building timber structures, so as to at least solve the above-mentioned technical problems existing in the prior art.

[0005] According to a first aspect of this application, a method for inspecting the appearance quality of a wooden building structure is provided, comprising: step S1, acquiring original point cloud data of the ancient wooden building structure at the site using a three-dimensional laser scanning device, and preprocessing the original point cloud data to obtain a documentary point cloud model reflecting the current state of the building; step S2, performing forward deconstruction and point cloud cutting on the documentary point cloud model to obtain component point cloud data corresponding to each independent unit, and performing reverse parametric modeling using the component point cloud data as a size reference to obtain a standard component correction model; step S3, performing random surface sampling and Euclidean distance ratio calculation on the standard component correction model and the component point cloud data to obtain a reference shape distribution curve and a current shape distribution curve; step S4, calculating the shape similarity index between the reference shape distribution curve and the current shape distribution curve; step S5, comparing the shape similarity index with a preset structural stability threshold, and combining the morphological offset characteristics of the current shape distribution curve relative to the reference shape distribution curve to identify the defect type, so as to obtain an appearance quality inspection result including defect level and maintenance strategy.

[0006] In one embodiment, step S1 includes: acquiring original data from multiple perspectives of the ancient building entity to be inspected using a 3D laser scanner to obtain an original set of station cloud data containing spatial 3D coordinate information and reflection intensity information; performing multi-station cloud registration and stitching on the original set of station cloud data to obtain a spatially continuous and complete stitched point cloud; performing statistical filtering and noise reduction processing on the complete stitched point cloud to obtain clean point cloud data; and performing uniform voxel sampling on the clean point cloud data based on a voxel grid to obtain a documentary point cloud model.

[0007] In one embodiment, step S2 includes: performing semantic rule-based component spatial semantic segmentation on the documentary point cloud model to obtain component point cloud data corresponding to each independent unit; performing robust geometric feature inverse fitting on the component point cloud data to obtain an ideal geometric parameter set; generating a three-dimensional entity based on the ideal geometric parameter set through a building information modeling engine, and assigning the three-dimensional entity a spatial coordinate system consistent with the component point cloud data to perform parametric entity model reconstruction to obtain a standard component correction model.

[0008] In one embodiment, step S3 includes: performing surface discretization unified sampling on the standard component correction model and component point cloud data to obtain an ideal sample point set and a current sample point set; performing D2 global distance distribution calculation on the ideal sample point set and the current sample point set to obtain an ideal distance set and a current distance set; and performing maximum value normalization and histogram statistics on the ideal distance set and the current distance set respectively to obtain a reference shape distribution curve and a current shape distribution curve.

[0009] In one embodiment, step S4 includes: calculating the discrete interval absolute error of the reference shape distribution curve and the current shape distribution curve to obtain the original average absolute error; introducing a preset linear amplification factor to perform error numerical sensitivity amplification correction on the original average absolute error to obtain the corrected average absolute error; and performing similarity normalization inversion mapping on the corrected average absolute error to obtain the shape similarity index.

[0010] In one possible implementation, step S4 includes: calculating the structural drift degree between the reference shape distribution curve and the current shape distribution curve; calculating the local abrupt change degree between the reference shape distribution curve and the current shape distribution curve; and performing nonlinear similarity mapping on the structural drift degree and the local abrupt change degree to obtain a shape similarity index.

[0011] In one possible implementation, step S5 includes: based on a preset structural stability critical threshold, performing threshold determination and state classification on the shape similarity index to obtain the structural stability state; when the structural stability state is abnormal, performing morphological difference feature pattern recognition on the reference shape distribution curve and the current shape distribution curve to determine the semantic label of the defect type; and querying a preset maintenance knowledge base by combining the structural stability state, the semantic label of the defect type and the shape similarity index to obtain the appearance quality detection result.

[0012] According to a second aspect of this application, a device for inspecting the appearance quality of a wooden building structure is provided, comprising: a point cloud acquisition module, used to acquire original point cloud data of the ancient wooden building structure at the site through a three-dimensional laser scanning device, and preprocess the original point cloud data to obtain a documentary point cloud model reflecting the current state of the building; a component deconstruction module, used to perform forward deconstruction and point cloud cutting on the documentary point cloud model to obtain component point cloud data corresponding to each independent unit, and to perform reverse parametric modeling using the component point cloud data as a size reference to obtain a standard component correction model; a curve generation module, used to perform random surface sampling and Euclidean distance ratio calculation on the standard component correction model and the component point cloud data to obtain a reference shape distribution curve and a current shape distribution curve; a similarity calculation module, used to calculate the shape similarity index between the reference shape distribution curve and the current shape distribution curve; and a defect identification module, used to compare the shape similarity index with a preset structural stability threshold, and combine the morphological offset characteristics of the current shape distribution curve relative to the reference shape distribution curve to identify the defect type, so as to obtain an appearance quality inspection result including defect level and maintenance strategy.

[0013] According to a fourth aspect of this application, a non-transitory computer-readable storage medium is provided storing computer instructions for causing a computer to perform the methods described in this application.

[0014] Compared with existing technologies, this application provides a method for inspecting the surface quality of wooden structures. It utilizes three-dimensional laser scanning technology to achieve non-contact inspection throughout the entire process, completely eliminating the risk of secondary damage to ancient artifacts caused by manual hammering. More importantly, by introducing the shape distribution curve of the Euclidean distance ratio as a feature descriptor, this application significantly reduces the excessive reliance of traditional detection techniques on the spatial registration accuracy of point cloud models, effectively solving the registration misalignment problem caused by the complex surface texture and non-uniform local deformation of ancient wooden components. This method of transforming geometric differences into probabilistic statistical differences not only greatly improves the detection algorithm's resistance to interference and robustness against changes in lighting, surface stains, and natural wood textures (such as knots and grain), but also allows for the sensitive capture of millimeter-level micro-cracks and deformation signals through quantified shape similarity indicators. This provides objective, accurate, and traceable data support for the preventative protection and scientific restoration of ancient wooden structures.

[0015] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent from the following description. Attached Figure Description

[0016] The above and other objects, features, and advantages of exemplary embodiments of this application will become readily apparent from the following detailed description taken in conjunction with the accompanying drawings. Several embodiments of this application are illustrated in the drawings by way of example and not limitation, in which: In the accompanying drawings, the same or corresponding reference numerals indicate the same or corresponding parts.

