AI-driven metal 3D printing support type automatic identification method and system

Abstract

The application provides an AI-driven metal 3D printing support type automatic identification method and system, belonging to the field of additive manufacturing and intelligent control, comprising: obtaining the part type and material type of a metal 3D printing part as printing information; based on the printing information, matching and obtaining a common support type set in a support type database; extracting support shape feature parameters in metal 3D printing and performing support type identification to obtain a support type identification result; combining the support type database to obtain the identification confidence of the support type identification result; based on the identification confidence, performing discrimination, and when the identification confidence is greater than or equal to a confidence threshold, performing output; when the identified confidence is less than or equal to the confidence threshold, optimizing the extraction parameters of the support shape feature parameters, and iteratively performing identification. The application solves the problem that the prior art cannot meet the production requirements of efficient, accurate and comprehensive identification of metal 3D printing complex support structure types.

Description

Technical Field

[0001] This invention relates to the field of additive manufacturing and intelligent control, specifically to an AI-driven method and system for automatic identification of support types in metal 3D printing. Background Technology

[0002] Metal additive manufacturing (3D printing) technology has significant advantages in the integrated molding of complex structural parts. In metal forming processes such as laser powder bed melting or directional energy deposition, auxiliary support structures must be added to the overhanging areas of the parts or the bottom of large spans to prevent the collapse of suspended structures and to counteract warping deformation caused by thermal stress.

[0003] In the existing production process, after the parts are printed, the first step is the removal of the support structure. Because metal parts have high mechanical performance requirements, and the support structure and the solid part are made of the same metal, the choice of support removal process—such as wire cutting, CNC milling, manual grinding, or chemical dissolution—is closely related to the distribution, geometry, and density of the support structure. Therefore, before entering the post-processing workshop, operators need to identify and classify the support types of the printed blanks on the worktable to plan the subsequent removal path and estimate processing time.

[0004] In existing technologies, the identification and positioning of support structures on metal 3D printed parts are mainly achieved through manual visual inspection and experience-based judgment. This involves operators holding drawings or portable terminals and visually observing the support distribution on the surface of the blank part against a 3D model on a screen. Another method is two-dimensional vision-assisted positioning, which involves setting up an industrial camera to take photos of the part's outer surface and using traditional image edge detection algorithms to mark the boundary contours between the supports and the part's surface.

[0005] However, existing support analysis methods face the following problems in actual production: Metal support structures are diverse in type and shape; when faced with complex or numerous support groups, manual identification is not only time-consuming but also prone to omissions or misjudgments due to visual fatigue. If thin-walled sheet supports are mistakenly identified as solid supports and subjected to heavy milling, the component substrate is easily over-cut and scrapped. Two-dimensional vision methods are limited by a single perspective and struggle to obtain complete contour information of the support in three-dimensional space, especially for concealed supports in internal flow channels and interlocking support groups.

[0006] In summary, existing technologies are insufficient to meet the production requirements of efficiently, accurately, and comprehensively identifying complex support structure types in metal 3D printing. There is an urgent need for a support perception and analysis method that can adapt to complex three-dimensional structures and has automated classification capabilities. Summary of the Invention

[0007] This application provides an AI-driven method and system for automatic identification of support types in metal 3D printing, which solves the problem that existing technologies are unable to meet the production needs of efficiently, accurately, and comprehensively identifying complex support structure types in metal 3D printing.

[0008] In view of the above problems, this application provides an AI-driven method and system for automatic identification of support types in metal 3D printing.

[0009] Firstly, this application provides an AI-driven method for automatic identification of support types in metal 3D printing, including:

[0010] Obtain the part type and material type of the metal 3D printed part as printing information;

[0011] Based on the printed information, a set of commonly used support types is obtained by matching and retrieving the support type database;

[0012] Extract the support shape feature parameters from metal 3D printing, and identify the support type based on the support shape feature parameters to obtain the support type identification result;

[0013] Based on the support type database, the recognition credibility of the support type recognition result is obtained;

[0014] Based on the recognition credibility, a judgment is made, and when the recognition credibility is greater than or equal to the credibility threshold, the support type recognition result is output;

[0015] When the recognition confidence is less than or equal to the confidence threshold, the extraction parameters of the support shape feature parameters are optimized, and the recognition is iterated until a support type recognition result with a recognition confidence greater than or equal to the confidence threshold is obtained.

