Traditional wood carving innovation method and system based on 3D scanning

Abstract

The invention relates to the technical field of carving, in particular to a traditional wood carving innovation method and system based on 3D scanning, and the method comprises the steps: carrying out the high-precision scanning of a core region through a 3D scanner, carrying out the lightweight scanning of a secondary region, obtaining multi-view point cloud data, and carrying out the preprocessing; inputting the preprocessed multi-view point cloud data into the point cloud network model, outputting the filled point cloud, and manually carrying out refinement on the filled point cloud to obtain a repaired point cloud; geometric features are extracted, pattern semantics are marked, and a dynamic element library is constructed and stored; the parameterized design and modeling processing platform is in butt joint with the dynamic element library, generates and screens a design scheme, manually corrects the design scheme, verifies the corrected design scheme, and performs finite element analysis on the verified design scheme to obtain a final scheme; and matching the corrected final scheme process parameters through an SWRL rule. According to the scheme, the digital precision and efficiency of the woodcarving are improved through multi-resolution scanning.

Description

Technical Field

[0001] This invention relates to the field of carving technology, and in particular to an innovative method and system for traditional wood carving based on 3D scanning. Background Technology

[0002] In the field of digital preservation and innovative design of traditional wood carvings, there has always been a challenge in balancing data accuracy and processing efficiency, as well as the risk of losing authenticity during restoration. Traditional digital acquisition methods often employ single-resolution scanning, which either results in massive data volumes and low processing efficiency due to the pursuit of high precision, making it unsuitable for the actual needs of small and medium-sized workshops, or sacrifices accuracy for efficiency, failing to meet the high standards of museum-level artifact restoration, especially when dealing with core and secondary artistic areas. Furthermore, in the point cloud restoration stage, relying solely on manual restoration is not only time-consuming and labor-intensive but also struggles to ensure the complete preservation of key handcrafted features; while relying solely on AI-automated restoration may lead to the loss of traditional craftsmanship, failing to balance accuracy and authenticity during digital conversion. In addition, existing technologies often suffer from data loss or inaccurate reconstruction when dealing with complex structures such as deep hollows and tool marks, severely impacting the quality of the foundational data for subsequent innovative designs. Summary of the Invention

[0003] This invention constructs a dynamic element library by using multi-resolution scanning, AI, and manual point cloud restoration, and combines parametric design and finite element analysis to improve the accuracy and efficiency of wood carving digitization.

[0004] The technical solution proposed in this invention is: an innovative method for traditional wood carving based on 3D scanning, the method comprising: The wood carvings to be scanned are pre-processed. The core area is scanned with high precision using a 3D scanner, while the secondary areas are scanned with lightweight technology to obtain multi-view point cloud data. The multi-view point cloud data is then pre-processed. The preprocessed multi-view point cloud data is input into the point cloud network model, and the filled point cloud is output. The filled point cloud is then manually refined to obtain the repaired point cloud. Based on the point cloud generated after Poisson surface reconstruction and repair, a triangular network is generated, geometric features are extracted, the semantics of the patterns are labeled, and a dynamic element library is constructed and stored by combining manually uploaded knowledge. The parametric design and modeling processing platform connects to a dynamic element library to generate and filter design schemes. The design schemes are manually corrected, and the corrected design schemes are verified. Finite element analysis is performed on the verified design schemes to obtain the final scheme, which is then manually corrected. The final process parameters of the scheme are corrected by matching SWRL rules, and the SWRL rule weights are updated using a reinforcement learning algorithm.

[0005] Preferably, the specific preprocessing procedure is as follows: Before scanning, use a soft brush and compressed air to clean the dust off the surface of the wood carving, and attach optical positioning markers to non-critical areas. Perform high-precision scanning on the core areas that carry the main cultural meaning and craftsmanship value, and perform lightweight scanning on secondary areas with weaker decorative features and lower craftsmanship complexity. Automatic stitching of multi-view point clouds is completed by matching positioning points with curvature features. Statistical outlier filtering and radius filtering are used to remove isolated points and noise. The point cloud of the core area is resampled, and voxel grid filtering is used to unify the indirectness of the secondary areas.

[0006] Preferably, the specific process for obtaining the filled point cloud is as follows: A point cloud network model is constructed using the PointFlow++ generative network, which includes an encoder and a decoder. The encoder extracts point cloud features based on the preprocessed multi-view point cloud data and the network model. The point cloud features are mapped to a low-dimensional latent space, and the missing region point cloud is generated based on the latent space features. The missing region point cloud is then stitched together with the preprocessed multi-view point cloud data to obtain the filled point cloud.

[0007] Preferably, the specific process of obtaining the repaired point cloud is as follows: On a VR manual correction platform, craftsmen roughly refine the filled point cloud by referring to reference materials; the roughly refined point cloud is then imported into Rhino, and using SubD modeling tools and a digital screen, the carving depth is controlled by pressure sensitivity to refine key features; manual feature points are identified by local curvature abrupt changes and normal vector jumps, the number of feature points before and after the repair is compared, the normal vector direction is unified, and the repaired point cloud is obtained.