[0017] Figure 1 A schematic diagram illustrating the implementation process of the method for inspecting the appearance quality of building timber structures according to an embodiment of this application is shown.

[0018] Figure 2 A schematic diagram of the overall technical process of the method for inspecting the appearance quality of building timber structures according to an embodiment of this application is shown.

[0019] Figure 3 A detailed flowchart illustrating the construction process of an ancient building modification model according to an embodiment of this application is shown.

[0020] Figure 4 A schematic diagram illustrating the technical route for constructing a three-dimensional point cloud model of an ancient building according to an embodiment of this application is shown.

[0021] Figure 5 A schematic diagram of the point cloud model of the Jade Buddha Hall, acquired using a 3D laser scanner according to an embodiment of this application, is shown.

[0022] Figure 6A schematic diagram of the forward deconstruction of the Jade Buddha Hall according to an embodiment of this application is shown.

[0023] Figure 7 A schematic diagram of the construction process for establishing a standard component BIM family library according to an embodiment of this application is shown.

[0024] Figure 8 A schematic diagram of a modified model of the Jade Buddha Hall constructed based on Revit according to an embodiment of this application is shown.

[0025] Figure 9 A schematic diagram of the texture model of the Jade Buddha Hall after texture mapping is shown according to an embodiment of this application.

[0026] Figure 10 The definitions and geometric characteristics of five shape distribution functions are shown in the diagram.

[0027] Figure 11 A schematic diagram showing the shape distribution curves of brackets, columns, and their shapes under different numbers of random sampling points is presented.

[0028] Figure 12 The diagram shows three-dimensional models of four typical components: brackets, tables, columns, and walls.

[0029] Figure 13 The diagram shows the evolution of shape distribution curves for four types of component models under different numbers of sampling points.

[0030] Figure 14 The diagram shows the shape distribution curves of four types of component models at 100,000 random points.

[0031] Figure 15 A schematic diagram of 12 test models for verifying the feasibility of defect detection is shown.

[0032] Figure 16 A schematic diagram of the composition of a building timber structure appearance quality testing device according to an embodiment of this application is shown. Detailed Implementation

[0033] To further illustrate the technical means and effects adopted by this application in order to achieve the intended purpose, the following detailed description of the specific implementation methods, structures, features and effects of this application is provided in conjunction with the accompanying drawings and preferred embodiments.

[0034] Figure 1 The illustration shows a schematic flowchart of a method and system for inspecting the appearance quality of a building timber structure according to an embodiment of this application. Figure 2 A schematic diagram illustrating the overall technical flow of a method for inspecting the appearance quality of building timber structures according to an embodiment of this application is shown. Figure 1 and Figure 2As shown, this application provides a method for testing the appearance quality of building timber structures, including: S1: Obtain the original point cloud data of the ancient wooden structure on site using a 3D laser scanning device, and preprocess the original point cloud data to obtain a documentary point cloud model reflecting the current state of the building. This step aims to solve the problem that traditional manual inspection methods cannot accurately record the complex geometric shape and minute deformation of ancient buildings. By using high-precision digital methods, a digital twin base that can objectively reflect the current geometric state of the building (including current features such as cracks and deformation) is constructed.

[0035] In one specific implementation of this application, step S1 includes: acquiring original data from multiple perspectives of the ancient building entity to be inspected using a 3D laser scanner to obtain an original set of station cloud data containing spatial 3D coordinate information and reflection intensity information; performing multi-station cloud registration and stitching on the original set of station cloud data to obtain a spatially continuous and complete stitched point cloud; performing statistical filtering and noise reduction processing on the complete stitched point cloud to obtain clean point cloud data; and performing uniform voxel sampling on the clean point cloud data based on a voxel grid to obtain a documentary point cloud model.

[0036] The specific technical approach implemented is as follows: Figure 4 As shown, the process begins with the collection and planning of point cloud data for ancient architecture. Due to the complex structure of ancient buildings and their often limited by lighting, shading, and spatial form, a site survey is necessary beforehand. Based on the site lighting conditions, shading, and structural form, the station layout and scanning parameters of the 3D laser scanner are planned. When using a 3D laser scanning device (such as the Trimble X7) to collect raw data from multiple perspectives of the ancient building to be inspected, it is essential to ensure that the scanning range overlap of adjacent stations reaches more than 30% to guarantee complete data stitching. During the collection process, the instrument acquires the spatial 3D coordinate information and reflection intensity information from that perspective, generating a raw point cloud set containing data from multiple independent stations.

[0037] After acquiring the raw data, the original set of station cloud data needs to be registered and stitched together to obtain a spatially continuous and complete stitched point cloud. This process typically employs an iterative nearest-point algorithm to transform the data from each independent station to the same world coordinate system. By finding corresponding point pairs within the overlapping region, the optimal rigid transformation matrix is ​​calculated to minimize the sum of squared errors between adjacent station clouds. The optimized registration error function is then determined. The calculation is as follows: in, This represents the total number of corresponding point pairs involved in the calculation. The first point in the source point cloud (the station to be registered) One point, For the target point cloud (reference station) and The corresponding nearest point, Let the rotation matrix represent the spatial attitude transformation. To represent a translation vector of spatial displacement, Let be the Euclidean distance norm. By minimizing this error function, discrete station data are merged into a geometrically continuous, complete patchwork point cloud.

[0038] Subsequently, to remove outliers in the complete stitched point cloud caused by equipment errors, airborne dust, or transient interference, statistical filtering is required to obtain clean point cloud data. This step employs a statistical filtering algorithm for any point in the point cloud. Calculate its relationship with the nearest The average distance of each neighboring point And calculate the mean of the Gaussian distribution of the average distance to all points. and standard deviation If the average distance of a point is greater than a preset threshold, it is identified as a noise point and removed. The threshold determination logic is as follows: in, To preserve the pure point cloud data set, For the complete stitched point cloud input, For point To its The average distance between the nearest neighbors. This is a standard deviation factor used to control the noise reduction sensitivity (typically ranging from 1.0 to 3.0).

[0039] To eliminate the impact of uneven scanning density (denser near, sparser farther away) caused by varying distances during scanning on subsequent shape analysis and to reduce data redundancy, uniform voxel sampling of the clean point cloud data based on a voxel mesh is required to obtain a documentary point cloud model. This process constructs a model with a side length of... For a 3D voxel mesh (e.g., 1 mm), calculate the centroid of all points within each voxel containing points, and retain only the centroid as the representative point of that voxel. The formula for calculating the centroid is as follows: in, The coordinates of the centroid points that are ultimately preserved within the voxel, which are the constituent elements of the documentary point cloud model; The number of original points contained within a single voxel mesh. For the current voxel mesh, the first The coordinate vectors of each point. The documentary point cloud model output after the above processing ( ), refers to a three-dimensional point cloud dataset that meets the accuracy requirements, has uniform point density, and can accurately reflect the geometric status of ancient buildings at the time of detection (including real appearance features such as cracks, bends, and missing parts).

