A method of additive manufacturing model security structure formation and protection system

Abstract

This invention relates to the field of additive manufacturing technology and discloses a method and protection system for forming a secure structure for an additive manufacturing model. The method includes: acquiring a 3D model file to be protected, obtaining geometric baseline data and multi-view rendered images; identifying candidate geometric features in the 3D model file based on the geometric baseline data and multi-view rendered images, and obtaining a set of confirmed geometric features through automatic verification and / or user review; generating candidate secure embedding schemes based on the confirmed geometric feature set and a security feature knowledge base, and determining the selected secure embedding scheme; performing topological segmentation, geometric cutting, and / or geometric fusion operations on the 3D model file according to the selected secure embedding scheme to generate an encrypted 3D model file with embedded security features. This invention can improve the automation and adaptability of 3D model security feature embedding, and increase the difficulty of unauthorized copying, reverse engineering, or unauthorized printing.

Description

Technical Field

[0001] This invention belongs to the field of additive manufacturing technology, and more specifically, relates to a method for forming a safety structure for an additive manufacturing model and a protection system thereon. Background Technology

[0002] Additive manufacturing technology, due to its advantages such as eliminating the need for specialized molds, short manufacturing cycles, and the ability to form complex structures, has been widely applied in aerospace, medical implants, precision manufacturing, and customized industrial parts. In additive manufacturing, the product's structural design, functional layout, and manufacturing parameters are often initially represented by 3D model files, such as those in STEP, STL, IGES, and OBJ formats. Compared to traditional manufacturing methods, additive manufacturing relies more heavily on digital model files. If these 3D model files are illegally copied, leaked, or tampered with, unauthorized entities could use general-purpose slicing software and 3D printing equipment to directly manufacture corresponding parts, leading to intellectual property losses, uncontrollable product quality, and difficulties in tracing safety responsibilities.

[0003] Currently, security protection methods for 3D model files mainly include file encryption, access control, digital watermarking, manual addition of geometric perturbations, and manufacturing process detection. File encryption and access control primarily function during file storage, transmission, or access, effectively limiting unauthorized access to model files. However, once the model file is legally downloaded, distributed offline, or access is bypassed, these methods become ineffective in protecting the model itself. While digital watermarking or security marking can be used to trace the model's origin to some extent, they typically focus on embedding and retrieving identification information and may not necessarily cause substantial interference with the unauthorized printing and manufacturing process.

[0004] Furthermore, existing geometric perturbation or security feature embedding methods typically require designers to manually select embedding locations, feature types, and embedding parameters based on experience, making it difficult to adapt to 3D models of different shapes, sizes, and complexities. When dealing with batch model protection or protection of complex engineering models, manual methods are not only inefficient but also prone to affecting the normal manufacturing performance of the model due to inaccurate feature recognition, unreasonable embedding locations, or improper parameter settings. Moreover, existing security feature embedding methods are mostly static schemes. Once attackers understand their embedding rules, feature types, or parameter patterns, they may weaken or bypass the protection effect by modifying the slicing direction, adjusting slicing parameters, repairing the model topology, or reconstructing the model geometry.

[0005] On the other hand, additive manufacturing equipment, materials, slicing software, and process parameters vary considerably. The same model safety features may exhibit different forming results under different printing orientations, slicing parameters, printing equipment, or material systems. If the safety feature embedding strategy cannot be dynamically adjusted based on model geometry, manufacturing environment, and user feedback, it will be difficult to continuously increase the difficulty of unauthorized copying and printing while ensuring normal manufacturing for authorized users.

[0006] Therefore, there is an urgent need for a method and protection system for forming a safe structure of additive manufacturing models, which can automatically identify geometric features in 3D models, automatically recommend corresponding safety feature embedding schemes based on geometric feature attributes, and dynamically optimize the identification rules and embedding strategies through feedback information such as user review, slice preview or manufacturing verification, thereby improving the automation, adaptability and long-term effectiveness of 3D model safety protection. Summary of the Invention

[0007] This invention provides a method for forming a safety structure for an additive manufacturing model and a protection system to overcome the aforementioned defects in the prior art.

[0008] The present invention provides a method for forming a safety structure of an additive manufacturing model, which is achieved by the following specific technical means: A method for forming a safety structure for an additive manufacturing model, comprising: The three-dimensional model file to be protected is obtained, and the three-dimensional model file is read and preprocessed to obtain the geometric baseline data and multi-view rendering image corresponding to the three-dimensional model file. Based on the geometric baseline data and the multi-view rendered image, the geometric features in the 3D model file are identified to obtain a candidate geometric feature set; wherein, the candidate geometric feature set includes geometric feature type, geometric feature location information and identification confidence level; The candidate geometric feature set is verified to obtain the confirmed geometric feature set; wherein the confirmed geometric feature set is determined based on the automatic verification results and / or the user review results. Based on the confirmed set of geometric features and the preset security feature knowledge base, candidate security embedding schemes are generated; wherein, the security feature knowledge base includes mapping rules between geometric feature types and security feature types, and the mapping rules determine the corresponding security feature type based on at least one of the curvature attribute, closed contour attribute and solid thickness attribute of the geometric feature; Based on a preset selection strategy and / or user selection results, a selected security embedding scheme is determined from the candidate security embedding schemes; Based on the selected security embedding scheme, perform at least one geometric embedding operation among topological segmentation, geometric cutting, and geometric fusion on the 3D model file to generate an encrypted 3D model file with embedded security features; Obtain feedback samples for the candidate geometric feature set, the candidate secure embedding scheme, and / or the encrypted 3D model file, and adjust the recognition threshold, geometric consistency rule weight, mapping rule priority, and / or security feature parameter boundaries based on the feedback samples.

[0009] A further technical solution involves acquiring the 3D model file to be protected, and reading and preprocessing the 3D model file to obtain the geometric baseline data and multi-view rendered images corresponding to the 3D model file, including: Read the boundary representation entities of the 3D model file; The number of topological elements in the boundary representation entity is counted. Extract the bounding box size information of the 3D model file; Geometric parameters are acquired for candidate geometric regions in the 3D model file; wherein, the geometric parameters include at least one of face type, edge type, radius, depth, axial direction, curvature information, and position coordinates; The geometric baseline data is generated based on the geometric parameters; The 3D model file is rendered according to a preset perspective to obtain the multi-view rendered image.

[0010] A further technical solution involves identifying geometric features in the 3D model file based on the geometric baseline data and the multi-view rendered image to obtain a candidate geometric feature set, including: The geometric baseline data, the multi-view rendered image, the recognition prompt information, and the structured output specification are encapsulated into recognition request data; The identification request data is input into the geometric feature recognizer to obtain the structured identification result; The structured recognition results include geometric feature type, location description, evidence view, geometric reference identifier, and recognition confidence level. The geometric feature recognizer includes a rule recognizer, a visual language model recognizer, or a hybrid recognizer formed by combining a rule recognizer and a visual language model recognizer.

[0011] A further technical solution involves performing verification processing on the candidate geometric feature set to obtain a confirmed geometric feature set, including: The recognition confidence level is constrained to a preset numerical range, and the candidate geometric feature set is filtered based on the confidence level threshold; Based on the number of evidence views corresponding to the multi-view rendered image, the evidence consistency verification is performed on the candidate geometric feature set. Based on the geometric baseline data, a geometric consistency check is performed on the candidate geometric feature set; Deduplication is performed on candidate geometric features that have the same geometric reference identifier or the same semantic signature; The verified candidate geometric features are provided to a visual interactive interface to receive verification labels, rejection reasons and / or correction categories for the candidate geometric features; Based on the results of the verification process and / or the review results received by the visual interactive interface, the confirmed geometric feature set is generated.

[0012] A further technical solution, wherein generating candidate security embedding schemes based on the confirmed geometric feature set and a preset security feature knowledge base, includes: Read the available security feature types, applicable geometric feature types, embedding prerequisites, recommended parameter ranges, and manufacturing impact parameters from the security feature knowledge base; Match the geometric feature types in the confirmed geometric feature set with the applicable geometric feature types in the security feature knowledge base; If a match is successful, a candidate secure embedding scheme is generated based on the corresponding mapping rules; The candidate security embedding scheme includes security feature type, suggested embedding location, suggested parameter range, visibility parameter, and manufacturing impact parameter.

[0013] A further technical solution is that the mapping rule includes spline curve segmentation rules corresponding to curvature features; The spline curve segmentation rules include: When a target geometric feature with curvature attribute exists in the confirmed set of geometric features, the local parameter domain of the target host surface corresponding to the target geometric feature is extracted; Generate one or more two-dimensional spline curves within the local parameter domain; The two-dimensional spline curve is mapped onto the target host surface to obtain a three-dimensional surface segmentation line; Based on the three-dimensional curved surface segmentation lines, a topology segmentation operation is performed on the target host surface to form a secure segmentation interface on the target host surface; The appearance of the safety segmentation interface remains a continuous solid surface, and at least one of the segmentation position, segmentation length, curvature control amount, and number of segmentation strips of the safety segmentation interface is an adjustable parameter.

[0014] A further technical solution is that the mapping rules include closed contour segmentation rules; The closed contour segmentation rules include: When a target geometric feature with a closed contour attribute exists in the confirmed set of geometric features, the contour type of the closed contour corresponding to the target geometric feature is identified; Based on the contour type of the closed contour, a curvature closed segmentation interface or a polygonal closed segmentation interface is generated. Wherein, when the closed contour is a circular contour or an elliptical contour, the curvature closed segmentation interface is generated based on the circular contour or the elliptical contour; When the closed contour is formed by straight line boundaries, multiple boundaries that are connected end to end are selected from the straight line boundaries, and a closed polygonal contour is generated based on the multiple boundaries to form the polygonal closed segmentation interface; At least one of the following parameters is adjustable: the outline size, position offset, distance from the original geometric feature boundary, and number of superpositions of the curvature closed segmentation interface or the polygonal closed segmentation interface.

[0015] A further technical solution is that the mapping rules include hierarchical authentication code embedding rules; The hierarchical authentication code embedding rules include: When there is a target volume region in the confirmed geometric feature set whose entity thickness is greater than a preset thickness threshold and whose bounding box size meets the preset embedding conditions, obtain the authentication code image or authentication identifier matrix. Convert the authentication code image or the authentication identifier matrix into a binary matrix; The set of units to be embedded is determined based on the binary matrix; Each unit to be embedded in the set of units to be embedded is mapped to a discrete voxel cavity located inside the target volume region; Along the thickness direction of the target volume region, the discrete voxel cavity is distributed to different embedding layers; A layered authentication code is formed within the target volume region through geometric excision operations; Among them, at least one of the following parameters of the layered authentication code—number of layers, layer spacing, unit size, embedding depth, and embedding direction—is adjustable.