[0008] Preferably, the dynamic element library construction process is as follows: A triangular mesh is generated by reconstructing a Poisson surface. Based on the mesh curvature distribution and abrupt changes in normal vectors, the pattern boundaries are identified, and geometric features are obtained. Craftsmen semantically annotate the extracted patterns and upload relevant knowledge through a 3D interactive platform. The uploaded knowledge is semantically understood using a BERT model to extract key information. If the knowledge confidence is ≥0.8 and does not conflict with existing data, the parameters of the corresponding elements are updated. The relationships between elements are stored using the Neo4j graph database.

[0009] Preferably, the process of obtaining the final solution is as follows: A parametric design and modeling platform based on Rhino and Grasshopper is built to convert 3D patterns in a dynamic element library into editable NURBS models and associate them with process constraint parameters. Multiple design schemes are generated by adjusting element parameters. Custom battery packs are used to calculate the hollowing thickness, extract the minimum radius of curvature, and set thresholds to filter feasible schemes. SubD modeling tools and a digital screen are used to carve and correct key details by hand, and the system automatically synchronizes the parametric data from Grasshopper. The corrected schemes are then subjected to final process verification in Grasshopper. Real-time wood properties and tool parameters are input, and finite element analysis is used to simulate the carving process, predict the rationality of the tool path, and obtain the final scheme.

[0010] Preferably, the specific matching process of the SWRL rule is as follows: The Neo4j graph database is used to store triple relationships, including design entities, process entities, and production entities. SWRL rules are written and imported into the knowledge graph through the Protege ontology editor. Data is collected by sensors on the production side, processed by ETL, and then mapped to the knowledge graph. A reinforcement learning algorithm is used, with the process feasibility index as the reward function, to update the weights of the SWRL rules.

[0011] The present invention also provides a traditional wood carving innovation system based on 3D scanning, the system being used to execute the aforementioned traditional wood carving innovation method based on 3D scanning.

[0012] The present invention also provides a computer-readable storage medium storing a computer program that is executed by a processor to implement the aforementioned innovative method for traditional wood carving based on 3D scanning.

[0013] The beneficial effects of this invention are: 1. Using an industrial-grade 3D scanner, the wood carving artifacts are scanned in sections. The core artistic areas are output with high-precision point clouds at the 0.05mm level, while secondary areas use lightweight point clouds at the 0.2mm level, effectively balancing data accuracy and processing efficiency. The developed virtual-real collaborative restoration algorithm first uses AI to automatically fill in the point cloud blind spots, and then allows craftsmen to manually correct key handcrafted features in a VR environment. The generated 3D model has an error of ≤0.1mm and retains the charm of traditional craftsmanship. This solves the problems of accuracy attenuation and authenticity loss in data conversion, providing a 3D prototype for innovative design of traditional wood carvings that combines digital precision with the warmth of handcraftsmanship. It meets the high-precision requirements of museum-level artifact restoration while also adapting to the efficiency needs of small and medium-sized workshops, ensuring the accuracy and completeness of cultural element extraction from the source.

[0014] 2. By utilizing edge detection and semantic recognition algorithms to extract wood carving pattern features, a dynamic database containing over 2000 traditional elements and supporting real-time uploading of artisan knowledge was constructed, continuously enriching the cultural heritage. Based on a dual-track innovation plugin developed using Rhino and Grasshopper, the parametric track allows designers to generate basic designs within one hour and associate them with dynamic process feasibility tags. The manual correction track allows designers to adjust key details from hand-drawn sketches, with the system automatically synchronizing parametric data. This breaks through the bottlenecks of difficult cultural element extraction and low reconstruction efficiency in traditional design. Through a human-machine collaboration mechanism, it not only improves the speed of innovative solution generation but also avoids the dilution of traditional craft styles by parametric design, promoting young designers' understanding and inheritance of the "freehand" aesthetic concept of traditional wood carving.

[0015] 3. The developed digital twin system for the process can access real-time dynamic data of wood and combine it with finite element analysis to simulate the cutting process. It also provides a calibration channel based on craftsman experience, enabling the correction of algorithm-recommended toolpaths and process parameters. The constructed bidirectional knowledge evolution graph achieves intelligent matching of "design scheme - process parameters" through SWRL rules, allowing production-end data to flow back to the knowledge graph to drive algorithm self-learning. It can also bind CNC code with videos of experienced craftsmen's operations to form an integrated link, solving the problems of lack of quantitative records of key parameters and difficulty for young designers to quickly master them in traditional craft inheritance. Through the deep integration of digital twins and human experience, it realizes the dynamic evolution and intergenerational inheritance of process knowledge, while improving the feasibility of implementing innovative solutions, shortening the design-to-production cycle, and reducing costs. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating an innovative method for traditional wood carving based on 3D scanning, as described in this invention. Figure 2 This is an innovative flowchart of a traditional wood carving innovation method based on 3D scanning, according to the present invention. Detailed Implementation

[0017] The following description is intended to disclose the present invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art. The basic principles of the invention defined in the following description can be applied to other embodiments, modifications, improvements, equivalents, and other technical solutions that do not depart from the spirit and scope of the invention.