[0040] In one specific embodiment, taking the inspection of a jade Buddha temple as an example, a Trimble X7 3D scanner was used. Scanning parameters were set according to the differences between indoor and outdoor environments: the indoor scanning density was set to 0.5 mm / point, and the outdoor density to 1 mm / point. A total of 29 monitoring stations were deployed, with the overlap between adjacent stations controlled to be above 30%. Data collection was conducted during sunny, windless days with low pedestrian traffic. The scanning time per station was approximately 8 minutes, with a total time of approximately 230 minutes, collecting approximately 700 million point cloud data points. In the data preprocessing stage, Trimble RealWorks software was used for stitching, and statistical filtering was employed to remove noise points (with a filtering threshold set to 3). Finally, the point cloud density was uniformly sampled to 1 mm / point. After data quality checks, it was ensured that the point cloud coverage of key components (such as brackets and columns) reached over 95%, with no obvious holes or distortions, thus generating a high-quality documentary point cloud model of the Jade Buddha Hall. The collected and processed point cloud model of the Jade Buddha Hall is shown below. Figure 5 As shown.

[0041] S2: The documentary point cloud model undergoes forward deconstruction and point cloud segmentation to obtain component point cloud data corresponding to each independent unit. Then, using this component point cloud data as a dimensional reference, inverse parametric modeling is performed to obtain a standard component correction model. This step aims to address the problems of component coupling, semantic gaps, and lack of standard reference benchmarks encountered when directly using the original point cloud for detection. Forward deconstruction discretizes the complex overall building into independent component units, eliminating spatial occlusion and adhesion interference between components. Inverse parametric modeling uses algorithms to reconstruct the undamaged ideal geometric shape from the damaged existing data, thereby establishing a standard reference system that objectively reflects the original design dimensions of the components and removes distortion features such as cracks, deformations, or human-caused damage. This provides a high-precision benchmark model for subsequent shape difference quantification calculations. The construction process is as follows: Figure 3 As shown.

[0042] In one specific implementation of this application, step S2 includes: performing semantic rule-based component spatial semantic segmentation on the documentary point cloud model to obtain component point cloud data corresponding to each independent unit; performing robust geometric feature inverse fitting on the component point cloud data to obtain an ideal geometric parameter set; generating a three-dimensional entity based on the ideal geometric parameter set through a building information modeling engine, and assigning the three-dimensional entity a spatial coordinate system consistent with the component point cloud data to perform parametric entity model reconstruction to obtain a standard component correction model.

[0043] Specifically, the documentary point cloud model is first subjected to semantic rule-based component spatial semantic segmentation to obtain component point cloud data corresponding to each independent unit. The documentary point cloud model reflecting the current state of the building obtained in the previous steps (…) Using this as input, and based on the structural form of ancient architecture (such as the modular rules in the *Yingzao Fashi*) or architectural construction logic, the documentary point cloud model is forward deconstructed. This process utilizes spatial semantic segmentation technology, combined with region growing algorithms or bounding box-based slicing tools, to decompose the overall point cloud model in three-dimensional space into independent units such as columns, beams, brackets, walls, and lintels. A schematic diagram of the forward deconstruction is shown below. Figure 6 As shown. At the operational level, the point cloud documentary model can be exported to a common format (such as .rcs), imported into 3D processing software (such as 3ds Max), and then sliced ​​according to the physical boundaries of the components to separate the discrete data corresponding to each independent unit, i.e., the component point cloud data. During this process, it is necessary to strictly preserve the absolute spatial coordinate information of the components in the original model to ensure that the segmented component point cloud data consists of multiple independent but precisely positioned component point sets.

[0044] After acquiring the component point cloud data, robust geometric feature inverse fitting is required to extract the ideal geometric parameter set. To eliminate the influence of natural deformation, cracking, peeling, or protrusions on the component surface on the modeling accuracy, robust geometric fitting based on the least squares method is performed for each independent component point set. Taking the cylindrical wooden component commonly found in ancient architecture as an example, a spatial cylindrical fitting optimization model is constructed to find an optimal set of geometric parameters (including the center of the base). , axis direction vector and radius This minimizes the sum of squared distances from all valid points in the point cloud to the ideal cylindrical surface. This fitting process automatically filters out outliers caused by cracks, depressions, or bulges, thereby extracting the ideal set of geometric parameters characterizing the original design dimensions of the component. The formula for calculating the objective function of the cylinder fitting is as follows: in, To fit the coordinate vector of the center of the cylinder's base, which is used to determine the starting point of the component's spatial position; To fit the unit direction vector of the cylinder's axis, used to characterize the growth direction or tilting posture of the component; To fit the ideal radius of the cylinder, representing the design cross-sectional dimensions of the wooden component in its undeformed state; This represents the total number of sampling points involved in the fitting calculation, i.e., the number of points in the current component point cloud. The index variable for the sampling point; For the component point cloud data, the first Three-dimensional spatial coordinate vectors of points; Let be the Euclidean norm of the vector; This is the cross product operation for vectors.

[0045] Based on the extracted ideal geometric parameter set, a parametric solid model is reconstructed using a Building Information Modeling (BIM) engine. This process uses component point cloud data as a dimensional reference and utilizes BIM software (such as Revit) or 3D modeling software (such as 3ds Max, AutoCAD) to construct a standard component BIM family library. The workflow is as follows: Figure 7 As shown. The specific operations include: in 3ds Max software, outlining the ideal contour lines of the components based on point cloud data, editing the polygonal construction surface model, and converting it to a CAD wireframe model (.dwg format); then importing the wireframe model into Revit software, and creating a parametric component family by using annotation constraints, parameter settings, and Boolean operations such as extrusion, lofting, and hollow shape shearing to remove deformation and damage features. Finally, establishing a baseline grid and elevation in Revit software, calling the generated standard component family library, assembling and fine-tuning the parameters of each 3D entity according to the spatial position of the point cloud data, assigning the 3D entity a spatial coordinate system consistent with the component point cloud data, thereby generating a standard component correction model composed of multiple standard components that restores the ideal geometric shape of the original architectural design. ).