[0016] A further technical solution, wherein obtaining feedback samples for the candidate geometric feature set, the candidate secure embedding scheme, and / or the encrypted 3D model file, and adjusting the recognition threshold, geometric consistency rule weight, mapping rule priority, and / or security feature parameter boundaries based on the feedback samples, includes: Obtain at least one of the following: identification and correction samples, scheme acceptance samples, and manufacturing verification samples; If the feedback samples include identification and error correction samples, the identification threshold and / or the geometric consistency rule weights are adjusted based on the identification and error correction samples. If the feedback sample includes a scheme acceptance sample, the priority of the mapping rule and / or the default parameters of the security feature are adjusted based on the scheme acceptance sample. When the feedback sample includes a manufacturing verification sample, the safety feature parameter boundary is adjusted based on the manufacturing verification sample. The manufacturing verification sample is obtained by converting the encrypted 3D model file into a slice preview file and obtaining the slice preview result. The slice preview result includes at least one of the following: whether the preview passed, whether the defects are displayed, whether the support structure has been added, whether the parameters need to be adjusted, and an explanation of the reasons. The security feature parameter boundaries include upper and lower limits of at least one of the following: embedding size, position offset, number of layers, interlayer spacing, embedding depth, and number of segments.

[0017] A protection system is also provided, including: The model input and preprocessing module is used to acquire the 3D model file to be protected, and to read and preprocess the 3D model file to obtain the geometric baseline data and multi-view rendering image corresponding to the 3D model file. A geometric feature recognition module is used to recognize geometric features in the 3D model file based on the geometric baseline data and the multi-view rendered image to obtain a candidate geometric feature set; wherein, the candidate geometric feature set includes geometric feature type, geometric feature location information and recognition confidence level; The recognition result verification module is used to verify the candidate geometric feature set to obtain the confirmed geometric feature set; wherein, the confirmed geometric feature set is determined based on the automatic verification result and / or the user review result; A security feature scheme generation module is used to generate candidate security embedding schemes based on the confirmed set of geometric features and a preset security feature knowledge base; wherein, the security feature knowledge base includes mapping rules between geometric feature types and security feature types, and the mapping rules determine the corresponding security feature type based on at least one of the curvature attribute, closed contour attribute and solid thickness attribute of the geometric feature; The visualization interaction and user confirmation module is used to display the candidate geometric feature set, candidate security embedding scheme, recommended parameter range and manufacturing impact prompts in the 3D preview interface, and to receive user input for acceptance, rejection, category correction, parameter adjustment or reason annotation for the candidate geometric feature set and / or the candidate security embedding scheme; The scheme determination module is used to determine the selected security embedding scheme from the candidate security embedding schemes based on a preset selection strategy and / or the user selection result received by the visualization interaction and user confirmation module. The geometry embedding and export module is used to perform at least one geometry embedding operation among topological segmentation, geometric cutting and geometric fusion on the 3D model file according to the selected security embedding scheme, to generate an encrypted 3D model file with embedded security features; The feedback learning module is used to obtain feedback samples for the candidate geometric feature set, the candidate secure embedding scheme and / or the encrypted 3D model file, and adjust the recognition threshold, geometric consistency rule weight, mapping rule priority and / or security feature parameter boundary based on the feedback samples.

[0018] Compared with the prior art, the present invention has the following beneficial effects: This invention reads and preprocesses the 3D model file to be protected, generating geometric baseline data and multi-view rendered images. Based on the geometric baseline data and multi-view rendered images, candidate geometric features are identified, making the security feature embedding process no longer entirely dependent on manual observation and experience judgment, thus improving the automation level of 3D model geometric feature recognition. At the same time, after geometric feature recognition, the candidate geometric feature set is further subjected to confidence screening, evidence consistency verification, geometric consistency verification, deduplication, and user review, which can reduce the risk of misidentified features directly entering the security embedding process, thereby improving the reliability of the security feature embedding location and the embedded object.

[0019] This invention generates candidate security embedding schemes based on a confirmed set of geometric features and a security feature knowledge base. The security feature knowledge base includes mapping rules between geometric feature types and security feature types, enabling the system to automatically match security feature types such as spline curve segmentation, closed contour segmentation, and hierarchical authentication codes based on different geometric conditions such as curvature attributes, closed contour attributes, and solid thickness attributes. This improves the adaptability between the security feature embedding strategy and the model's geometric structure. Furthermore, the system can perform at least one geometric embedding operation—topological segmentation, geometric cutting, and geometric fusion—on the 3D model file according to the selected security embedding scheme, thereby forming security features in the surface topology, local geometric structure, or internal volume regions of the 3D model. These security features do not solely rely on file permission control but are embedded within the model's geometric or topological structure itself, thus continuing to provide security protection even after the model has been copied, downloaded, or processed offline.

[0020] This invention forms a combined security feature embedding mechanism through spline curve segmentation rules, closed contour segmentation rules, and hierarchical authentication code embedding rules. This enables different 3D models to generate differentiated security feature combinations. Without corresponding embedding rules, parameter settings, and manufacturing verification information, unauthorized users will find it difficult to accurately identify, restore, or eliminate the embedded security features, thereby increasing the difficulty of unauthorized copying, reverse engineering, and unauthorized printing.

[0021] This invention can acquire identification and correction samples, scheme acceptance samples, and manufacturing verification samples. Based on these feedback samples, it adjusts the identification threshold, geometric consistency rule weights, mapping rule priorities, and safety feature parameter boundaries. This ensures that the system does not perform safety feature embedding in a one-time, static manner, but rather continuously optimizes based on user review results, candidate scheme selection results, slicing preview results, and manufacturing verification results. Therefore, this invention can adapt to different model types, user preferences, slicing software, and additive manufacturing equipment, improving the long-term effectiveness and dynamic evolution capability of model safety protection strategies.

[0022] When generating encrypted 3D model files with embedded security features, this invention can also simultaneously generate scheme summaries and operation logs, recording security feature types, embedding parameters, target geometry references, location indexes, user confirmation schemes, geometry embedding operation types, export paths, preview verification results, and feedback records, thereby forming traceable export records, which facilitates subsequent security verification, accountability tracking, policy maintenance, and feedback learning. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the additive manufacturing model safety structure forming system provided in an embodiment of the present invention; Figure 2 This is a schematic flowchart of the additive manufacturing model safety structure formation method provided in the embodiments of the present invention; Figure 3 This is a schematic diagram of the security feature mapping rules provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the slice verification and feedback feedback process provided in an embodiment of the present invention; Figure 5 This is a three-dimensional schematic diagram of the encryption protection formed after adding a single spline curve segmentation security feature in an embodiment of the present invention; Figure 6 This is a three-dimensional schematic diagram of the encryption protection formed after adding rectangular segmentation security features in an embodiment of the present invention; Figure 7 This is a three-dimensional schematic diagram of the encryption protection formed after adding elliptical segmentation security features in an embodiment of the present invention; Figure 8This is a three-dimensional schematic diagram of how a QR code pattern is embedded in a cube model in a partitioned and layered manner to form an embedded solid cavity for encryption protection in an embodiment of the present invention; Figure 9 yes Figure 8 A complete diagram of the QR code as seen from a top-down view; Figure 10 yes Figure 8 A schematic diagram of randomly scattered unreadable QR codes observed in the main view; Figure 11 yes Figure 8 A schematic diagram of randomly scattered, unreadable QR codes as observed in the right-middle view. Detailed Implementation

[0024] The embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and should not be construed as limiting the scope of the invention.

[0025] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of the present invention, including various details to aid understanding. These details should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the invention. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0026] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. Those skilled in the art will understand that the embodiments described in this application can be combined with other embodiments.

[0027] This embodiment provides a method for forming a secure structure for an additive manufacturing model. This method can be applied to independent software systems, CAD plugins, cloud-based model processing platforms, or additive manufacturing preprocessing systems. It is used to identify geometric features of a 3D model file to be protected and automatically generate a security feature embedding scheme based on the identified geometric features. This allows the security features to be embedded in the 3D model, thereby increasing the difficulty of cracking the 3D model when it is illegally copied, reverse engineered, or printed without authorization.

[0028] It should be noted that the "3D model file" referred to in this embodiment can be STEP, STP, STL, IGES, OBJ, or other data files capable of expressing 3D geometry. Preferably, the 3D model file is a STEP format file so that the system can read geometric information such as boundary representation entities, topological surfaces, topological edges, surface types, and bounding box dimensions. The "encrypted 3D model file" referred to in this embodiment is not limited to an encrypted file in the traditional cryptographic sense, but refers to a protected 3D model file formed by embedding security features into the geometric structure, topological structure, or internal volume region of the original 3D model.

[0029] like Figure 1 As shown, this embodiment provides a protection system, including a model input and preprocessing module 1, a geometric feature recognition module 2, a recognition result verification module 3, a security feature scheme generation module 4, a scheme determination module 5, a geometric embedding and export module 6, and a feedback learning module 7.

[0030] The model input and preprocessing module 1 is used to acquire the 3D model file to be protected and generate geometric baseline data and multi-view rendered images; the geometric feature recognition module 2 is used to recognize candidate geometric feature sets based on geometric baseline data and multi-view rendered images; the recognition result verification module 3 is used to verify the candidate geometric feature sets to obtain the confirmed geometric feature sets; the security feature scheme generation module 4 is used to generate candidate security embedding schemes based on the confirmed geometric feature sets and the security feature knowledge base; the scheme determination module 5 is used to determine the selected security embedding scheme based on the preset selection strategy and / or the user selection result; the geometric embedding and export module 6 is used to perform geometric embedding operations according to the selected security embedding scheme and generate encrypted 3D model files; the feedback learning module 7 is used to adjust the recognition threshold, geometric consistency rule weight, mapping rule priority and / or security feature parameter boundaries based on feedback samples.

[0031] In one implementation, the visualization interaction and user confirmation module can be integrated into the scheme determination module 5, or it can be set as an independent module. It is used to display candidate geometric features, candidate safe embedding schemes, recommended parameter ranges and manufacturing impact prompts, and to receive user input for acceptance, rejection, category correction, parameter adjustment or reason annotation.

[0032] like Figure 2 As shown in the figure, this embodiment provides a method for forming a safety structure of an additive manufacturing model, which includes the following steps.

[0033] S101. Obtain the 3D model file to be protected, and read and preprocess the 3D model file to obtain the geometric baseline data and multi-view rendering images corresponding to the 3D model file.

[0034] In one implementation, the user uploads the 3D model file to be protected through a visual interactive interface. The system generates a unique model identifier for the 3D model file and calls a geometry engine to read the boundary representation entities in the 3D model file. Specifically, the system can read the STEP file using a geometry kernel such as OCCT or PythonOCC to obtain the overall topological entity and traverse the topological elements such as faces, edges, shells, and solids within the topological entity. The system further counts the number of topological elements in the model and extracts the bounding box size information of the model. The bounding box size information may include the minimum and maximum coordinates, as well as the length, width, and height dimensions of the model in the X, Y, and Z directions.