[0018] It is understood that the term "a" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of an element can be one, while in another embodiment, the number of the element can be multiple, and the term "a" should not be understood as a limitation on the number.

[0019] like Figure 1 and Figure 2 As shown, an industrial-grade structured light 3D scanner (such as ArtecEva or GOMATOS) was used with a high-precision rotating stage (rotation accuracy ±0.01°) to scan the wood carving, supporting multi-angle, high-resolution data acquisition. Core artistic areas carrying the main cultural semantics and craft value (including details of traditional patterns (such as the turning points of scroll patterns and dragon patterns), the facial features of figures / animal carvings, key nodes of limb dynamics, deep reliefs with obvious hand-carved marks, and openwork structures (depth ≥ 5mm)) were scanned at a 0.05mm (high-precision) resolution, with a point cloud density of ≥ 100 points / mm² (corresponding point spacing ≤ 0.1mm). Secondary areas with weaker decorative features and lower craftsmanship complexity (including large base panels, non-core decorative surfaces such as borders, shallow relief or flat carving areas (depth < 2mm), and non-critical structural connections (such as mortise and tenon joints in furniture components, which are not visible)) were scanned at a 0.2mm (lightweight) resolution, with a point cloud density of ≥ 25 points / mm² (corresponding point spacing ≤ 0.2mm).

[0020] Before scanning, the wood carving needs to be pre-processed. Use a soft brush and compressed air to remove dust from the surface to prevent impurities from affecting scanning accuracy. Affix 5mm diameter optical positioning markers (white base + black center) to non-critical areas of the carving for subsequent point cloud stitching. When scanning the core artistic areas, set the scanning resolution to 0.05mm, and automatically adjust the exposure time to the reflectivity of the wood surface (increase exposure for dark wood). The rotating stage collects data every 15° to ensure comprehensive coverage of deep openwork, intaglio, and other structures (openwork patterns require scanning from both sides). For details such as tool marks and pattern transitions, use a local magnification scanning mode (field of view diameter ≤ 100mm) to increase point cloud density. When scanning secondary areas, switch to 0.2mm resolution to increase scanning speed (reducing single-frame acquisition time by 50%). Use a "global + local" stitching strategy: first complete the overall stitching using positioning points, then fill in any missing areas. Using GeomagicControl software, the scanned data was compared with the outline of the actual wood carving to ensure that the point cloud coverage of the core area was ≥99%. Ten key measurement points (such as pattern depth and openwork hole diameter) were selected, and the measured values ​​were compared with the point cloud data using a coordinate measuring machine (accuracy ±0.02mm). The error should be ≤±0.05mm.

[0021] After scanning, ArtecStudio software was used to automatically stitch together point clouds from multiple perspectives by matching positioning points with curvature features. Statistical Outlier Removal and Radius Outlier Removal were employed to remove isolated points and noise generated during the scanning process. The point cloud of the core art area was resampled to ensure a point spacing of ≤0.1mm, retaining 0.05mm-level tool mark texture. Voxel Grid Filter was used for secondary areas to unify the point spacing to 0.2mm, compressing the data volume by more than 70% and improving subsequent modeling efficiency. The core art area data was saved in high-precision .ply format (including color and normal vectors) for subsequent pattern extraction and repair, while the secondary area data was converted to lightweight .obj format for overall model construction and parametric design.

[0022] Isolated points are caused by reflections from the wood carving surface, dust particles, equipment vibration, or ambient light interference during scanning. These isolated points deviate from the actual surface coordinates (e.g., the nearest point is >0.5mm), interfering with point cloud stitching accuracy and resulting in "spiky" or "protruding" appearances on the model surface, affecting subsequent pattern feature extraction. Noise points, on the other hand, are caused by irrelevant objects in the scanning background (such as supports, environmental clutter), sensor random errors, or data transmission errors. They increase the data volume, interfere with point cloud curvature calculations, and lead to deviations in the recognition of process features (such as tool mark depth). First, statistical outlier filtering is applied to the original point cloud dataset. Perform filtering processing, where In three-dimensional coordinates. The filtering process is as follows: for each point... Calculate its to Euclidean distance between the nearest neighbors : Calculate the average neighborhood distance for all points. and standard deviation : , Set a confidence coefficient; if the point... Average neighborhood distance satisfy Then determine Outliers are filtered out. Finally, the filtered point cloud dataset is output. This removes isolated points that significantly deviate from the true surface. Then, the point cloud dataset after outlier filtering is processed... Filtering is performed using a half-price filter. For each point... Search radius Calculate the number of neighboring nodes for all nodes within the range. : Set a minimum neighborhood threshold. ,like Then determine These are considered noise points and are filtered out. The final filtered point cloud dataset is obtained. This process removes background noise and sparse noise points, preserving high-density, effective point clouds. For example, the original point cloud of a Ming and Qing dynasty wooden dragon carving component contains 1.2 million points, including details such as dragon scales and whiskers, as well as support noise in the scanned background. Statistical outlier filtering (core region) is then applied. , Secondary areas , By dynamically adjusting the threshold to balance denoising and fidelity, approximately 12,000 outliers (such as isolated points caused by reflection) were removed. Radius filtering was then used. , (Core area) Remove approximately 3,000 speckled points (such as residual points from the support structure). Point cloud after filtering: 1.185 million points, retaining 98.75% of the effective points, with no loss of 0.05mm-level details such as dragon scale texture and dragon whiskers.