[0046] In a specific embodiment, taking the inspection of a Jade Buddha Hall as an example, the implementers first exported the pre-processed documentary point cloud model of the Jade Buddha Hall as a .rcs format and imported it into 3ds Max software. Based on the architectural form of the Jade Buddha Hall, it was deconstructed into units such as columns, beams, brackets, and walls, and the point cloud areas corresponding to each component were cut out using the slicing tool. Then, using the cut-out column component point cloud as a reference, its ideal circular outline was sketched in 3ds Max, a surface model was constructed, and exported as a .dwg format; after generating a wireframe model in AutoCAD, it was imported into Revit, and a parametric cylindrical solid family was created using the "Extrude" command. For complex components such as brackets, polygon editing and lofting modeling were also performed based on the point cloud outline. Finally, a baseline grid for the Jade Buddha Hall was established in Revit. Referring to the coordinate positions of the point cloud data, starting from the base and columns, beams, brackets, and walls were gradually assembled, ultimately generating a standard component correction model that removed existing damaged features and restored the original design form of the Jade Buddha Hall. The Jade Buddha Hall correction model assembled in Revit is shown below. Figure 8 As shown. Furthermore, to provide a more realistic model foundation, models such as... Figure 9 The texture model shown is created by processing collected photos of the surface and then attaching them to a modified model to reflect the true appearance of the ancient building.

[0047] S3: Random surface sampling and Euclidean distance ratio calculation are performed on the standard component correction model and component point cloud data to obtain the reference shape distribution curve and the current shape distribution curve. This step aims to solve the problems of registration difficulties and high computational complexity encountered when directly comparing point clouds and models in 3D space. By introducing a shape distribution function that is insensitive to noise and can effectively distinguish shapes, a digital fingerprint that can characterize the initial and current forms of the component is established, providing a unified dimensional comparison benchmark for subsequent quantitative inspection.

[0048] In one specific implementation of this application, step S3 includes: performing surface discretization and unified sampling on the standard component correction model and component point cloud data to obtain an ideal sample point set and a current sample point set; performing D2 global distance distribution calculation on the ideal sample point set and the current sample point set to obtain an ideal distance set and a current distance set; and performing maximum value normalization and histogram statistics on the ideal distance set and the current distance set respectively to obtain a reference shape distribution curve and a current shape distribution curve.

[0049] Specifically, firstly, to compare the continuous solid model and discrete point cloud data on the same dimension, surface discretization and unified sampling need to be performed on the standard component correction model and the component point cloud data to obtain the ideal sample point set and the current sample point set. Considering both curve stability and computational complexity, for the standard component correction model (usually a mesh model), the area weight of each triangular facet is calculated, and points are uniformly and randomly sampled from the model surface. Points (selected in this embodiment) ), constituting an ideal sample point set ( For component point cloud data, a random resampling strategy is used to extract or supplement the data to the same number of components. These points constitute the current status sample point set. This process ensures that different types of components (such as brackets, columns, walls, etc.) are properly secured. Figure 12 The shape distribution curves (as shown) all exhibit good stability and eliminate calculation bias caused by differences in data density. Figure 11 The curves showing the shape distribution of brackets and columns at different numbers of sampling points are presented. Figure 13 The stability of the curves of the four models under different numbers of random points was demonstrated, and the rationality of using 100,000 points as a uniform number of sampling points was verified.

[0050] Subsequently, the D2 global distance distribution is calculated for the ideal sample point set and the current sample point set to obtain the ideal distance set and the current distance set. The D2 distribution (distance distribution between two points) function is chosen for this process because it can effectively distinguish between three-dimensional objects of different shapes and is insensitive to noise and small deformations, making it particularly suitable for describing the shape of ancient architectural components. Figure 10A schematic diagram of five different shape distribution functions is shown. (For an ideal sample point set...) Any two points in ) and Calculate its Euclidean distance By traversing or using Monte Carlo methods to extract a large number of point pairs for calculation, an ideal distance set is generated. Similarly, for the current sample point set ( Perform the same operation to generate a current distance set ( The formula for calculating the D2 distance is as follows: in, For the generated distance set, For the set of The point and the first Euclidean distance between points The operator for calculating the L2 norm, i.e., the Euclidean distance. For the input sample point set Any two points in three-dimensional space.

[0051] Finally, to eliminate the influence of component dimensions on shape description and unify the comparison domain, maximum value normalization and histogram statistics were performed on the ideal distance set and the current distance set respectively to obtain the reference shape distribution curve and the current shape distribution curve. First, maximum value normalization was performed on the input distance set, mapping the distance value range to... Intervals. Next, using histogram statistics, the normalized distance range is divided into intervals. For each interval, the number of distance data points falling within each interval is counted, and their proportion in the total number is calculated to generate a probability density function (PDF). This process is repeated for both input sets to obtain a reference shape distribution curve with the ratio of distance to maximum distance as the x-axis and the probability of that ratio occurring as the y-axis. ) and the current shape distribution curve ( ). Figure 14 The shape distribution curves of four types of components at 100,000 random points are shown. The normalization and probability calculation formulas are as follows: in, The distance value is the normalized value; The maximum distance value in the current distance set; The distribution curve at the th The ordinate values ​​of each interval; To fall into the first The count value of distance data within each interval; This represents the total number of distance data samples used in the statistics.

[0052] In one specific embodiment, taking the inspection of a jade Buddha hall as an example, the implementers used Matlab software as the analysis and calculation tool. First, components in the jade Buddha hall (such as columns and brackets) were selected as the inspection objects, and separate structures were built... Time-based correction model and Point cloud model data at each moment. For each component, 100,000 random sample points were collected on the model surface. Then, the Euclidean distance between any two points was calculated, and two D2 shape distribution curves were plotted with distance / maximum distance as the x-axis and occurrence probability as the y-axis. These two curves compress the complex three-dimensional morphology of the component into one-dimensional probability distribution curves, providing direct data input for subsequent quantitative assessment of crack and deformation degree by calculating curve similarity (MAE' and S-value).

[0053] S4: Calculate the shape similarity index between the reference shape distribution curve and the current shape distribution curve. The main purpose of this step is to solve the technical problem of accurately quantifying the degree of defects by directly observing the curve shape. By using mathematical methods, the complex probability distribution differences are compressed into a standardized value, thereby providing objective and unified data support for subsequent threshold-based structural stability assessment and defect classification. This achieves a leap from qualitative image analysis to quantitative data decision-making, enabling the detection results to intuitively reflect the integrity of ancient building components relative to their ideal design state. To adapt to different precision detection requirements, this invention provides two specific implementation methods: one is a linear calculation method based on mean absolute error (MAE), and the other is a nonlinear shape relaxation measurement method based on Wasserstein distance.