[0035] To facilitate subsequent identification and verification, the system also collects geometric parameters from candidate geometric regions in the 3D model file. These geometric parameters include at least one of the following: face type, edge type, radius, depth, axial direction, curvature information, and position coordinates.

[0036] For example, when the system needs to identify a circular hole feature, it can traverse all surfaces in the model and determine whether the current surface is a cylindrical surface through surface adaptation. If the current surface is a cylindrical surface, it further extracts parameters such as the radius, axial direction, cylindrical surface coverage angle, and end opening position of the cylindrical surface. For multiple cylindrical surfaces with the same axis, the same radius, and axial distance meeting preset conditions, the system can cluster them into the same candidate geometric feature. If the cylindrical surface corresponding to a candidate geometric feature is concave and the circumferential coverage angle is equal to or close to the full circumference, the candidate geometric feature can be initially identified as a hole feature; if the circumferential coverage angle is less than the full circumference, the candidate geometric feature can be initially identified as a cylindrical groove feature.

[0037] For example, when the system needs to identify straight line boundaries or closed contours, it can traverse the topological edges in the model and perform geometric curve adaptation on each topological edge to determine its geometric type as a straight line, circle, ellipse, or spline curve. When several topological edges can be connected end to end to form a closed contour, the system can further determine whether the closed contour is a circular contour, elliptical contour, or polygonal contour.

[0038] Through the above preprocessing, the system can generate geometric baseline data. This geometric baseline data can be saved in JSON or other structured data file format, and includes at least the model identifier, bounding box size, number of topological elements, candidate geometric regions, geometric parameters of the candidate geometric regions, geometric reference identifiers, and preliminary geometric type. This geometric baseline data is used not only for subsequent geometric feature recognition but also for consistency verification of the recognition results, thereby reducing the risk of incorrect embedding due to misidentification.

[0039] After generating geometric baseline data, the system renders the 3D model file according to preset viewpoints to obtain multi-view rendered images. The preset viewpoints may include at least one of the following: front view, rear view, left view, right view, top view, bottom view, and isometric view. By generating multi-view rendered images, the system can further obtain the model's appearance, contours, openings, concave and convex structures, and relative positional relationships beyond the geometric baseline data, facilitating subsequent geometric feature recognition by an auxiliary geometric feature recognizer to identify complex geometric features.

[0040] Through the above steps, the system can simultaneously obtain geometric baseline data based on the geometric kernel and image evidence based on view rendering, so that the subsequent recognition process does not rely solely on a single image or a single geometric rule, but can form a cross-validation basis between topological data and view data.

[0041] In a more specific implementation, when generating the geometric baseline data, the system does not only collect the original geometric parameters, but further uses geometric kernel rules to perform preliminary analysis on the candidate geometric regions, so that the geometric baseline data carries preliminary geometric type reference information for subsequent constraint and verification of the geometric feature recognizer's recognition process.

[0042] In one example, to identify circular hole features, the system traverses the topological surfaces in the 3D model file and calls a geometric kernel function to determine whether the current topological surface is a cylindrical surface. If the current topological surface is a cylindrical surface, the system further extracts at least one parameter from the cylindrical surface, including its radius, axial direction, circumferential coverage angle, and end opening position, and performs concavity / convexity attribute discrimination on the cylindrical surface. Since a geometric feature may be formed by multiple topological surfaces, the system clusters cylindrical surfaces that meet the conditions of being coaxial, having the same radius, and having an axial distance that satisfies a preset clustering condition into the same candidate geometric region. After clustering, the system performs a preliminary type determination of the candidate geometric region based on the concavity / convexity attribute and circumferential coverage angle of the cylindrical surface; for example, when the circumferential coverage angle of the concave surface cluster group is equal to or close to a complete circumference, the candidate geometric region is preliminarily determined to be a hole feature; when the circumferential coverage angle of the concave surface cluster group is less than a complete circumference, the candidate geometric region is preliminarily determined to be a cylindrical groove feature.

[0043] The system assigns a unique geometric reference identifier to each candidate geometric region determined by the rules, and associates and stores the initially determined geometric feature type, the geometric reference identifier, and at least one parameter among the calculated radius, depth, axial direction, and position coordinates with the geometric baseline data. In this way, the geometric baseline data contains both original geometric topology information and preliminary engineering semantic information obtained based on geometric rules.

[0044] S102. Based on geometric baseline data and multi-view rendered images, identify geometric features in the 3D model file to obtain a set of candidate geometric features.

[0045] In one implementation, the system encapsulates geometric baseline data, multi-view rendered images, recognition prompts, and structured output specifications into recognition request data. The recognition request data may include model identifiers, multi-view rendered image paths, bounding box dimensions, candidate geometric regions, geometric reference identifiers, prompts, and the desired output data structure.

[0046] In one implementation, the identification request data includes model identifiers, multi-view rendered image paths, geometric baseline data, geometric reference identifiers of candidate geometric regions, identification prompts, and structured output specifications. The structured output specifications constrain the geometric feature recognizer to output identification results according to preset fields, which include at least one of geometric feature type, location description, evidence view, geometric reference identifier, identification confidence level, manufacturing risk warning, and secure embedding potential.

[0047] The system inputs the recognition request data into a geometric feature recognizer to obtain a structured recognition result. The geometric feature recognizer can be a rule recognizer, a visual language model recognizer, or a hybrid recognizer formed by combining a rule recognizer and a visual language model recognizer.

[0048] In implementations employing rule-based recognition, the system performs rule matching on candidate geometric regions based on geometric parameters in the geometric baseline data. For example, for cylindrical surface clustering results, the system can determine whether it is a hole, blind hole, cylindrical groove, or cylindrical boss based on the concavity / convexity attributes of the cylindrical surface, the circumferential coverage angle, the number of end openings, and the depth value. For closed boundary sets, the system can determine whether it is a rectangular groove, triangular boss, circular opening, or elliptical opening based on the boundary type, the number of boundaries, the connection relationship between the beginning and end, and the plane on which the contour lies.

[0049] In the implementation using a visual language model recognizer, the system inputs multi-view rendered images and geometric baseline data into the visual language model recognizer. Based on geometric appearance evidence from the multi-view images, as well as candidate regions, bounding box information, and geometric reference identifiers provided by the geometric baseline data, the visual language model recognizer performs semantic recognition and description of features such as holes, grooves, bosses, fillets, chamfers, threads, and freeform surfaces in the 3D model. The structured recognition results output by the visual language model recognizer may include fields such as geometric feature type, location description, evidence view, geometric reference identifier, recognition confidence, manufacturing risk warning, and security embedding potential.

[0050] In the implementation using a hybrid recognizer, a rule recognizer is used to extract candidate geometric regions and geometric parameters at the geometric kernel level, while a visual language model recognizer is used to semantically supplement and positionally describe the candidate geometric regions by combining multi-view rendered images. By combining the two, the system can reduce the illusion recognition problem caused by simple image recognition and also make up for the shortcomings of pure geometric rules in recognizing complex free-form surfaces, combined grooves, or irregular protrusions.

[0051] When the geometric feature recognizer is a visual language model recognizer, the system inputs the multi-view rendered images and geometric baseline data into the visual language model recognizer, and constrains the output format of the visual language model recognizer through standard output examples, so that it prioritizes the output of structured recognition results consistent with the geometric baseline data based on multiple evidence views and geometric reference marks.

[0052] Through this step, the system obtains a set of candidate geometric features. Each candidate geometric feature in the set may include a geometric feature type, geometric feature location information, geometric reference identifier, evidence view, and identification confidence level. The geometric feature type may include at least one of the following: hole, groove, boss, fillet, chamfer, thread feature, curvature region, closed contour region, and region of sufficient thickness. The geometric feature location information may be three-dimensional coordinates, relative position description, corresponding face / edge number, or geometric reference identifier.

[0053] It should be noted that, within the hybrid recognition framework provided in this embodiment, the visual language model recognizer does not perform unconstrained target detection on multi-view rendered images. Instead, guided by geometric prior information provided by the geometric baseline data, it verifies, describes, and supplements candidate geometric regions. Specifically, the geometric baseline data encapsulated in the recognition request data includes a list of candidate geometric regions, geometric reference identifiers corresponding to each candidate geometric region, and their position range and size parameters in three-dimensional space. The recognition prompt information is used to instruct the visual language model recognizer to prioritize the region corresponding to the geometric reference identifier, combining at least one of the following information: position description from the multi-view rendered image, evidence view, recognition confidence, manufacturing risk warning, and secure embedding potential.

[0054] In this way, the visual language model recognizer can supplement semantics based on the topological priors provided by the geometric kernel, rather than simply guessing the location of geometric features based on the appearance of the image. This reduces the risk of misidentification and improves the consistency between the structured recognition results and the underlying topological structure of the model. In other implementations, when the visual language model recognizer discovers geometric regions that are inconsistent with the geometric baseline data or are suspected of being missed based on multi-view rendered images, it can also mark these regions as candidate geometric features to be verified and proceed to the subsequent verification process.

[0055] S103. Verify the candidate geometric feature set to obtain the confirmed geometric feature set.

[0056] Since the results of geometric feature recognition may be affected by model complexity, viewpoint occlusion, image rendering quality or recognition misjudgment, this embodiment does not directly use all candidate geometric features for security feature embedding after generating the candidate geometric feature set. Instead, the candidate geometric feature set is verified by the recognition result verification module 3.

[0057] In one implementation, the system first constrains the recognition confidence level to a preset numerical range. For example, the system can constrain the recognition confidence level to the range of 0 to 1. If the recognition confidence level of a candidate geometric feature is lower than a preset confidence threshold, the system can mark the candidate geometric feature as an unconfirmed feature or remove it from the candidate geometric feature set.

[0058] Subsequently, the system performs evidence consistency verification on the candidate geometric feature set based on the number of evidence views corresponding to the multi-view rendered images. For example, when a candidate geometric feature has visible evidence only in a single-view image, but lacks corresponding geometric appearance or positional support in other views, the system can lower its credibility level; when a candidate geometric feature has consistent evidence in two or more view images, the system can increase its credibility level.

[0059] Furthermore, the system performs geometric consistency checks on the candidate geometric feature set based on the geometric baseline data. For example, when a candidate geometric feature is identified as a hole feature, the system checks whether there is a corresponding cylindrical surface, circular boundary, or opening pairing relationship in the geometric baseline data; when a candidate geometric feature is identified as a closed contour, the system checks whether there is a set of closed boundaries connected end to end in the geometric baseline data; when a candidate geometric feature is identified as a region with sufficient thickness, the system checks whether the solid thickness and bounding box size of the corresponding region meet the preset embedding conditions.