[0023] Next, an AI-powered automatic point cloud restoration algorithm automatically fills in the scanned blind spots. Craftsmen then manually correct deep cutouts, tool marks, and other handcrafted features on a VR platform using a VR environment and a digital pen, generating a 3D model with an error ≤0.1mm that retains its original authenticity. The AI-powered automatic point cloud restoration algorithm's restoration process is shown below: Based on the pre-processed point cloud... With network model Blind spot detection and modeling are performed based on mesh curvature. With normal vector Detect boundary points : ,in , Threshold for curvature and normal vector variation, typical value , PointFlow++ generative network is used for completion, which includes an encoder-decoder structure. The encoder converts point cloud features (boundary points) into... Global point cloud feature vectors extracted with PointFlow++ Mapped to a low-dimensional latent space The encoder is based on latent space features Generate point cloud of missing regions and the original point cloud spliced ​​together Loss function ,in , Ensure consistency between the geometry and the normal vector. Obtain the point cloud after automatic filling. It initially fills in blind spots such as deep hollowing and intaglio, and the point cloud integrity is ≥95%.

[0024] On a VR-based manual correction platform, craftsmen reference high-resolution photos of the original wood carvings, historical documents, and other materials to refine the filled point cloud. The process involves refinement. Craftsmen use a digital pen to "carve" in VR, adjusting the local coordinates of the point cloud to simulate the depth variations of hand-carved strokes (accuracy 0.1-0.5mm). For the smooth edges generated by AI, jagged features are manually added to recreate the "stiffness" of traditional carving. Hand-polishing marks are added to large smooth areas (such as the base plate), using a curvature brush tool to achieve basic local roughening. The VR-refined point cloud is then... Import the image into Rhino and refine it using its high-precision SubD modeling tools. Use a Wacom pen display with Rhino's SubD Sculpting tool, controlling the sculpting depth (pressure sensitivity) through pressure sensitivity. Corresponding depth This involves making local adjustments to key features such as the endpoints of tool marks and the turning points of patterns at the 0.05mm level. For example, the depth of tool marks is corrected to an accuracy of 0.1mm, and the serrated edges are refined to a 0.05mm level of undulation. The GaussianSmooth command in Rhino (with parameters...) is used. Smooth local surfaces or use the Sharpen tool in SculptingBrush to sharpen edges, ensuring that the subtle undulations of handcrafted features match the original wood carving. Capture refinement data in Rhino in real-time using Grasshopper's GeometryPipeline, automatically updating point cloud coordinates. ,in To refine and adjust the amount, to meet , , All values ​​are less than or equal to 0.05 mm. The final point cloud obtained is the result of manual correction. , , , , All values ​​are less than or equal to 0.05mm, retaining over 98% of the original handcrafted features. The hand-corrected point cloud... Physical data obtained with a coordinate measuring machine Perform ICP (Iterative Closest Point) registration and calculate the root mean square error (RMSE). ) and maximum deviation ( ) , core area , For qualified, secondary areas , If the deviation exceeds the threshold, the system returns to the VR platform to correct the high-deviation areas (such as the bottom of the tool marks or the corners of the cutouts) until the accuracy requirements are met.

[0025] Through local curvature mutation ( ) and normal vector jump ( Identify handcrafted feature points (such as tool mark endpoints and pattern transition points), compare the number of feature points before and after restoration, and calculate the retention rate. Perform lightweight processing on the corrected point cloud (e.g., voxel mesh filtering), retaining a point spacing of 0.05mm in the core area and reducing it to 0.1mm in secondary areas. Unify the normal vector direction to ensure compatibility with subsequent modeling software (such as Rhino). Obtain the restored point cloud. The format is .ply (including color and normal vectors) or .obj (lightweight model), with an error ≤0.1mm and a manual feature retention rate ≥98%.

[0026] Based on the repaired point cloud Coordinate measuring machine data of physical wood carvings Perform three-dimensional deviation analysis (accuracy ±0.02mm). The deviation calculation formula is as follows: The core area is required The deviation of key feature points (such as the bottom of tool marks) is ≤0.05mm. Calculate the retention rate of manual features: The manual feature points are identified through features such as curvature abrupt changes and normal vector anomalies, requiring a retention rate of ≥98%. If the deviation exceeds the threshold, the system returns to the VR platform for correction until the requirement is met. And the retention rate is ≥98%. Point cloud after final repair. .