[0054] In the first embodiment, a strategy combining discrete interval absolute error calculation with linear amplification is mainly adopted. In a specific implementation of this application, step S4 includes: calculating the discrete interval absolute error of the reference shape distribution curve and the current shape distribution curve to obtain the original average absolute error; introducing a preset linear amplification factor to perform error numerical sensitivity amplification correction on the original average absolute error to obtain the corrected average absolute error; and performing similarity normalization inversion mapping on the corrected average absolute error to obtain the shape similarity index.

[0055] First, the input reference shape distribution curve and the current shape distribution curve are treated as vector data with the same resolution (i.e., the same number of discrete intervals). Discrete interval absolute error calculation is then performed to obtain the raw mean absolute error. This process iterates through each corresponding interval, calculating the absolute value of the difference between the probability density value of the reference shape distribution curve in that interval and the corresponding probability density value of the current shape distribution curve. The absolute differences of all intervals are summed and divided by the total number of intervals to obtain the raw mean absolute error reflecting the average deviation between the two curves. The calculation formula is as follows: in, and These represent the probability density values ​​of the reference shape distribution curve and the current shape distribution curve in the i-th interval, respectively. The total number of discrete intervals. This represents the original mean absolute error.

[0056] Considering that the probability difference reflected on the D2 distribution curve for minute cracks or slight deformations in ancient wooden structures is usually extremely small (on the order of magnitude), to Directly using the original error values ​​would make subsequent evaluations unintuitive and difficult to classify levels; therefore, error sensitivity amplification correction is necessary. This process introduces a preset linear amplification factor (e.g., 1000) to numerically amplify the original mean absolute error calculated above, generating a corrected mean absolute error. The correction formula is as follows: in, Indicates the linear amplification factor. This indicates the corrected mean absolute error.

[0057] Finally, a similarity normalization inversion mapping is performed on the corrected mean absolute error to obtain the shape similarity index. This process uses linear inversion logic, subtracting the corrected error value from 1, and adds boundary constraint logic. If the error is extremely large, causing the calculated result to be less than 0, it is forcibly truncated to 0, ensuring that the final output shape similarity index always falls within the valid non-negative interval of 0 to 1. The calculation formula is as follows: in, This indicates the shape similarity index.

[0058] In the first implementation of the algorithm for detecting the apparent quality of ancient wooden structures, although it can obtain basic shape distribution curves, it has significant theoretical and practical shortcomings in the crucial similarity assessment stage. Specifically, the traditional linear error calculation method mainly relies on simple discrete interval average absolute error to calculate the difference between two curves. This approach firstly demonstrates a lack of consideration for neighborhood topological relationships.

[0059] At the physical level, the overall bending deformation of wood caused by stress often leads to a general shift or skewness of the probability distribution curve on the coordinate axis, while local cracks or spalling are more likely to manifest as in-situ oscillations and fluctuations within a specific probability density range. The original MAE algorithm adopts point-to-point independent difference logic, mechanically treating the decrease in value of interval i and the increase in value of neighboring interval j as two unrelated independent events. This makes it impossible for the algorithm to mathematically identify the physical fact that the probabilistic mass simply flows from i to i+1 (i.e., a slight bending occurs rather than material loss), thus failing to effectively distinguish between structural drift and random noise, and severely weakening its ability to characterize deformation-type defects.

[0060] Secondly, the original method simply introduces a linear factor λ to amplify the error, which is essentially a hard coding lacking physical field meaning and mathematical robustness. The safety risks of ancient building structures often exhibit a non-linear exponential increase with the accumulation of deformation, rather than a simple linear superposition. The first embodiment adopts... The linear truncation logic is prone to mathematical anomalies such as similarity index S<0 when dealing with boundary values ​​with large errors, and it is completely unable to simulate the Gaussian decay characteristic that is common in structural reliability theory. As a result, the final output detection score is difficult to truly reflect the safety margin of the component.

[0061] To overcome the aforementioned technical bottlenecks, this solution constructs a nonlinear shape relaxation metric mechanism based on Wasserstein distance. By introducing the concept of bulldozer distance and Gaussian kernel nonlinear mapping, it achieves accurate quantitative evaluation of the apparent quality of ancient wooden structures.

[0062] That is, in another specific implementation of this application, step S4 includes: calculating the structural drift degree between the reference shape distribution curve and the current shape distribution curve; calculating the local abrupt change degree between the reference shape distribution curve and the current shape distribution curve; and performing nonlinear similarity mapping on the structural drift degree and the local abrupt change degree to obtain the shape similarity index.

[0063] First, to address the issue that single-pixel alignment cannot capture macroscopic deformation, structural drift measurement is implemented, requiring the calculation of the structural drift degree between the reference shape distribution curve and the current shape distribution curve. This step is no longer limited to local comparison of probability density functions, but introduces the concept of bulldozer distance from optimal transport theory, transforming it into a difference measure of the cumulative distribution function in a one-dimensional distribution scenario. Specifically, the cumulative distribution functions of the ideal model and the current measured data are first calculated separately, followed by the calculation of the absolute difference between their integral areas. It should be understood that EMD can measure the minimum physical work required to reshape the current geometric distribution back to the original design distribution, thus mathematically aligning the physical bending and shear deformation of the wooden components caused by stress, achieving effective capture of structural drift from a macroscopic perspective. The formula for calculating the structural drift degree is as follows: in, and These represent the probability density values ​​of the reference shape distribution curve and the current shape distribution curve in the i-th interval, respectively. and These are the cumulative distribution function values ​​of the corresponding curve in the b-th interval; The total number of discrete intervals; This represents the output structural drift.