[0060] In one implementation, when the candidate geometric feature is identified as a hole feature, the system needs to find cylindrical surfaces, circular edges, opening boundaries, or corresponding geometric references in the geometric baseline data as supporting evidence. When the candidate geometric feature is identified as a closed contour feature, the system needs to find a set of topological edges that can be connected end-to-end or corresponding closed boundary markers in the geometric baseline data as supporting evidence. When the candidate geometric feature is identified as a region with sufficient solid thickness, the system needs to find solid thickness, bounding box size, or target volume region markers in the geometric baseline data as supporting evidence. If the number of evidence views for a candidate geometric feature is insufficient, or the geometric baseline data cannot support the identification result of the candidate geometric feature, the system marks the candidate geometric feature as an unconfirmed feature or excludes it from the subsequent secure embedding decision process.

[0061] If multiple candidate geometric features with the same geometric reference identifier or the same semantic signature exist in the candidate geometric feature set, the system performs deduplication. The semantic signature can be formed by combining geometric feature type, location description, geometric reference identifier, and evidence view. For duplicate candidate geometric features, the system can retain the candidate geometric features with higher identification confidence and remove the remaining duplicate entries.

[0062] In one implementation, the system can also provide the verified candidate geometric features to a visual interactive interface to receive user feedback labels, rejection reasons, and / or revised categories for the candidate geometric features. For example, a user can perform operations such as retaining, canceling retention, category correction, or adding comments for a candidate geometric feature. If a user believes that the system has misidentified a cavity as a hole, the category of the candidate geometric feature can be corrected from hole to cavity. If a user believes that a candidate geometric feature is unsuitable as a security feature host region, the user can cancel the retention of the candidate geometric feature and fill in the rejection reason.

[0063] The system generates a set of confirmed geometric features based on the results of automatic verification and / or the review results received through a visual interactive interface. This set of confirmed geometric features is directly used as input for the generation of subsequent security feature schemes, thereby preventing unconfirmed or incorrectly identified geometric features from entering the embedding decision process.

[0064] Through the above steps, this embodiment can introduce geometric consistency verification and optional human-machine review mechanism after automatic identification, reduce the risk of incorrect embedding, and improve the matching degree between the security feature embedding scheme and the actual geometric structure of the 3D model.

[0065] In one implementation, after verifying candidate geometric features, the system generates a reviewed feature file. This file records candidate geometric features retained by the user, candidate geometric features removed from the user's list, and the corresponding reasons for removal. The system also generates a recognition feedback file, which records the user's reasons for reviewing the candidate geometric features, notes, overall comments, and next steps suggestions. The reviewed feature file serves as input for generating candidate secure embedding schemes, while the recognition feedback file serves as the basis for subsequently adjusting recognition thresholds, geometric consistency rule weights, or recognition prompts.

[0066] S104. Based on the confirmed set of geometric features and the preset security feature knowledge base, generate candidate security embedding schemes.

[0067] In one implementation, the system pre-establishes a security feature knowledge base. This knowledge base can be stored in the form of rule tables, databases, knowledge graphs, or structured configuration files. The knowledge base records available security feature types, applicable geometric feature types, embedding prerequisites, recommended parameter ranges, visibility parameters, and manufacturing impact parameters.

[0068] Available security feature types may include spline curve segmentation security features, closed contour segmentation security features, layered authentication code security features, microcavity security features, microlattice security features, character cavity security features, or other geometric security features that can be embedded in a 3D model. The applicable geometric feature type is used to define the host geometry suitable for embedding a particular security feature. Embedding prerequisites may include host region thickness thresholds, curvature conditions, closed contour conditions, minimum manufacturable size, distance from the model's outer surface, and distance from functional areas. Recommended parameter ranges may include embedding size, position offset, number of segments, interlayer spacing, embedding depth, number of layers, voxel size, and orientation parameters. Manufacturing impact parameters may include the level of impact of the security feature on the model's appearance, strength, support structure, printing time, or forming stability.

[0069] The system matches the geometric feature types in the confirmed geometric feature set with the applicable geometric feature types in the security feature knowledge base. If a match is successful, the system generates candidate security embedding schemes based on the corresponding mapping rules.

[0070] In this embodiment, the security feature knowledge base includes at least the following three types of mapping rules, see [link to relevant documentation]. Figure 3 As shown, see the specific segmentation results. Figures 5 to 11 .

[0071] The first category is the spline curve segmentation rules corresponding to curvature features.

[0072] When a target geometric feature with curvature attributes is confirmed to exist in the set of geometric features, such as a curved groove, a rounded corner region, a freeform surface region, a circular boss, or a threaded feature region, the system matches the target geometric feature to the spline curve segmentation safety feature.

[0073] Specifically, the system first extracts the local parameter domain of the target host surface corresponding to the target geometric features. For a target host surface, the system can obtain its surface parameter range and determine the start point, end point, control points, and positional offset relative to the original geometric feature boundary within the surface parameter domain. Subsequently, the system generates one or more two-dimensional spline curves within the local parameter domain. The two-dimensional spline curves can be generated by interpolation or by fitting.

[0074] In one implementation, the system determines the length, curvature control amount, and number of spline curves based on the parameter domain size, curvature variation, and preset safety level of the target host surface. For example, when the safety level is high, the system can generate multiple spline curves with different curvatures; when the host area is small or the manufacturing precision is limited, the system can reduce the number of spline curves or increase the distance between the spline curves and the functional boundary.

[0075] In one implementation, the parameters for segmenting the safety feature using a spline curve include a position offset δ, a length parameter λ, a curvature control parameter κ, and the number of segments m. The position offset δ defines the minimum distance between the spline curve and the original geometric feature boundary; the length parameter λ defines the extent to which the spline curve extends relative to the local parameter domain size or the straight-line distance to its endpoints; the curvature control parameter κ adjusts the curvature of the spline curve through control point offset, control point weight, or sampling point offset; and the number of segments m determines the number of spline segments generated on the same host surface. The system can constrain and adjust these parameters based on the model size, local curvature, safety level, and the minimum manufacturable feature size of the additive manufacturing equipment.

[0076] After generating the 2D spline curve, the system maps the 2D spline curve to the target host surface to obtain a 3D surface segmentation line. Based on this 3D surface segmentation line, the system performs a topology segmentation operation on the target host surface to form a safe segmentation interface. The topology segmentation operation can be implemented using the surface segmentation function, boundary representation entity segmentation function, or equivalent topology segmentation algorithm within the geometric kernel; for example, in an implementation using the OCCT geometric kernel, the topology segmentation operation can be implemented using its boundary representation entity segmentation tool. This safe segmentation interface does not significantly change the model's surface thickness, allowing the model to appear as a continuous solid surface, but its local topology has been safely identified. Specifically, the safe segmentation interface can be a topology segmentation interface with zero or near-zero thickness, or a hidden segmentation interface that appears continuous but can produce differences in model topology, slicing paths, or manufacturing parameter responses.

[0077] In one implementation, the two-dimensional spline curve can be represented as a B-spline curve or a rational B-spline curve. The system determines multiple control points in the local parameter domain of the target host surface and generates a two-dimensional spline curve based on the control points, curve order, basis functions, and optional weights. For rational B-spline curves, the system can also change the curvature of the spline curve in the local parameter domain by adjusting the control point weights. The spline curve can be generated by interpolation or by approximation.

[0078] Since the position, length, curvature, number of segments, and position offset of the secure segmentation interface can be parameterized, authorized users can identify or control this security feature based on preset manufacturing parameters, slicing parameters, or subsequent verification methods. However, unauthorized users, lacking corresponding parameters and identification rules, will find it difficult to accurately determine the design intent and parameter combination of the spline segmentation security feature even if they obtain the model file, thus increasing the difficulty of model replication and cracking.

[0079] In one implementation, the spline curve segmentation security feature can be coupled with the slicing direction and / or slicing parameters. When the encrypted 3D model file is sliced ​​according to the first slicing direction or the first slicing parameters, the slicing path can be substantially consistent with the original model without embedded security features. When the encrypted 3D model file is sliced ​​according to the second slicing direction or the second slicing parameters, the slicing path exhibits path splitting, path breaking, or material path repetition at the positions corresponding to the 3D surface segmentation lines. These path splitting, path breaking, or material path repetition phenomena can be distributed along the model thickness direction, making it difficult for unauthorized users to consistently obtain the expected printing results without the corresponding slicing direction and slicing parameters.

[0080] In one embodiment, the spline curve segmentation security feature may include multiple spline curve segmentation lines. These multiple spline curve segmentation lines may be parallel, intersecting, perpendicular, inclined, or spaced apart from each other; different spline curve segmentation lines may have different curvature control values, lengths, positional offsets, and segmentation depths. By changing the number, relative angles, and spatial positions of the spline curve segmentation lines, the encrypted 3D model file can produce differentiated slice path behaviors under different slice directions and / or slice parameters, thereby increasing the complexity of the security feature combination and the difficulty of cracking.

[0081] In this way, the system can form a safe segmentation interface corresponding to the spline curve parameters on the target geometric features with curvature attributes, so that the safe features are combined with the curvature region of the original model, reducing the possibility of the safe features being directly identified or deleted as a whole, and improving the adaptability between the embedding position of the safe features and the geometric structure of the model.

[0082] The second category is closed contour segmentation rules.

[0083] When a target geometric feature with a closed contour attribute is confirmed to exist in the set of geometric features, the system identifies the contour type of the closed contour corresponding to the target geometric feature and generates a curvature closed segmentation interface or a polygonal closed segmentation interface based on the contour type of the closed contour.

[0084] In one implementation, the system traverses the topological edges in the target geometric region and performs geometric curve adaptation on each topological edge to determine whether it is a straight line, circle, ellipse or other curve.

[0085] If the closed contour is a circular or elliptical contour, the system generates a curvature closed segmentation interface based on the circular or elliptical contour. The curvature closed segmentation interface can be formed by offsetting equidistantly along the original closed contour, or it can be formed by scaling the original closed contour according to a preset ratio inside or outside the original closed contour.

[0086] If the closed contour is formed by straight line boundaries, such as a rectangular groove, a triangular boss, a square pocket structure, or a chamfered boundary area, the system selects multiple consecutive boundaries from the straight line boundaries and generates a closed polygonal contour based on these boundaries. In a specific example, the system can select four consecutive straight line boundaries to form a rectangular or general quadrilateral contour, and use this closed polygonal contour as the bottom contour for the segmentation operation, forming a polygonal closed segmentation interface in the target area.

[0087] The contour dimensions, positional offset, distance from the original geometric feature boundary, and number of superpositions of the curvature closed segmentation interface or polygonal closed segmentation interface can all be adjustable parameters. For example, when the target geometric feature area is large, the system can generate multiple closed contour segmentation interfaces in that region and distribute them nested at a preset interval; when the target geometric feature is close to the functional boundary of the model, the system can increase the distance between the closed segmentation interface and the original geometric feature boundary to reduce the impact on the functional area of ​​the model.

[0088] By using closed contour segmentation rules, the system can transform the geometric boundaries, openings, or grooves that originally exist in the model into safe feature host regions, making the safe feature embedding more concealed, and can form differentiated embedding structures according to different closed contour types.