[0027] Using the repaired point cloud Feature extraction is performed based on a geometric feature extraction module. Poisson surface reconstruction is used to generate a watertight triangular mesh. The core art area of ​​the network resolution has a surface density of ≥1000 faces / cm², and secondary areas have a surface density of ≥200 faces / cm², ensuring that pattern details are not lost. This is based on the grid curvature distribution. Identifying pattern boundary edges by abrupt change in normal vectors. ,in , For the edge The vertex, , Curvature distribution ( , (for Gaussian curvature) (Along the normal vector direction), pattern width ( For the pattern area, (As the center line).

[0028] Cultural relic experts used a 3D interactive platform to semantically annotate the extracted patterns, establishing a hierarchical structure based on plant categories (scroll patterns, lotus flower patterns), animal categories (dragon patterns, bat patterns), and object categories (eight auspicious symbols). Each element needed to include geometric feature parameters, cultural connotations (e.g., "bat" symbolizing "fortune"), craftsmanship techniques (relief / openwork carving), and historical origin. By annotating 100 Ming and Qing dynasty wood carvings, an initial library containing over 800 basic elements could be built, with plant elements accounting for 45% and animal elements for 30%. Craftsmen uploaded their knowledge through methods such as verbal transcription, hand-drawn annotations, and video annotations. For example, the language transcription process integrates dialect recognition models (such as Chaozhou dialect and Dongyang dialect) to automatically transcribe the craftsman's oral experience (such as "the tip of the bat pattern wing needs to be upturned by 15°") into text; hand-drawn annotations require drawing variations on a two-dimensional pattern image, and the system maps the hand-drawn information to three-dimensional geometric parameters through feature matching; video annotations require uploading videos of the carving process and annotating key process nodes (such as "three cuts are needed to position the face"), and the system automatically extracts the timestamp and key points of operation. The uploaded knowledge is first semantically understood using a BERT model to extract key information (such as tool selection and carving angle), and then its accuracy is reviewed by experts. If the knowledge confidence is ≥0.8 and there is no conflict with existing data, the parameters of the corresponding element are updated (such as adjusting the curvature threshold of a regional variation); if it is a new element, the DBSCAN clustering algorithm is used to determine whether it is an existing element variation to avoid duplicate entry. For example, after accessing the knowledge of 5 Dongyang woodcarving masters, the element library can be expanded to 2000+ entries within 3 months, adding process difference parameters between "Dongyang bat pattern" and "Chaozhou bat pattern".

[0029] The Neo4j graph database is used to store the relationships between elements, including element nodes (such as scroll patterns), knowledge nodes (such as "layered cutting process"), and craftsman nodes (knowledge contributors). Elements are associated with categories, features, and knowledge sources through edges such as "belongs to," "possesses," and "origin," forming a reasonable knowledge network. Elements can be quickly obtained in the following ways: combining geometric features (such as "curvature > 0.3 and depth > 3mm") and semantic tags (such as "symbolizing 'fortune, prosperity, and longevity'") to filter target patterns; uploading hand-drawn sketches or 3D model fragments, the system calculates similarity using Hausdorff distance and returns elements with a matching degree ≥ 85%, improving efficiency by more than 10 times compared to traditional manual searches.

[0030] A parametric design and modeling platform was built based on Rhino and Grasshopper, and this platform was then integrated with a dynamic element library. An interface for the dynamic element library was established within the Rhino and Grasshopper platforms to enable parametric access to cultural elements and binding of process attributes. 3D patterns from the cultural element dynamic library were converted into editable Rhino NURBS models, while simultaneously associating process constraint parameters, providing standardized data input for parametric design. A Python script was written using Grasshopper's VBScript component to read .3dm format pattern models (such as scrolling grass and bat patterns) from the dynamic library, parse geometric parameters (curvature, depth) using the RhinoCommonAPI, and store them as a Grasshopper parametric data structure. Referring to Grasshopper's Parameter.Attributes function, process attribute tags were added to each element (such as "cutout safety thickness ≥ 1.5mm" and "recommended Φ1mm ball end mill"), and custom attributes were used to bind parameters to process rules.

[0031] In Grasshopper, multiple design schemes are generated by adjusting element parameters, and the feasibility of the process is verified in real time using built-in algorithms. Leveraging Grasshopper's parametric modeling capabilities, design schemes can be generated in batches within one hour, while geometric calculations eliminate schemes that clearly do not meet process requirements, improving design efficiency. Using Grasshopper's Transform series components (such as Scale, Rotate, and Translate), parameters can be adjusted via input devices such as sliders and knobs (scaling 10%-200%, rotation 0°-360°) to generate transformed 3D models in real time. Utilizing Grasshopper's Loop structure and DataTree, parameter combinations (such as scaling step size of 10%, rotation step size of 15°) are traversed to automatically generate multiple schemes and store them as different Branch branches. A custom battery pack is developed using Grasshopper. MeshThicknessAnalysis is used to calculate the cutout thickness, CurvatureAnalysis is used to extract the minimum radius of curvature, thresholds are set (such as thickness < 1mm, radius of curvature < 0.5mm) to automatically mark risky schemes, and BooleanToggle is used to filter feasible schemes.