[0064] Secondly, to address the challenge that minor local defects such as cracks and insect damage are often masked by background noise, a feature-weighted local abrupt change calculation is performed. This involves calculating the local abrupt change between the reference shape distribution curve and the current shape distribution curve. Considering that the main geometric features of ancient architectural components (such as the main diameter of a cylinder) inevitably correspond to high-probability density peak regions in the distribution curve, while random scanning noise is usually scattered in low-probability tail regions, this step constructs an adaptive weight mask based on the saliency of the original features. By assigning higher computational weights to the main feature regions, the signal-to-noise ratio can be effectively improved, and the high-frequency oscillation signals caused by cracks and spalling can be amplified in a targeted manner. This processing method can intelligently suppress noise interference in non-critical areas, thereby accurately quantifying local texture abrupt changes at the microscopic level, ensuring that subtle defects are not missed. The formula for calculating the local abrupt change can be expressed as: in, Indicates the first Adaptive feature weights for each interval; This is the attention gain factor, used to adjust the sensitivity to the main structural region. For cylindrical components with simple and significant geometric features, it is recommended to set it to [value missing]. To appropriately increase the weight of the main peak area; and for dougong-type components with complex geometric features, it is recommended to take... This significantly suppresses background noise in non-critical areas and highlights structural damage signals. This represents the maximum value of the probability density function of the ideal distribution curve. This is the local mutation degree of the output.

[0065] Finally, to address the issue that linear scoring cannot reflect the nonlinear decay law of structural safety, a nonlinear similarity mapping based on radial basis functions (RBF) is performed. Specifically, the structural drift and local mutation degrees are nonlinearly mapped to obtain a shape similarity index. This step abandons the traditional linear subtraction logic. First, a weighted fusion method is used to integrate the macroscopic structural drift and microscopic local mutation degrees into a total energy loss. Then, a Gaussian kernel function is introduced as the activation function for mapping. This approach aims to endow the scoring system with a smooth-steep-smooth nonlinear response characteristic: maintaining high similarity (robustness) within a healthy range with minimal error, and a sharp drop in score once the error exceeds a critical point (high sensitivity), accurately simulating the abrupt process of structural failure from safety. This not only avoids the negative value anomalies that may occur in linear calculations but also ensures that the final output similarity index conforms to the physical laws of structural mechanics, providing a scientific and reliable basis for maintenance decisions. The formula for calculating the shape similarity index is: in, This represents the total energy loss after fusion; The drift-mutation fusion weighting coefficient can be adjusted according to the component type (such as column or beam). For example, for major load-bearing components such as beams and columns where overall bending deformation is the primary concern, it is recommended to use [a specific weighting coefficient]. For decorative components such as brackets where surface texture and integrity are of primary concern, it is recommended to take... ; This is the sensitivity bandwidth parameter, which controls the rate of similarity decay. For high-precision detection scenarios, it is recommended to set it to [value missing]. ; This is the shape similarity index for the final output.

[0066] This specific implementation aims to solve core problems in the surface quality inspection of ancient wooden structures, such as unclear defect identification and distorted scoring mechanisms. By introducing the Wasserstein distance concept to construct structural drift, it effectively decouples the overall deformation of components from local noise, significantly improving the accuracy of identifying macroscopic deformations such as bending and shearing. By using feature weighting to calculate local abrupt changes, it enhances the algorithm's ability to capture minute texture defects such as cracks and spalling, effectively suppressing environmental noise interference. Finally, it uses Gaussian kernel nonlinear mapping to generate similarity indices, establishing a scoring system that conforms to the decay law of the structural safety index, completely eliminating mathematical boundary anomalies in linear scoring.

[0067] In one specific embodiment, taking the inspection of a jade Buddha hall as an example, for the first implementation method, the implementers used Matlab software as the analysis and calculation tool. Components within the jade Buddha hall were selected as the inspection objects, and data was obtained... Time-correction model and After obtaining the D2 shape distribution curve of the point cloud model, the original mean absolute error is calculated. Then, this error value is multiplied by 1000 to obtain... This eliminates the problem of insignificant differences caused by excessively small values. Ultimately, through the formula... The similarity index of the component was calculated. If the calculated A value of 0.95 indicates that the component is highly similar to the original design and is in good condition. Regarding the second implementation method, if a wooden beam undergoes overall bending due to long-term stress, causing a shift in the distribution curve, the ordinary MAE algorithm may misjudge due to point-to-point differencing. However, the second implementation method can accurately capture this physical bending by calculating the structural drift degree. Simultaneously, the Gaussian kernel function mapping is used to obtain... The value can more sensitively reflect the nonlinear effect of bending on structural safety, avoiding the negative value anomalies that may occur with simple linear subtraction, and providing a more scientific basis for maintenance decisions. To verify the feasibility of defect detection, this application also constructed a system as follows: Figure 15 The 12 test models shown include brackets, tables, columns, and walls with varying degrees of detachment.

[0068] S5: The shape similarity index is compared with a preset structural stability threshold, and the defect type is identified by combining the morphological deviation characteristics of the current shape distribution curve relative to the reference shape distribution curve, so as to obtain the appearance quality inspection results including defect level and maintenance strategy. This step aims to solve the problem that simple numerical indicators cannot specifically guide the repair work. By establishing a mapping mechanism of numerical value-state-strategy, a closed-loop technology for the entire process from defect identification to quantitative assessment and maintenance guidance is realized, providing a standardized inspection solution for the protection of ancient buildings.

[0069] In one specific implementation of this application, step S5 includes: based on a preset structural stability critical threshold, performing threshold determination and state classification on the shape similarity index to obtain the structural stability state; when the structural stability state is abnormal, performing morphological difference feature pattern recognition on the reference shape distribution curve and the current shape distribution curve to determine the semantic label of the defect type; and querying a preset maintenance knowledge base by combining the structural stability state, the semantic label of the defect type and the shape similarity index to obtain the appearance quality detection result.

[0070] Specifically, firstly, a threshold determination and state classification based on a preset structural stability critical threshold are performed to obtain the structural stability state. According to the "Technical Standard for Maintenance and Reinforcement of Ancient Wooden Structures GB / T50165-2020" (see Table 1), when the deformation of wooden components does not exceed 2mm, it does not affect structural stability; the corresponding average similarity is approximately 0.9. Therefore, a structural stability critical threshold is set (…). The input shape similarity index is set to 0.9. Compare the values ​​with the threshold: If It not only determines that the components are in a healthy state, but also marks the differences as natural aging within an acceptable range, and outputs the structural stability status. ) is stable; if If the condition is not met, the component is determined to have a substantial defect, requiring further classification analysis at the next level, and the output structural stability state is deemed abnormal. The determination logic formula is as follows: Subsequently, when the structural stability state is abnormal, morphological difference feature pattern recognition is performed on the reference shape distribution curve and the current shape distribution curve to determine the semantic label of the defect type. This process identifies the specific physical defect type based on the geometric feature changes of the curve: if the curve difference is mainly manifested as a slight shift or oscillation in a local area (i.e., local curve shift), it is determined to be a crack defect. If the difference manifests as a shift in the overall shape or a misalignment of the peak (i.e., a change in the overall curve shape), it is judged as a deformation defect. If the current shape distribution curve shows a broken interval with a probability value of 0 or a significant missing integral area (i.e., a missing segment of the curve), it is determined to be a component detachment defect. The final output is the corresponding defect type semantic label. If the state is stable, the label is empty or none.