[0089] In one implementation, the closed contour segmentation security feature can alter the slicing software's identification of model material regions, support material regions, or outer contour paths. For example, when the closed contour segmentation interface is a polygonal closed segmentation interface, in the first slicing direction, the local area defined by the polygonal closed segmentation interface can be identified by the slicing software as a support material region, an empty region, or an independent object region; in the second slicing direction, the local area can remain consistent with the original model material region. As another example, when the closed contour segmentation interface is a circular or elliptical closed segmentation interface, in a specific slicing direction, the slicing path can form path splits or path breaks at the position corresponding to the curvature closed segmentation interface. Thus, the closed contour segmentation security feature enables the same encrypted 3D model file to present different manufacturing path results under different slicing directions, slicing parameters, or object splitting strategies.

[0090] In this way, the system can use the existing circular, elliptical or polygonal contours in the 3D model to generate corresponding closed segmentation interfaces, so that the safety features are combined with the original geometric boundaries of the model, thereby improving the concealment of the safety feature embedding and the structural adaptability.

[0091] The third category is hierarchical authentication code embedding rules.

[0092] When a target volume region with an entity thickness greater than a preset thickness threshold and a bounding box size that meets the preset embedding conditions is confirmed in the geometric feature set, the system matches the target volume region to the hierarchical authentication code security feature.

[0093] Specifically, the system acquires an authentication code image or an authentication identifier matrix. The authentication code image can be a QR code, character identifier, number matrix, enterprise identifier, or other authentication information that can be converted into a matrix structure. The system converts the authentication code image or authentication identifier matrix into a binary matrix and determines the set of units to be embedded based on the binary matrix. For example, the system can determine the black units in the binary matrix as the units to be embedded and the white units as the non-embedded units.

[0094] After determining the set of units to be embedded, the system maps each unit in the set to be embedded as a discrete voxel cavity located within the target volume region. Each discrete voxel cavity can correspond to a unit to be embedded in a binary matrix. The lateral position of the discrete voxel cavity can be determined by the row and column positions in the binary matrix, and the vertical position of the discrete voxel cavity can be determined by a random layering algorithm or a preset layering rule.

[0095] To make the authentication code difficult to fully recover from a single viewpoint or a single slice direction, the system distributes discrete voxel cavities to different embedding layers along the thickness direction of the target volume region. Different embedding layers can have the same interlayer spacing or non-uniform interlayer spacing. The system forms a layered authentication code within the target volume region through a geometric cut operation. This geometric cut operation can be implemented using Boolean cutting, where the corresponding discrete voxel cavity is used as the tool body to cut away from the target volume region of the original model, thereby forming a layered, discrete authentication code structure within the model that cannot be directly observed from the outside.

[0096] At least one of the following parameters of the layered authentication code—number of layers, interlayer spacing, unit size, embedding depth, and embedding direction—can be adjustable. The system can determine the range of these parameters based on the minimum manufacturable feature size of the additive manufacturing equipment, material type, printing accuracy, model strength requirements, and target volume region thickness. For example, for high-precision metal additive manufacturing equipment, the system can set smaller unit sizes and interlayer spacing; for ordinary fused deposition modeling equipment, the system can set larger unit sizes and interlayer spacing to ensure the manufacturability of the layered authentication code.

[0097] In one implementation, when the authentication code image is a QR code image, the system performs gridding processing on the QR code image. If the number of grids, grid_n, is not pre-specified, the system can automatically estimate the number of grids from multiple candidate modules. The number of candidate modules can be determined according to the rules for the number of QR code modules, for example, n = 2^1 + 4k, where k is a non-negative integer. The system scores the number of candidate modules and selects the module with the higher score as the gridding result. For example, the selection can be performed according to the following scoring formula: Among them, FinderAvg represents the matching score of the QR code positioning graphic region, which is used to characterize the degree of consistency between the positioning graphic and the standard QR code positioning graphic under the number of candidate modules; Separator represents the integrity score of the separation area around the positioning graphic, which is used to characterize whether the white separation area around the positioning graphic meets the preset separation requirements; Timing represents the consistency score of the timing graphic, which is used to characterize the degree of consistency between the alternating distribution of black and white modules in the QR code timing line and the timing structure of the standard QR code.

[0098] The system converts the meshed authentication code image into a binary matrix and identifies the target color modules within the binary matrix as the set of units to be embedded. Each unit to be embedded corresponds to a discrete voxel cavity. The planar position of each discrete voxel cavity within the target volume region is determined by the row and column positions of the corresponding module in the binary matrix, while the vertical position is determined by a random layering algorithm or a preset layering rule. The system then uses a Boolean cut operation to remove the discrete voxel cavities from the target volume region, thus splitting the authentication information across multiple thickness layers.

[0099] For example, in order to achieve the layered distribution of the internal cavity, the following operations are performed during the three-dimensional allocation stage: each black module corresponds to only one discrete voxel; the planar position information X and Y are determined by the QR code matrix; the height position information Z is determined by the random layering algorithm; and finally, these voxels are cut out from the interior of the solid body by Boolean cut-off.

[0100] In one embodiment, the set of units to be embedded corresponding to the authentication code image can be divided into multiple discrete sub-units and allocated to different depth positions within the target volume region according to a random layering algorithm, a pseudo-random layering algorithm, or a preset layering rule. The projections of each discrete sub-unit in the first viewing direction, the first cutting direction, or the first slice direction can combine to form a recognizable authentication pattern; while the projections in the second viewing direction, the second cutting direction, or the second slice direction present a discrete and irregular distribution, thereby increasing the difficulty for the authentication code to be directly recognized or completely reconstructed.

[0101] In one embodiment, the discrete voxel cavity can exhibit different material states under different additive manufacturing processes. For example, in fused deposition modeling (FDM) processes, the discrete voxel cavity can be filled with supporting material or left as an empty area; in powder bed fusion (PBF) processes, the discrete voxel cavity can correspond to an unsintered powder area; in photopolymerization (PBL) processes, the discrete voxel cavity can correspond to an uncured or removable material area. The system can adjust the size, interlayer spacing, and embedding depth of the discrete voxel cavity according to the additive manufacturing process type, material type, and post-processing method.

[0102] By embedding hierarchical authentication codes, the system can form a traceable authentication structure within the entity without directly exposing it to the model surface. This authentication structure can be identified under transparent viewing, cross-sectional viewing, slicing in a specific direction, or specialized detection conditions, while ordinary copyists cannot fully obtain its encoded information through conventional visual observation.

[0103] In this way, the authentication information is split into discrete voxel cavities distributed in different embedding layers, making it difficult to fully recover through a single visual perspective, a single cutting direction, or conventional model viewing methods. At the same time, the system can adjust the number of layers, the layer spacing, and the unit size according to the thickness of the target volume region, the bounding box size, and the manufacturing precision, so as to balance the concealment of authentication information and the manufacturability of the model.

[0104] Using the three types of mapping rules described above, the system can generate candidate safety embedding schemes for different safety feature types based on the curvature attributes, closed contour attributes, and solid thickness attributes of different geometric features. Each candidate safety embedding scheme can include a safety feature type, a suggested embedding location, a suggested parameter range, visibility parameters, and manufacturing impact parameters. If multiple confirmed geometric features exist in the same 3D model, the system can generate safety feature embedding sub-schemes for different geometric features, thereby forming a combined safety embedding scheme.

[0105] S105. Based on the preset selection strategy and / or the user selection result, determine the selected security embedding scheme from the candidate security embedding schemes.

[0106] In one implementation, the system can automatically determine the selected security embedding scheme from candidate security embedding schemes based on a preset selection strategy. The preset selection strategy can comprehensively rank schemes based on security level, manufacturing impact parameters, visibility parameters, host region size, model functional area location, and printing equipment capabilities. For example, the system can prioritize candidate security embedding schemes with lower manufacturing impact, lower visual visibility, higher security strength, and greater distance from functional areas.

[0107] In another implementation, the system can display candidate security embedding schemes to the user through a visual interactive interface. The user can perform operations such as accepting, rejecting, adjusting parameters, reordering, or adding comments for each candidate security embedding scheme. For example, the user can accept a spline curve segmentation scheme for a certain curvature region while rejecting a layered authentication code scheme for a region with insufficient thickness; or the user can adjust the embedding depth, interlayer spacing, and cell size of the layered authentication code.

[0108] The system generates a selected security embedding scheme based on a preset selection strategy and / or user selection results. The selected security embedding scheme can be saved as a structured scheme file, database record, configuration table, or other machine-readable data structure, serving as input for subsequent geometric embedding and export. The selected security embedding scheme may include a model identifier, a selected security feature type, a host geometric feature identifier, an embedding location, an embedding direction, embedding parameters, manufacturing impact parameters, and a traceable log identifier.

[0109] This step allows the system to retain room for manual confirmation and parameter fine-tuning on top of automatic recommendations, making the safety feature embedding scheme both automated and efficient, while also meeting the personalized needs of specific models, users, or manufacturing conditions.

[0110] In this step, a visual interaction and user confirmation module is used to display the candidate geometric feature set, candidate safe embedding scheme, recommended parameter range, and manufacturing impact prompts in the 3D preview interface, and to receive user input for acceptance, rejection, category correction, parameter adjustment, or reason annotation for the candidate geometric feature set and / or candidate safe embedding scheme.

[0111] In a user-oriented interactive workflow, the user first imports the 3D model file to be protected. The system generates geometric baseline data and multi-view rendered images, and identifies the geometric features in the 3D model. Subsequently, the system displays candidate geometric features, recognition confidence levels, and evidence views in the 3D preview interface. The user retains, deletes, classifies, or annotates candidate geometric features, forming the first feedback loop. Based on the confirmed set of geometric features, the system generates candidate security embedding schemes and displays security feature types, recommended parameter ranges, manufacturing impact prompts, and visibility prompts to the user. The user accepts, rejects, reorders, or fine-tunes parameters of candidate security embedding schemes, forming the second feedback loop. Based on the selected security embedding scheme confirmed by the user, the system generates an encrypted 3D model file, which can be further previewed in slices or verified in manufacturing. When the slice preview or manufacturing verification results show path changes, defect visibility, support structure changes, or parameter adjustments, the system adjusts the security feature parameter boundaries based on the results, forming the third feedback loop. Through the above interactive workflow, the system can achieve closed-loop processing of model import, geometric feature recognition, user verification, security feature recommendation, user confirmation, encrypted model generation, slice verification, and feedback updates.

[0112] S106. Based on the selected security embedding scheme, perform at least one geometric embedding operation among topological segmentation, geometric cutting, and geometric fusion on the 3D model file to generate an encrypted 3D model file with embedded security features.

[0113] In one implementation, the system parses the selected security embedding scheme and converts the security feature type, host geometry region, and embedding parameters defined therein into geometric operation instructions. These geometric operation instructions may include topology segmentation instructions, Boolean cutting instructions, Boolean fusion instructions, or other instructions capable of altering the geometry or topology of the 3D model.