[0032] Using Rhino's SubD modeling tool and a graphics tablet, key hand-carved details were refined, with the system automatically synchronizing Grasshopper parametric data. Based on the parametric scheme, handcrafted artistic creation was incorporated, preserving the traditional handcrafted texture of wood carving while ensuring the modified model remained consistent with the parametric driving logic. A Wacom graphics tablet was connected to Rhino, and SubD Sculpting mode was enabled, with pressure sensitivity controlling the carving depth (pressure sensitivity value). Corresponding depth This involves smoothing and adjusting details such as character expressions and knife strokes, consistent with the logic of Rhino's built-in SculptingTools. Rhino's GaussianSmooth command is used to smooth local surfaces with the Sigma parameter set to 0.2mm, or the Sharpen tool in SculptingBrush is used to sharpen edges, simulating the texture of hand-carved knife marks. Modifications to the Rhino model are captured in real-time using Grasshopper's GeometryPipeline. The modified geometric parameters (such as curvature and depth) are extracted using the Surface.PointAt and Surface.NormalAt methods, and then updated in reverse to update the parametric model in Grasshopper, achieving bidirectional synchronization between manual correction and parameter updates.

[0033] In Grasshopper, the revised design is finalized for process verification, outputting a 3D model and process documentation suitable for production. Through a closed-loop process of "parametric initial screening - manual correction - final acceptance," the design is ensured to possess both artistic innovation and compliance with actual manufacturing requirements, achieving a seamless transition from design to production. In Grasshopper, a process verification battery pack is built, and Rhino's MeshThickness and CurvatureAnalysis commands are used to recalculate the cutout thickness of the revised model. With minimum radius of curvature Compare with the process rule library (such as hardwood) Cutting tool diameter Generate a verification report. Using Grasshopper's Export component, export the final solution as .STL (for 3D printing) or .STEP (for CNC machining) format, and generate a process specification document via TextPanel, including recommended tool models, cutting parameters, and other information to ensure direct use in production.

[0034] Inputting real-time wood properties and tool parameters, the carving process is simulated using finite element analysis to predict toolpath rationality. Digital simulation proactively identifies risks such as overcutting and chipping during the cutting process, optimizing the toolpath and avoiding material waste caused by traditional trial-and-error machining, thus improving process planning efficiency. The specific steps are as follows: Export the final solution generated in Rhino as a .step format, import it into finite element analysis software (such as ANSYS), and generate a tetrahedral mesh using the MeshGeneration function. Set the mesh size to 0.5 mm to balance computational accuracy and efficiency. Define the orthotropic properties of wood, considering the effects of grain direction and temperature on the modulus of elasticity, including moisture content. Texture direction and hardness The moisture content is At that time, the elastic modulus of wood along the grain direction (longitudinal direction) Elastic modulus in the vertical direction of the texture (lateral direction) And the elastic modulus in the direction perpendicular to the texture (radial) The calculation formula is: , , ,in =12 GPa (standard moisture content) =10%, standard temperature =Longitudinal elastic modulus at 20℃). =1.2 GPa (transverse elastic modulus). =1.5GPa (radial elastic modulus). =0.05 is the influence coefficient of moisture content. =0.001 / ℃ is the temperature influence coefficient. Texture direction. The angle between the cutting force and the toolpath affects the cutting force; the wood fiber direction vector is defined using a local coordinate system. .hardness Using Brinell hardness and depth of cut Related cutting resistance , This defines the cutting path length. It also sets the tool parameters, including the milling cutter diameter. The value ranges from 1 to 5 mm, affecting the distribution of cutting force. (Ball end mill cutting force formula) , The cutting force coefficient is the feed rate. The unit is mm / min. Tool material. The wear rate of carbide cutting tools is affected by wear rate. Ceramic knives .

[0035] Explicit Dynamics analysis was used to set the cutting step size. The material removal process is calculated using an updated Lagrange algorithm, outputting an equivalent stress contour map, toolpath cutting force, and tool wear. The threshold for the equivalent stress contour map is set at 40 MPa, representing the tensile strength of the wood; exceeding this threshold indicates a risk of chipping. The peak value of the toolpath cutting force is also considered. It needs to be less than the rated load capacity of the tool. Tool wear is calculated as follows: , This represents the cutting time.

[0036] Experienced craftsmen refine their toolpaths and mark solutions for wood defects in the simulation interface, forming a composite verification system. This system integrates the craftsmen's implicit experience (such as wood defect handling and toolpath optimization) into the digital twin model, overcoming the limitations of pure algorithm simulation and improving the accuracy and efficiency of process feasibility verification. The specific process is as follows: Develop a custom panel in Rhino, integrating ANSYS simulation result visualization capabilities, and supporting toolpath editing and defect marking. Adjust toolpath control points and modify cutting angles using the CurveEdit tool. (Default 45°) and step size (Default 0.5mm) Complete toolpath editing. Mark the locations of wood knots on the model. Set the avoidance radius Generate toolpaths that bypass defects. ,in For Centered on A sphere with radius .