[0071] Finally, the appearance quality inspection results are obtained by querying the pre-set maintenance knowledge base based on the overall structural stability status, defect type semantic tags, and shape similarity index. According to the judgment rules, if... And the status is stable, the matching maintenance and update strategy is that maintenance is not required for the time being or routine inspection; if If the status is abnormal, specific repair suggestions are generated based on the semantic tags of the specific defect type (such as cracks, deformation, or detachment), and the maintenance and update strategy is marked as requiring maintenance or requiring immediate maintenance. Finally, the above strategies and quantitative indicators are... The defect types are encapsulated into structured data to generate an appearance quality inspection result report that includes defect levels and maintenance strategies.

[0072] In one specific embodiment, taking the inspection of a jade Buddha hall as an example, the implementers calculated the shape similarity index of a certain component. Then, it is compared with the threshold of 0.9. If a certain column component... If the value is 0.85 (less than 0.9), and the current shape distribution curve shows a significant local oscillation offset relative to the reference shape distribution curve, the system determines that the component has a crack defect and outputs a strategy suggestion for maintenance or immediate maintenance, prompting maintenance personnel to repair the crack in the column. Conversely, if a beam component has a value of 0.85 (less than 0.9), and the current shape distribution curve shows a significant local oscillation offset relative to the reference shape distribution curve, the system determines that the component has a crack defect and outputs a strategy suggestion for maintenance or immediate maintenance, prompting maintenance personnel to repair the crack in the column. A value of 0.95 indicates that maintenance is not required at this time. This process verifies the effectiveness and operability of the detection method in practical engineering.

[0073] In addition, in another embodiment of this application, in order to address the problem that it is often difficult to obtain measured three-dimensional model data of components at historical moments in ancient building inspection scenarios, which makes it impossible to directly detect defect differences through multi-temporal comparison, this application also includes the step of constructing a component evolution BIM model and performing multi-temporal continuous inspection.

[0074] Specifically, this application proposes using a reconstructed ideal geometric model as an initial benchmark, and dynamically updating the benchmark model as the inspection cycle progresses, thereby establishing a digital twin archive of ancient buildings over time. The method for inspecting the apparent quality of wooden structures further includes steps S6 and S7 after step S5: S6: Update the geometric state of the baseline model based on the current detection results to generate a component evolution BIM model.

[0075] The specific steps include: First, using the standard component modification model obtained in step S2 as the initial component BIM baseline model. This model represents the geometric information of the component in its ideal or initial state.

[0076] Secondly, based on the component point cloud data acquired at the current detection time, the current component BIM model reflecting the current component's geometric state is generated according to the reverse parametric modeling method in step S2 (i.e., geometric feature fitting and parametric reconstruction of the point cloud).

[0077] Subsequently, in the Building Information Modeling (BIM) modeling software, the component BIM baseline model file is loaded; in the 3D view, the corresponding target component in the component BIM baseline model is located and deleted; the corresponding component in the current component BIM model is copied and pasted into the component BIM baseline model file. During the pasting process, in-situ positioning and replacement are strictly performed using the same elevation and coordinate method. Through the above operations, the component BIM baseline model is updated to a component evolution BIM model reflecting the geometric defect state of the component at the current inspection time.

[0078] S7: Perform multi-temporal continuous detection and dynamic updates based on the component evolution BIM model.

[0079] In the next testing cycle, the component evolution BIM model obtained in the previous cycle will be used as the new component BIM benchmark model.

[0080] The point cloud data of the building's timber structure is reacquired using a 3D laser scanning device, and steps S1 to S5 are repeated. It is important to note that when repeating step S3 (constructing the shape distribution curve), the new component BIM baseline model replaces the standard component correction model in the calculation. That is, the newly acquired component point cloud data is compared and analyzed with the baseline model updated at the previous moment to capture any new deformations or damage that have occurred since the last update.

[0081] When the detection results show that the geometric changes of the component at the new detection time exceed the allowable range (i.e., the shape similarity index is lower than the preset evolution update threshold), step S6 is executed again to update the geometric state of the new component BIM baseline model to reflect the latest state; when the detection results show that the geometric changes of the component at the new detection time are within the allowable range, the current component BIM baseline model is retained unchanged.

[0082] Through the aforementioned iterative execution method, the component BIM model can gradually evolve as the actual geometric state of the component changes, realizing multi-temporal continuous detection of the appearance quality of the building timber structure and dynamic updating of the BIM model.

[0083] In summary, the surface quality inspection method for wooden structures provided in this application has been elucidated. It utilizes three-dimensional laser scanning technology to achieve non-contact inspection throughout the entire process, completely eliminating the risk of "secondary damage" to ancient artifacts that could be caused by manual hammering. More importantly, by introducing the shape distribution curve of the Euclidean distance ratio as a feature descriptor, this application successfully avoids the excessive reliance of traditional detection techniques on the spatial registration accuracy of point cloud models, effectively solving the registration misalignment problem caused by the complex surface texture and non-uniform local deformation of ancient wooden components. This method of transforming geometric differences into probabilistic statistical differences not only greatly improves the detection algorithm's resistance to interference and robustness against changes in lighting, surface stains, and natural wood textures (such as knots and grain), but also allows for the sensitive capture of millimeter-level micro-cracks and deformation signals through quantified shape similarity indicators. This provides objective, accurate, and traceable data support for the preventative protection and scientific restoration of ancient wooden structures.