[0114] Specifically, after obtaining a selected secure embedding scheme, the system parses the selected secure embedding scheme into one or more tool bodies or equivalent geometric operations. Based on the target geometric reference, embedding position, embedding direction, embedding size, position offset, and secure feature type in the selected secure embedding scheme, the system determines the embedding coordinates, normal direction, and scope of action on the target entity, target surface, or target volume region. It then calls the geometric kernel to perform topological segmentation, Boolean cutting, Boolean fusion, or equivalent geometric operations to form corresponding secure segmentation interfaces, closed contour segmentation structures, or internal embedding structures in the original 3D model. The geometric kernel operations may include BRepFeat_SplitShape, BRepAlgoAPI_Cut, BRepAlgoAPI_Fuse, or their equivalents.

[0115] For spline curve segmentation of security features, the system generates 3D surface segmentation lines on the target host surface based on the curve parameters in the selected security embedding scheme, and divides the target host surface into multiple topological sub-surfaces through topological segmentation. This process can change only the topological partitioning relationship of the target host surface without significantly changing the outer surface thickness. Therefore, the model appearance remains a continuous solid surface, but its topological structure already embeds security identifiers corresponding to the spline curve parameters.

[0116] For closed contour segmentation safety features, the system generates curvature closed segmentation interfaces or polygonal closed segmentation interfaces in the target area based on the closed contour type and offset parameters, and performs topological segmentation or geometric cutting operations on the target area. If the safety feature is a superimposed structure of multiple closed contours, the system can generate multiple closed segmentation interfaces sequentially and arrange them within the target area at set intervals.

[0117] For the layered authentication code security feature, the system generates multiple discrete voxel cavity tool bodies based on the authentication code matrix and layering parameters, and arranges these discrete voxel cavity tool bodies inside the target volume region. Subsequently, the system performs a Boolean cut operation on the original 3D model to remove the discrete voxel cavity tool bodies from the target volume region, thereby forming an internal cavity-type layered authentication code structure.

[0118] In other implementations, when a safety feature needs to form a raised, filled, or combined structure with the original model, the system can also use a geometric fusion operation to fuse the tool body corresponding to the safety feature into the original three-dimensional model.

[0119] After geometric embedding is complete, the system exports an encrypted 3D model file with embedded security features. This encrypted 3D model file can be in STEP format or converted to STL or other additive manufacturing-recognizable formats according to subsequent manufacturing requirements. The system can also simultaneously generate a solution summary and an operation log. The solution summary may include a list of embedded security features, corresponding geometric feature identifiers, embedding parameters, location indexes, and the export file path. The operation log can record model identifiers, processing time, recognition results, user selection results, embedding operation type, and output file identifiers for subsequent traceability and feedback learning.

[0120] Through this step, this embodiment does not merely generate abstract security policies or recommendations, but actually writes security features into the geometry or topology of the 3D model, thereby forming a protected model file that can be used for additive manufacturing preprocessing and security verification.

[0121] In one implementation, after generating an encrypted 3D model file, the system can convert the encrypted 3D model file into an additive manufacturing-recognizable slice preview file, or prompt the user to import the encrypted 3D model file into slicing software for preview. The system obtains the slice preview results, which include at least one of the following: whether safety features can be identified; whether the forming path of the area corresponding to the safety features has changed; whether support structures have been added; whether local paths have splitted or broken; whether the model material path has changed; whether the encrypted 3D model file meets authorized manufacturing requirements; whether parameters need to be adjusted; and explanations of the reasons. The slice preview results can serve as manufacturing verification samples in subsequent feedback learning steps.

[0122] S107. Obtain feedback samples for candidate geometric feature sets, candidate secure embedding schemes, and / or encrypted 3D model files, and adjust the recognition threshold, geometric consistency rule weights, mapping rule priorities, and / or secure feature parameter boundaries based on the feedback samples.

[0123] The feedback samples include at least one of the following: identification and correction samples, scheme acceptance samples, and manufacturing verification samples.

[0124] Error correction samples are derived from user reviews of the candidate geometric feature set. For example, deleting a misidentified feature, correcting the type of a candidate geometric feature, adding missed features, or providing reasons for removal can all generate error correction samples. The system adjusts the identification threshold and / or geometric consistency rule weights based on these error correction samples. For instance, if a certain type of geometric feature is repeatedly marked as misidentified by the user, the system can increase the identification confidence threshold for that type of geometric feature or increase the geometric consistency check weight corresponding to that type of geometric feature; if a certain type of geometric feature is repeatedly added as a missed feature by the user, the system can decrease the identification threshold for that type of geometric feature or add corresponding identification prompt samples.

[0125] The scheme acceptance sample is derived from users' selection results of candidate secure embedding schemes. For example, users accepting, rejecting, adjusting parameters, or reordering a secure embedding scheme can all form a scheme acceptance sample. The system adjusts the priority of mapping rules and / or the default parameters of security features based on the scheme acceptance sample. For example, if a spline curve segmentation scheme is repeatedly accepted by users in a certain type of curvature region with small parameter adjustments, the system can increase the priority of this mapping rule in similar contexts and move the default parameters closer to the range of parameters repeatedly accepted by users; if a hierarchical authentication code scheme is repeatedly rejected by users in a specific size region, the system can reduce the recommendation priority of this scheme in similar regions.

[0126] The manufacturing verification sample originates from the preview, slicing, or manufacturing verification results of an encrypted 3D model file. In one implementation, the system converts the encrypted 3D model file into a slice preview file and obtains the slice preview results. The slice preview results include at least one of the following: whether the preview passed, whether the path is split, whether the path is broken, whether the material path has changed, whether the support material area has changed, whether the internal cavity is displayed, whether defects are shown, whether the support structure has increased, whether parameters need to be adjusted, and an explanation of the reason. If the slice preview results show that a certain safety feature causes a significant increase in the support structure, makes a part of the model unmanufacturable, or affects a functional area, the system adjusts the safety feature parameter boundaries based on the manufacturing verification sample. The safety feature parameter boundaries may include upper and lower limits of at least one of the following: embedding size, position offset, number of layers, interlayer spacing, embedding depth, and number of segments.

[0127] For example, when the unit size of a certain layered authentication code is too small, causing the slicing software to be unable to stably identify the internal cavity, the system can increase the lower limit of the minimum unit size under the corresponding additive manufacturing equipment; when a certain slice curve is too close to the functional boundary of the model, resulting in an increase in local manufacturing risk, the system can increase the lower limit of the position offset of this type of safety feature; when the number of superpositions of a certain closed contour segment is too large, resulting in a significant increase in the local support structure of the model, the system can reduce the default number of superpositions in the corresponding context.

[0128] As a specific implementation of feedback learning, the system may employ a policy update mechanism to process the identification and correction samples, which includes the following comprehensive feature representations:

[0129] in, This represents the structural feature vector obtained from geometric analysis, whose dimensions may include abstract features such as aperture, aspect ratio, curvature value, number of cylindrical surfaces, and topological adjacency relationships. This represents the output confidence score or candidate category score of the visual language model for candidate features; y represents the score based on geometric consistency rules, such as the verification result of "whether the geometric kernel (OCC) analysis supports the geometric category" or "whether the local topology satisfies the feature definition"; y represents the user feedback label, where deleting a recognition result by the user is recorded as negative feedback, and adding or correcting it to the correct category by the user is recorded as positive feedback.

[0130] Based on this input, the policy update mechanism will output four types of interpretable update quantities to systematically improve the entire recognition process: 1) Recognition prompt word template update, used to optimize the guidance of the visual language model; 2) Recognition threshold update, used to adjust the difficulty of recognizing a certain type of geometric feature; 3) Geometric consistency rule weight update, used to adjust the importance of different evidence items in the final decision; 4) Error correction sample library update.

[0131] Furthermore, to quantitatively assess the reliability of a candidate geometric item belonging to a certain category k, the system can define a comprehensive recognition score Sk, which fuses multi-source evidence through a weighted summation:

[0132] in, This represents the confidence level output by the visual language model recognizer that the candidate geometric entry belongs to category k; This indicates the support level of the candidate geometric entry, based on the results of geometric kernel analysis and rule recognition, for conforming to the geometric definition of category k. The geometric consistency rule score represents the degree of consistency between geometric kernel analysis results, local topological relationships, and category definitions. This represents the weighting coefficient for the corresponding evidence item.

[0133] The system sets a corresponding recognition threshold for each type of geometric feature k. When Sk ≥ When Sk < 1, the system identifies the candidate geometric entry as a geometric feature of category k; when Sk < 1, Sk < 1. If this happens, the system will mark the candidate geometric entry as an unconfirmed feature or exclude it from the set of confirmed geometric features.

[0134] When a user deletes a candidate geometric item that the system identifies as category k, the system treats this feedback as a misidentification sample and raises the recognition threshold corresponding to category k. And / or reduce the weight of the evidence item that contributed to the misidentification; when a user supplements or corrects a candidate geometric item to category y, the system treats this feedback as a missed identification or category correction sample, and lowers the identification threshold corresponding to category y. And / or increase the weight of evidence items that support the correct identification.

[0135] In one implementation, the identification threshold and evidence item weights can be iteratively updated as follows:

[0136]

[0137] in, and To preset the update step size, and The direction and magnitude are determined by the type of feedback sample; when the feedback sample is a misidentified sample... It is a positive value; when the feedback sample is a missed sample, The value is negative. Through this method, the system can adjust the recognition threshold and geometric consistency rule weights based on the user's review results.

[0138] When the identification threshold is updated in N consecutive rounds The average change is below the threshold Furthermore, the average change in the weight of each piece of evidence is below the threshold. When the system reaches a certain point, it can consider the current recognition strategy to have reached a relatively stable state, and use the prompt word template, threshold, and weight in this state as the current benchmark for subsequent recognition. The number of updates... , and All parameters are configurable and can be set according to the number of samples, the number of feature categories, and the complexity of the recognition task. N is used to suppress the impact of single-feedback fluctuations on stability judgment. Preferably, A value of 3 to 10 can be selected to achieve a balance between response speed and noise immunity. Preferably, and The value can be set to 1% to 5% of the current value range of the corresponding parameter to adapt to different parameter scales and avoid premature or delayed determination of stability due to improper setting of absolute threshold.

[0139] In practice, the system collects user operations during the scheme selection and parameter fine-tuning stages, as well as feedback results from the optional slice preview and verification stages, into a unified sample of scheme acceptance and manufacturing feasibility, which is used to optimize the recommendation quality of the security embedding strategy.

[0140] As a specific embodiment, the system may employ the following update logic, the input of which is a comprehensive feature representation:

[0141] in: It represents contextual information, including host geometry type, thickness, bounding box size, curvature level, printer type, slicing software type, etc. This indicates the recommended security feature type, such as micro-dot matrix, letter matrix, spline segmentation, hierarchical authentication code, etc. A parameter vector representing this security feature; Indicates user feedback (acceptance, rejection, parameter modification, reordering); This indicates feedback on manufacturing feasibility (whether the slice is abnormal, whether the risk exceeds the limit, and whether the impact on forming is too great).