[0037] By integrating empirical parameters, toolpath optimization rules are formulated, and strain coefficients are set. Experienced craftsmen use their expertise to set toolpath optimization rules, such as specifying that "deep relief cutting requires layering, with each layer having a specific depth..." "Use Grasshopper scripts to calculate the total depth" Decomposed into Layers. Adjust toolpath coefficients for different wood grain directions. Used to correct feed rate Algorithm-recommended initial toolpath The simulation yields a set of risk points. The craftsman corrected the knife path and obtained Risk points were obtained through re-simulation. ,like and If so, the correction will be accepted. Calculate the process feasibility index. ,in , Initial wear rate ( time value), The wear acceleration factor, For the number of model nodes, when The feasibility of the proposed solution was determined at that time.

[0038] From the Rhino design model to the ANSYS simulation model, and then to the Rhino human-computer interaction calibration, all steps are based on the same geometric model. Data such as wood properties and tool parameters are kept consistent through an intermediate format (.step), ensuring the linkage between simulation and design. The parametric design and manual correction results of the preceding modules serve as simulation inputs, and problems discovered in the process simulation feed back into design optimization, forming a closed loop of "design-simulation-correction," which, together with the design logic of "parametric generation + manual correction" mentioned earlier, forms a complete technical chain. By quantitatively analyzing cutting risks through finite element simulation and combining the experience of veteran craftsmen to handle unstructured problems (such as wood defects), the "trial and error cost" of traditional processes is transformed into "simulation optimization" in the digital space. The final output process solution can directly drive CNC machining, achieving efficient transformation from innovative design to manufacturing.

[0039] Intelligent matching of design schemes to process parameters is achieved through SWRL rules. Simultaneously, production data is fed back to the knowledge graph to drive algorithm self-learning, constructing a closed-loop knowledge system of "design-production-feedback." This allows the process knowledge graph to continuously evolve with production practice, improving the accuracy of process parameter recommendations and reducing reliance on human experience. The Neo4j graph database is used to store triplet relationships (entity-relationship-entity), including design entities, process entities, and production entities. Design entities include texture type (e.g., bat pattern) and geometric parameters (curvature, depth). Process entities include tool type, cutting method (e.g., Φ1mm ball end mill), and cutting parameters (feed rate, spindle speed). Production entities include fracture locations. Tool wear and machining time Write SWRL rules (such as "IF pattern depth"). AND curvature THEN recommends layered cutting and tool diameter The knowledge graph is imported through the Protein ontology editor to achieve automatic matching of design schemes to process parameters. (Number of cutting layers) ,in =1mm is an empirical threshold. Recommended formula for tool diameter. , =0.8 is the safety factor. The minimum radius of curvature. Data is collected at the production end via sensors (such as cutting force during fracture, tool wear, etc.). After ETL processing, it is mapped to a knowledge graph. If a certain solution breaks down at a certain location... equivalent stress If so, a "high-risk area" label is added to the pattern node in the knowledge graph; the wear model is updated based on tool wear data. ,in Corrections based on feedback from the production side (such as the actual wear rate of carbide cutting tools) ).

[0040] Using reinforcement learning algorithms (such as Q-Learning), and with the process feasibility index I as the reward function, the SWRL rule weights are updated: ; Where: state For design parameter combinations, actions Rewards for recommending process parameters Learning rate Discount factor .

[0041] The system automatically generates CNC machining code based on optimized process parameters, and adds process instructions and teaching modules, achieving seamless integration from design to production, shortening the cycle to within 3 days. Simultaneously, it solidifies process knowledge into transferable digital assets, lowering the barrier to skill inheritance. A wood carving-specific post-processor was developed based on Grasshopper's MillPost component, converting toolpaths from C to G code, supporting layered cutting instructions and tool compensation, with each layer's depth... Tool compensation is based on the wear model Automatically calculate tool radius compensation value .

[0042] Automatically extract process parameters from the knowledge graph and generate Markdown format documents, including design parameters (pattern type, scaling ratio, curvature), process parameters (tool type, number of layers, feed rate), and experienced craftsmen's experience. Develop interactive 3D tutorials using Three.js to demonstrate toolpaths and cutting sequences, supporting zooming in to view key process points (e.g., "feed rate should be reduced by 20% at knots"). Calculate the time from design to production ,in For parametric design time, For process simulation and correction time, For code and documentation generation time, total cycle In conjunction with production preparation time (such as timber stocking), the overall cycle is controlled within 3 days.

[0043] The processes described above with reference to the flowcharts in the embodiments disclosed in this invention can be implemented as computer software programs. The embodiments disclosed in this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium. When the computer program is executed by a central processing unit (CPU), it performs the functions defined in the methods of this application. It should be noted that the computer-readable medium described above in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wire segments, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to: wireless segments, wire segments, optical fibers, RF, etc., or any suitable combination thereof.

[0044] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0045] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are merely examples and do not limit the present invention. The purpose of the present invention has been fully and effectively achieved. The functions and structural principles of the present invention have been shown and explained in the embodiments. Without departing from the stated principles, the implementation of the present invention may have any variations or modifications.