[0084] This application also provides a device for testing the appearance quality of building timber structures, used to perform the above-described method for testing the appearance quality of building timber structures, such as... Figure 16As shown, the architectural timber structure appearance quality inspection device 200 includes: a point cloud acquisition module 210, used to acquire the original point cloud data of the ancient architectural timber structure site through a three-dimensional laser scanning device, and preprocess the original point cloud data to obtain a documentary point cloud model reflecting the current state of the building; a component deconstruction module 220, used to perform forward deconstruction and point cloud cutting on the documentary point cloud model to obtain the component point cloud data corresponding to each independent unit, and to perform reverse parametric modeling using the component point cloud data as a size reference to obtain a standard component correction model; a curve generation module 230, used to perform random surface sampling and Euclidean distance ratio calculation on the standard component correction model and the component point cloud data to obtain a reference shape distribution curve and a current shape distribution curve; a similarity calculation module 240, used to calculate the shape similarity index between the reference shape distribution curve and the current shape distribution curve; and a defect identification module 250, used to compare the shape similarity index with a preset structural stability threshold, and combine the morphological offset characteristics of the current shape distribution curve relative to the reference shape distribution curve to identify the defect type, so as to obtain an appearance quality inspection result including defect level and maintenance strategy.

[0085] It should be noted that the surface quality testing device for building timber structures in this application is similar in principle to the aforementioned surface quality testing method for building timber structures. Therefore, the implementation process, implementation principle, and beneficial effects of the surface quality testing device for building timber structures can be found in the description of the implementation process, implementation principle, and beneficial effects of the aforementioned method, and will not be repeated.

[0086] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this application can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this application can be achieved, and this is not limited herein.

[0087] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "a plurality of" means two or more, unless otherwise explicitly specified.

[0088] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for inspecting the appearance quality of a building timber structure, characterized in that, include: Step S1: Obtain the original point cloud data of the ancient wooden structure site through a three-dimensional laser scanning device, and preprocess the original point cloud data to obtain a documentary point cloud model that reflects the current state of the building. Step S2: Perform forward deconstruction and point cloud cutting on the documentary point cloud model to obtain the component point cloud data corresponding to each independent unit, and perform reverse parametric modeling using the component point cloud data as a size reference to obtain the standard component correction model. Step S3: Random surface sampling and Euclidean distance ratio calculation are performed on the standard component correction model and component point cloud data to obtain the reference shape distribution curve and the current shape distribution curve; Step S4: Calculate the shape similarity index between the reference shape distribution curve and the current shape distribution curve; Step S5: Compare the shape similarity index with the preset structural stability threshold, and combine the morphological deviation characteristics of the current shape distribution curve relative to the reference shape distribution curve to identify the defect type, so as to obtain the appearance quality inspection result including the defect level and maintenance strategy.

2. The method for inspecting the appearance quality of wooden structures according to claim 1, characterized in that, Step S1 includes: A three-dimensional laser scanner is used to collect raw data from multiple perspectives of the ancient building entity to be inspected, so as to obtain a set of raw measurement stations containing spatial three-dimensional coordinate information and reflection intensity information. Multi-site cloud registration and stitching are performed on the original set of station cloud data to obtain a spatially continuous and complete stitched point cloud; Statistical filtering and noise reduction are performed on the complete stitched point cloud to obtain clean point cloud data; A documentary point cloud model is obtained by uniformly sampling pure point cloud data using a voxel grid.

3. The method for inspecting the appearance quality of wooden structures according to claim 1, characterized in that, Step S2 includes: The documentary point cloud model is subjected to component spatial semantic segmentation based on semantic rules to obtain the component point cloud data corresponding to each independent unit; Robust geometric feature inverse fitting is performed on the component point cloud data to obtain the ideal geometric parameter set; A 3D entity is generated based on an ideal set of geometric parameters using a building information modeling engine. The 3D entity is then given a spatial coordinate system consistent with the component point cloud data to reconstruct a parametric entity model, thereby obtaining a standard component correction model.

4. The method for inspecting the appearance quality of wooden structures according to claim 1, characterized in that, Step S3 includes: Surface discretization and unified sampling are performed on the standard component correction model and component point cloud data to obtain the ideal sample point set and the current sample point set; D2 global distance distribution calculation is performed on the ideal sample point set and the current sample point set to obtain the ideal distance set and the current distance set; Maximum value normalization and histogram statistics are performed on the ideal distance set and the current distance set respectively to obtain the reference shape distribution curve and the current shape distribution curve.

5. The method for inspecting the appearance quality of building timber structures according to claim 1, characterized in that, Step S4 includes: The absolute error of the reference shape distribution curve and the current shape distribution curve is calculated by discrete interval calculation to obtain the original mean absolute error. A preset linear amplification factor is introduced to perform error numerical sensitivity amplification correction on the original mean absolute error to obtain the corrected mean absolute error; The shape similarity index is obtained by performing a similarity normalization inversion mapping on the corrected mean absolute error.

6. The method for inspecting the appearance quality of building timber structures according to claim 1, characterized in that, Step S4 includes: Calculate the structural drift between the reference shape distribution curve and the current shape distribution curve; Calculate the local abruptness between the reference shape distribution curve and the current shape distribution curve; Nonlinear similarity mapping is performed on structural drift degree and local mutation degree to obtain shape similarity index.

7. The method for inspecting the appearance quality of building timber structures according to claim 1, characterized in that, Step S5 includes: Based on a preset critical threshold for structural stability, the shape similarity index is used to determine the threshold and classify the state to obtain the structural stability state. When the structural stability state is abnormal, morphological difference feature pattern recognition is performed on the reference shape distribution curve and the current shape distribution curve to determine the semantic label of the defect type. The appearance quality inspection results are obtained by querying the preset maintenance knowledge base based on the comprehensive structural stability status, defect type semantic tags, and shape similarity index.

8. The method for inspecting the appearance quality of building timber structures according to claim 1, characterized in that, It also includes step S6: Use the standard component modification model obtained in step S2 as the initial component BIM benchmark model; Based on the component point cloud data acquired at the current detection time, the current component BIM model reflecting the current component's geometric state is generated according to the reverse parametric modeling method in step S2. In the 3D view of the modeling software, delete the target component in the component BIM baseline model, and copy the corresponding component in the current component BIM model to the component BIM baseline model with the same elevation and coordinates, thereby updating the component BIM baseline model to the component evolution BIM model.

9. The method for inspecting the appearance quality of building timber structures according to claim 8, characterized in that, It also includes step S7: In the next inspection cycle, the component evolution BIM model obtained in the previous cycle will be used as the new component BIM benchmark model. Reacquire the site point cloud data of the building timber structure and repeat steps S1 to S5. When repeating step S3, use the new component BIM benchmark model instead of the standard component correction model for calculation. When the detection results indicate that the geometric changes of the component at the new detection time exceed the allowable range, step S6 is executed again to update the new component BIM baseline model; When the inspection results indicate that the geometric changes of the component at the new inspection time are within the allowable range, the current component BIM baseline model is retained.

Images