[0142] Based on this input, the update logic outputs the following four types of interpretable update quantities: mapping rule priority update; security feature recommendation score update; parameter default value update; parameter upper and lower bound update.

[0143] Regarding parameter boundary update rules: for security feature schemes that are accepted by users and have passed manufacturing verification, the system increases the recommendation weight of the corresponding parameter range; for parameter combinations that are rejected by users or fail manufacturing verification, the system reduces their recommendation weight and compresses or restricts the corresponding parameter range; for parameter values ​​that are accepted multiple times, the system can move the default parameter to its statistical center so that subsequent recommendations are closer to user preferences and manufacturing feasibility requirements.

[0144] Furthermore, to quantitatively evaluate the merits of different safety feature and parameter combinations under a given context, this system can employ a context-based strategy optimization algorithm. For the scheme acceptance sample and the manufacturing verification sample, the system can define a comprehensive reward value *r* to characterize the recommended performance of safety feature scheme *a* and its parameters *θ* under a given context *c*. The context *c* includes at least one of the following: host geometry feature type, host region thickness, bounding box size, curvature attribute, closed contour attribute, printing equipment type, slicing software type, and material type. The comprehensive reward value *r* can be determined as follows:

[0145] in, When a user accepts a reward, a positive value can be used if the user accepts it directly, a smaller positive value can be used if the user accepts it after making a modification, and a negative value can be used if the user rejects it. For manufacturing verification rewards, a positive value can be taken when the slice preview or manufacturing verification results meet the preset requirements, and a negative value can be taken when the risk exceeds the preset range, the part is unmanufacturable, or the quality of authorized manufacturing deteriorates. The smaller the change in recommendation parameters the user makes, the greater the reward. The higher. These are the weighting coefficients for each reward. In a specific example, the values ​​of each reward parameter can be set as follows: when the user directly accepts... = 1; When the user accepts the changes. = 0.5; when the user refuses = -1; When manufacturing verification passes = 1.

[0146] Based on accumulated feedback samples, the system adjusts the priority of mapping rules, default parameters of security features, and boundaries of security feature parameters with the goal of maximizing the comprehensive reward value r. For security feature schemes and their parameter combinations that repeatedly obtain high comprehensive reward values, the system increases the recommendation priority of the security feature scheme in similar contexts and moves the default parameters closer to the parameter range that has been accepted multiple times. For security feature schemes and their parameter combinations that repeatedly obtain low comprehensive reward values, the system decreases the recommendation priority of the security feature scheme in similar contexts and compresses or restricts the corresponding parameter boundaries.

[0147] Regarding the stability conditions of the strategy, when the absolute value of the average reward increment in N consecutive updates is lower than the threshold ε, or the change amplitude of the parameter boundary is lower than the threshold δ, the system can consider the current recommendation strategy to have reached a relatively stable state. Here, N, ε, and δ are all configurable parameters. N is used to window the observation of feedback in multiple consecutive rounds to reduce the impact of fluctuations in single user feedback or local anomalies on stability judgment. Preferably, N is 3-10 to achieve a balance between adaptive response speed and noise resistance. ε is set as a relative proportion of the width of the comprehensive reward interval, preferably 1% to 5% of the interval width, to adapt to different reward normalization scales and avoid premature or delayed determination of strategy stability due to improper absolute threshold settings. δ is set as a relative proportion of the current range of parameter changes, preferably 1% to 5% of the current range of corresponding parameter changes, to adapt to different parameter units and interval scales, and to reflect whether the parameter boundary has shrunk to a relatively stable feasible interval.

[0148] Through the above feedback learning process, the system can continuously adjust the identification threshold, geometric consistency rule weight, mapping rule priority, and safety feature parameter boundary based on actual identification and correction, scheme acceptance, and manufacturing verification results, so that the system can gradually adapt to different 3D models, different user preferences, different slicing software, and different additive manufacturing equipment.

[0149] Furthermore, as a specific implementation method of the manufacturing verification sample formation process, such as Figure 4 As shown, the manufacturing verification feedback sample is obtained through steps S201 to S208.

[0150] The system first converts the encrypted STEP model with embedded security features into STL or 3MF format, and then calls the local or cloud-based slicing engine via command-line interface or software development kit to load the target printer's process parameters for automatic slicing. After acquiring slicing preview data or slicing path data, the system focuses on determining whether the security features produce the expected path changes, defect visibility, support structure changes, or internal cavity identification results. If the security effect is not apparent, the embedding depth, number of segments, or interlayer spacing of the security features are adjusted; if the manufacturing impact is too significant (such as a sudden increase in support or local unmanufacturability), the position offset is increased or the number of stacked segments is reduced. The system records the judgment results and corresponding parameter combinations as manufacturing verification feedback samples, and updates the parameter boundaries, risk constraints, and mapping rule priorities in the security feature knowledge base accordingly, achieving feedback feedback.

[0151] Specifically, after generating the slice preview results, the system or user determines whether the safety feature meets the target based on the preview results. If the safety feature produces the expected path change under a specific slice direction or slice parameters, and does not significantly impair the manufacturability of authorized manufacturing, the corresponding solution is marked as preview passed, and the safety feature type, embedding parameters, printing direction, and slice parameter combination are recorded. If the safety feature is not displayed, the path does not change significantly, the internal cavity is not identified, or the defect visibility effect is insufficient, it is determined that the current safety feature size, depth, number of segments, interlayer spacing, embedding direction, or embedding position may be insufficient, and feedback information for adjusting the safety feature parameters is generated. If the support structure is significantly increased, the local path is excessively broken, the model's functional areas are affected, the local area is unmanufacturable, or the quality of authorized manufacturing is significantly affected, it is determined that the current safety feature parameters are too strong or the embedding position is unreasonable, and feedback information for reducing the embedding size, increasing the position offset, reducing the number of segments, adjusting the embedding direction, or changing the safety feature type is generated.

[0152] The system writes the above slice preview results into the feedback log, forming a manufacturing verification feedback sample. Based on the manufacturing verification feedback sample, the system updates the safety feature parameter boundaries, risk constraints, and mapping rule priorities, and sends the update results back to the safety feature knowledge base, candidate solution generation module, and feedback learning module.

[0153] It should be noted that the feedback learning in this embodiment is not limited to a specific machine learning algorithm. The feedback learning module can be implemented through rule updates, statistical updates, contextual strategy optimization, or manual maintenance of the knowledge base. The system can adjust the recognition prompts, recognition thresholds, and geometric consistency rule weights based on user-deleted, corrected, or supplemented geometric features; it can adjust the mapping rule priorities and default parameters based on user acceptance, rejection, or parameter modification results of candidate secure embedding schemes; and it can adjust the boundaries of secure feature parameters based on slicing previews or manufacturing verification results, thereby forming a dynamic protection mechanism that can be updated according to changes in model type, user preferences, slicing software, and printing equipment.

[0154] Corresponding to the above-described additive manufacturing model safety structure formation method, each module in the aforementioned additive manufacturing model safety structure formation system can respectively execute the corresponding steps in the above method embodiments. Specifically, the model input and preprocessing module executes S101, the geometric feature recognition module executes S102, the recognition result verification module executes S103, the safety feature scheme generation module executes S104, the scheme determination module executes S105, the geometric embedding and derivation module executes S106, and the feedback learning module executes S107. For the specific processing procedures of each module, please refer to the corresponding descriptions in the above method embodiments, which will not be repeated here.

[0155] It should also be noted that the additive manufacturing model safety structure formation system in this embodiment can be a standalone software system, or a plug-in system integrated into CAD software, 3D modeling software, slicing software, or additive manufacturing management platform. It can also be deployed on a cloud server or a local computing device. The above modules can be implemented using software programs, hardware circuits, or a combination of both.

[0156] It should be noted that although several modules of the system have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to embodiments of the present invention, the features and functions of two or more modules described above can be embodied in one module. Conversely, the features and functions of one module described above can also be further divided and embodied by multiple modules.

[0157] In some alternative implementations, the geometric feature recognizer is not limited to a visual language model recognizer, but may also be a rule-based geometric recognizer, a deep learning-based 3D point cloud recognizer, a mesh feature-based recognizer, a B-Rep topology-based recognizer, or a combination of the above recognizers.

[0158] In some alternative implementations, the security feature knowledge base is not limited to being stored in the form of a JSON file, but can also be stored in the form of a relational database, graph database, knowledge graph, rule engine or configuration table.

[0159] In some alternative implementations, the security feature type is not limited to spline curve segmentation security features, closed contour segmentation security features, and hierarchical authentication code security features, but may also include micro-hole arrays, micro-groove arrays, hidden text cavities, internal lattice defect control structures, orientation-sensitive cavity structures, or other geometric security features that can be coupled with additive manufacturing parameters.

[0160] In some alternative implementations, the manufacturing verification samples are not limited to obtaining them through slice previews, but can also be obtained through finite element simulation, manufacturing path analysis, printing test feedback, forming quality inspection, or 3D scanning comparison.

[0161] In some alternative implementations, geometric embedding operations are not limited to topological segmentation, geometric cutting, and geometric fusion, but may also include local offset, local scaling, surface reconstruction, internal cavity generation, texture mapping, or other geometric processing methods that can form security features in a 3D model.

[0162] Through the above embodiments, the present invention can automatically identify the geometric features of various types of 3D models that meet preset reading conditions, generate a combined security embedding scheme based on a security feature knowledge base, and then write the security features into the 3D model through geometric kernel operations. Furthermore, the present invention can continuously adjust the identification strategy, mapping rules, and parameter boundaries based on user review, scheme acceptance, and manufacturing verification results, thereby improving the automation level, adaptability, and long-term protection effect of security feature embedding.

[0163] The additive manufacturing model safety structure formation method provided in this embodiment can be executed in smart terminals, computer terminals, server equipment, cloud computing platforms, CAD plug-in operating environments, additive manufacturing pre-processing systems, or similar computing devices.

[0164] Corresponding to the above-described method for forming a safe structure for an additive manufacturing model, this embodiment also provides a computer device. The computer device includes a processor and a memory, the memory storing a computer program, and the processor executing the computer program to implement the steps in the above-described method for forming a safe structure for an additive manufacturing model.

[0165] In one embodiment, the computer device may be a personal computer, workstation, server, cloud server, edge computing device, terminal device with CAD software installed, or additive manufacturing preprocessing equipment. The computer device may also include input / output interfaces and / or network communication interfaces to receive 3D model files, display candidate geometric features and candidate secure embedding schemes, receive user review results or parameter adjustment results, and output encrypted 3D model files, scheme summaries, operation logs, or feedback samples.