Claims

1. A method for innovating traditional wood carving based on 3D scanning, characterized in that, The method includes: The wood carvings to be scanned are pre-processed. The core area is scanned with high precision using a 3D scanner, while the secondary areas are scanned with lightweight technology to obtain multi-view point cloud data. The multi-view point cloud data is then pre-processed. The preprocessed multi-view point cloud data is input into the point cloud network model, and the filled point cloud is output. The filled point cloud is then manually refined to obtain the repaired point cloud. Based on the point cloud generated after Poisson surface reconstruction and repair, a triangular network is generated, geometric features are extracted, the semantics of the patterns are labeled, and a dynamic element library is constructed and stored by combining manually uploaded knowledge. The parametric design and modeling processing platform connects to a dynamic element library to generate and filter design schemes. The design schemes are manually corrected, and the corrected design schemes are verified. Finite element analysis is performed on the verified design schemes to obtain the final scheme, which is then manually corrected. The final process parameters of the scheme are corrected by matching SWRL rules, and the SWRL rule weights are updated using a reinforcement learning algorithm.

2. The innovative method for traditional wood carving based on 3D scanning according to claim 1, characterized in that, The specific preprocessing procedure is as follows: Before scanning, use a soft brush and compressed air to clean the dust off the surface of the wood carving, and attach optical positioning markers to non-critical areas. Perform high-precision scanning on the core areas that carry the main cultural meaning and craftsmanship value, and perform lightweight scanning on secondary areas with weaker decorative features and lower craftsmanship complexity. Automatic stitching of multi-view point clouds is completed by matching positioning points with curvature features. Statistical outlier filtering and radius filtering are used to remove isolated points and noise. The point cloud of the core area is resampled, and voxel grid filtering is used to unify the indirectness of the secondary areas.

3. The innovative method for traditional wood carving based on 3D scanning according to claim 2, characterized in that, The specific process of obtaining the filled point cloud is as follows: A point cloud network model is constructed using PointFlow++ generative network, which includes an encoder and a decoder. The encoder extracts point cloud features based on the preprocessed multi-view point cloud data and the network model. Point cloud features are mapped to a low-dimensional latent space, and point clouds of missing regions are generated based on the latent space features. The point clouds of missing regions are then stitched together with preprocessed multi-view point cloud data to obtain the filled point cloud.

4. The innovative method for traditional wood carving based on 3D scanning according to claim 3, characterized in that, The specific process of obtaining the repaired point cloud is as follows: On a VR manual correction platform, craftsmen roughly refine the filled point cloud by referring to reference materials; the roughly refined point cloud is then imported into Rhino, and using SubD modeling tools and a digital screen, the carving depth is controlled by pressure sensitivity to refine key features; manual feature points are identified by local curvature abrupt changes and normal vector jumps, the number of feature points before and after the repair is compared, the normal vector direction is unified, and the repaired point cloud is obtained.

5. A method for innovating traditional wood carving based on 3D scanning according to claim 4, characterized in that, The process of constructing the dynamic element library is as follows: A triangular mesh is generated by reconstructing a Poisson surface. Based on the mesh curvature distribution and abrupt changes in the normal vector, the pattern boundary is identified and geometric features are obtained. Craftsmen use a 3D interactive platform to semantically annotate the extracted patterns and upload relevant knowledge. The uploaded knowledge is semantically understood using a BERT model to extract key information. If the knowledge confidence is ≥0.8 and does not conflict with existing data, the parameters of the corresponding element are updated. The Neo4j graph database is used to store the relationships between elements.

6. A method for innovating traditional wood carving based on 3D scanning according to claim 5, characterized in that, The process of obtaining the final solution is as follows: A parametric design and modeling platform was built based on Rhino and Grasshopper, which converts 3D patterns in a dynamic element library into editable NURBS models and associates them with process constraint parameters. Multiple design schemes are generated by adjusting element parameters, and the hollow thickness is calculated and the minimum radius of curvature is extracted using a custom battery pack. Thresholds are set to filter feasible schemes. Using the SubD modeling tool and digital display, key hand-carved details are corrected, and the system automatically synchronizes Grasshopper parametric data. Perform final process verification on the revised solution in Grasshopper; Input the real-time properties of the wood and the tool parameters, use finite element analysis to simulate the carving process, predict the rationality of the tool path, and obtain the final solution.

7. A method for innovating traditional wood carving based on 3D scanning according to claim 6, characterized in that, The specific matching process of the SWRL rule is as follows: The Neo4j graph database is used to store triple relationships, including design entities, process entities, and production entities. SWRL rules are written and imported into the knowledge graph through the Protege ontology editor. Data is collected by sensors on the production side, processed by ETL, and then mapped to the knowledge graph. A reinforcement learning algorithm is used, with the process feasibility index as the reward function, to update the weights of the SWRL rules.

8. A traditional wood carving innovation system based on 3D scanning, characterized in that, The system is used to perform a traditional wood carving innovation method based on 3D scanning as described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that is executed by a processor to implement the innovative method for traditional wood carving based on 3D scanning as described in any one of claims 1-7.

Images