[0166] This embodiment also provides a computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the steps in the above-described additive manufacturing model safety structure formation method. The computer-readable storage medium may store at least one of the following: geometric baseline data, multi-view rendered images, recognition request data, candidate geometric feature sets, confirmed geometric feature sets, a safety feature knowledge base, candidate safety embedding schemes, selected safety embedding schemes, encrypted 3D model files, scheme summaries, operation logs, and feedback samples.

[0167] In a possible implementation, the present invention can also be implemented as a program product, which includes program code. When the program product is run on a terminal device, server device, CAD plugin runtime environment, or cloud model processing platform, the program code is used to cause the corresponding device to perform the steps in the above-described additive manufacturing model safety structure formation method.

Claims

1. A method for forming a safety structure of an additive manufacturing model, characterized in that, include: The three-dimensional model file to be protected is obtained, and the three-dimensional model file is read and preprocessed to obtain the geometric baseline data and multi-view rendering image corresponding to the three-dimensional model file. Based on the geometric baseline data and the multi-view rendered image, the geometric features in the 3D model file are identified to obtain a candidate geometric feature set; wherein, the candidate geometric feature set includes geometric feature type, geometric feature location information and identification confidence level; The candidate geometric feature set is verified to obtain the confirmed geometric feature set; wherein the confirmed geometric feature set is determined based on the automatic verification results and / or the user review results. Based on the confirmed set of geometric features and the preset security feature knowledge base, candidate security embedding schemes are generated; wherein, the security feature knowledge base includes mapping rules between geometric feature types and security feature types, and the mapping rules determine the corresponding security feature type based on at least one of the curvature attribute, closed contour attribute and solid thickness attribute of the geometric feature; Based on a preset selection strategy and / or user selection results, a selected security embedding scheme is determined from the candidate security embedding schemes; Based on the selected security embedding scheme, perform at least one geometric embedding operation among topological segmentation, geometric cutting, and geometric fusion on the 3D model file to generate an encrypted 3D model file with embedded security features; Obtain feedback samples for the candidate geometric feature set, the candidate secure embedding scheme, and / or the encrypted 3D model file, and adjust the recognition threshold, geometric consistency rule weight, mapping rule priority, and / or security feature parameter boundaries based on the feedback samples.

2. The method for forming a safety structure of an additive manufacturing model as described in claim 1, characterized in that, The process of acquiring the 3D model file to be protected, and reading and preprocessing the 3D model file to obtain the geometric baseline data and multi-view rendered images corresponding to the 3D model file includes: Read the boundary representation entities of the 3D model file; The number of topological elements in the boundary representation entity is counted. Extract the bounding box size information of the 3D model file; Geometric parameters are acquired for candidate geometric regions in the 3D model file; wherein, the geometric parameters include at least one of face type, edge type, radius, depth, axial direction, curvature information, and position coordinates; The geometric baseline data is generated based on the geometric parameters; The 3D model file is rendered according to a preset perspective to obtain the multi-view rendered image.

3. The method for forming a safety structure of an additive manufacturing model as described in claim 2, characterized in that, Based on the geometric baseline data and the multi-view rendered image, the geometric features in the 3D model file are identified to obtain a candidate geometric feature set, including: The geometric baseline data, the multi-view rendered image, the recognition prompt information, and the structured output specification are encapsulated into recognition request data; The identification request data is input into the geometric feature recognizer to obtain the structured identification result; The structured recognition results include geometric feature type, location description, evidence view, geometric reference identifier, and recognition confidence level. The geometric feature recognizer includes a rule recognizer, a visual language model recognizer, or a hybrid recognizer formed by combining a rule recognizer and a visual language model recognizer.

4. The method for forming a safety structure of an additive manufacturing model as described in claim 1, characterized in that, The step of verifying the candidate geometric feature set to obtain the confirmed geometric feature set includes: The recognition confidence level is constrained to a preset numerical range, and the candidate geometric feature set is filtered based on the confidence level threshold; Based on the number of evidence views corresponding to the multi-view rendered image, the evidence consistency verification is performed on the candidate geometric feature set. Based on the geometric baseline data, a geometric consistency check is performed on the candidate geometric feature set; Deduplication is performed on candidate geometric features that have the same geometric reference identifier or the same semantic signature; The verified candidate geometric features are provided to a visual interactive interface to receive verification labels, rejection reasons and / or correction categories for the candidate geometric features; Based on the results of the verification process and / or the review results received by the visual interactive interface, the confirmed geometric feature set is generated.

5. The method for forming a safety structure for an additive manufacturing model as described in claim 1, characterized in that, The step of generating candidate security embedding schemes based on the confirmed geometric feature set and the preset security feature knowledge base includes: Read the available security feature types, applicable geometric feature types, embedding prerequisites, recommended parameter ranges, and manufacturing impact parameters from the security feature knowledge base; Match the geometric feature types in the confirmed geometric feature set with the applicable geometric feature types in the security feature knowledge base; If a match is successful, a candidate secure embedding scheme is generated based on the corresponding mapping rules; The candidate security embedding scheme includes security feature type, suggested embedding location, suggested parameter range, visibility parameter, and manufacturing impact parameter.

6. The method for forming a safety structure for an additive manufacturing model as described in claim 5, characterized in that, The mapping rules include spline curve segmentation rules corresponding to curvature features; The spline curve segmentation rules include: When a target geometric feature with curvature attribute exists in the confirmed set of geometric features, the local parameter domain of the target host surface corresponding to the target geometric feature is extracted; Generate one or more two-dimensional spline curves within the local parameter domain; The two-dimensional spline curve is mapped onto the target host surface to obtain a three-dimensional surface segmentation line; Based on the three-dimensional curved surface segmentation lines, a topology segmentation operation is performed on the target host surface to form a secure segmentation interface on the target host surface; The appearance of the safety segmentation interface remains a continuous solid surface, and at least one of the segmentation position, segmentation length, curvature control amount, and number of segmentation strips of the safety segmentation interface is an adjustable parameter.

7. The method for forming a safety structure of an additive manufacturing model as described in claim 5, characterized in that, The mapping rules include closed contour segmentation rules; The closed contour segmentation rules include: When a target geometric feature with a closed contour attribute exists in the confirmed set of geometric features, the contour type of the closed contour corresponding to the target geometric feature is identified; Based on the contour type of the closed contour, a curvature closed segmentation interface or a polygonal closed segmentation interface is generated. Wherein, when the closed contour is a circular contour or an elliptical contour, the curvature closed segmentation interface is generated based on the circular contour or the elliptical contour; When the closed contour is formed by straight line boundaries, multiple boundaries that are connected end to end are selected from the straight line boundaries, and a closed polygonal contour is generated based on the multiple boundaries to form the polygonal closed segmentation interface; At least one of the following parameters is adjustable: the outline size, position offset, distance from the original geometric feature boundary, and number of superpositions of the curvature closed segmentation interface or the polygonal closed segmentation interface.

8. The method for forming a safety structure of an additive manufacturing model as described in claim 5, characterized in that, The mapping rules include hierarchical authentication code embedding rules; The hierarchical authentication code embedding rules include: When there is a target volume region in the confirmed geometric feature set whose entity thickness is greater than a preset thickness threshold and whose bounding box size meets the preset embedding conditions, obtain the authentication code image or authentication identifier matrix. Convert the authentication code image or the authentication identifier matrix into a binary matrix; The set of units to be embedded is determined based on the binary matrix; Each unit to be embedded in the set of units to be embedded is mapped to a discrete voxel cavity located inside the target volume region; Along the thickness direction of the target volume region, the discrete voxel cavity is distributed to different embedding layers; A layered authentication code is formed within the target volume region through geometric excision operations; Among them, at least one of the following parameters of the layered authentication code—number of layers, layer spacing, unit size, embedding depth, and embedding direction—is adjustable.

9. The method for forming a safety structure for an additive manufacturing model as described in claim 1, characterized in that, The step of obtaining feedback samples for the candidate geometric feature set, the candidate secure embedding scheme, and / or the encrypted 3D model file, and adjusting the recognition threshold, geometric consistency rule weight, mapping rule priority, and / or security feature parameter boundaries based on the feedback samples, includes: Obtain at least one of the following: identification and correction samples, scheme acceptance samples, and manufacturing verification samples; If the feedback samples include identification and error correction samples, the identification threshold and / or the geometric consistency rule weights are adjusted based on the identification and error correction samples. If the feedback sample includes a scheme acceptance sample, the priority of the mapping rule and / or the default parameters of the security feature are adjusted based on the scheme acceptance sample. When the feedback sample includes a manufacturing verification sample, the safety feature parameter boundary is adjusted based on the manufacturing verification sample. The manufacturing verification sample is obtained by converting the encrypted 3D model file into a slice preview file and obtaining the slice preview result. The slice preview result includes at least one of the following: whether the preview passed, whether the defects are displayed, whether the support structure has been added, whether the parameters need to be adjusted, and an explanation of the reasons. The security feature parameter boundaries include upper and lower limits of at least one of the following: embedding size, position offset, number of layers, interlayer spacing, embedding depth, and number of segments.

10. A protection system for the additive manufacturing model safety structure formation method according to any one of claims 1-9, characterized in that, include: The model input and preprocessing module is used to acquire the 3D model file to be protected, and to read and preprocess the 3D model file to obtain the geometric baseline data and multi-view rendering image corresponding to the 3D model file. A geometric feature recognition module is used to recognize geometric features in the 3D model file based on the geometric baseline data and the multi-view rendered image to obtain a candidate geometric feature set; wherein, the candidate geometric feature set includes geometric feature type, geometric feature location information and recognition confidence level; The recognition result verification module is used to verify the candidate geometric feature set to obtain the confirmed geometric feature set; wherein, the confirmed geometric feature set is determined based on the automatic verification result and / or the user review result; A security feature scheme generation module is used to generate candidate security embedding schemes based on the confirmed set of geometric features and a preset security feature knowledge base; wherein, the security feature knowledge base includes mapping rules between geometric feature types and security feature types, and the mapping rules determine the corresponding security feature type based on at least one of the curvature attribute, closed contour attribute and solid thickness attribute of the geometric feature; The visualization interaction and user confirmation module is used to display the candidate geometric feature set, candidate security embedding scheme, recommended parameter range and manufacturing impact prompts in the 3D preview interface, and to receive user input for acceptance, rejection, category correction, parameter adjustment or reason annotation for the candidate geometric feature set and / or the candidate security embedding scheme; The scheme determination module is used to determine the selected security embedding scheme from the candidate security embedding schemes based on a preset selection strategy and / or the user selection result received by the visualization interaction and user confirmation module. The geometry embedding and export module is used to perform at least one geometry embedding operation among topological segmentation, geometric cutting and geometric fusion on the 3D model file according to the selected security embedding scheme, to generate an encrypted 3D model file with embedded security features; The feedback learning module is used to obtain feedback samples for the candidate geometric feature set, the candidate secure embedding scheme and / or the encrypted 3D model file, and adjust the recognition threshold, geometric consistency rule weight, mapping rule priority and / or security feature parameter boundary based on the feedback samples.

Images