[0016] Secondly, this application provides an AI-driven automatic recognition system for metal 3D printing support types, the system comprising:

[0017] The part information acquisition module is used to acquire the part type and material type of the metal 3D printed part as printing information;

[0018] The support type data management module is used to match and obtain a set of commonly used support types from the support type database based on the printed information.

[0019] The feature extraction and recognition reasoning module is used to extract the support shape feature parameters in metal 3D printing, and to perform support type recognition based on the support shape feature parameters to obtain the support type recognition result;

[0020] The credibility assessment and iteration control module is used to combine the support type database to obtain the recognition credibility of the support type recognition result. When the recognition credibility is less than or equal to the credibility threshold, the extraction parameters of the support shape feature parameters are optimized and the recognition is iterated until a support type recognition result with recognition credibility greater than or equal to the credibility threshold is obtained.

[0021] The result output and interface module is used to make a judgment based on the recognition credibility. When the recognition credibility is greater than or equal to the credibility threshold, the support type recognition result is output.

[0022] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0023] The technical solution of this application firstly reduces the search range of the recognition algorithm by pre-reading the type and material information of the parts and searching the support database. This not only improves the recognition response speed, but also effectively avoids fundamental misjudgments caused by the mismatch between material properties and support shape.

[0024] Furthermore, structured recognition of shape feature parameters can transform ambiguous geometric shapes into quantitative classification results, turning subjective "visual judgment" that relies on human experience into objective, repeatable, automated numerical judgment.

[0025] Furthermore, recognition confidence is introduced as an intermediate monitoring variable, and a threshold-triggered iteration mechanism is set. For difficult support areas with blurred features and severe powder adhesion, erroneous results are not forcibly output. Instead, secondary refined recognition is performed by dynamically optimizing feature extraction parameters, such as adjusting filtering and increasing sampling density, which significantly improves the robustness of recognition under complex working conditions. By directly passing simple features and deeply calculating complex features, the real-time performance and high efficiency of the overall process are taken into account while ensuring that the final output results are all of high confidence.

[0026] Furthermore, the output of the support's position coordinates and the verified high-confidence type label provides accurate process decision-making basis for downstream automated support removal equipment, enabling post-processing to call the corresponding removal strategies and cutting parameters for block, columnar, and grid-like supports respectively.

[0027] In summary, the technical solution of this application narrows the identification range through prior constraints of part information. By combining a dual mechanism of feature extraction and credibility iteration verification, it ensures high accuracy and robustness in the identification of metal 3D printing support types while achieving strong adaptability of the identification process to complex blank surface conditions. Ultimately, it provides highly reliable structured data support that can directly drive machine path planning for subsequent automated support removal processes, further solving the problem that existing technologies cannot meet the production needs of efficient, accurate, and comprehensive identification of complex support structure types in metal 3D printing. Attached Figure Description

[0028] Figure 1 This is a flowchart illustrating the AI-driven automatic identification method for metal 3D printing support types provided in this application embodiment.

[0029] Figure 2 This is a schematic diagram of the structure of the AI-driven automatic identification system for metal 3D printing support type provided in the embodiments of this application.

[0030] In the attached diagram, the component designations are as follows:

[0031] The module includes: 11 for obtaining part information, 12 for supporting type data management, 13 for feature extraction and recognition reasoning, 14 for credibility assessment and iteration control, and 15 for result output and interface. Detailed Implementation

[0032] This application provides an AI-driven method and system for automatic identification of support types in metal 3D printing, which solves the problem that existing technologies are unable to meet the production needs of efficiently, accurately, and comprehensively identifying complex support structure types in metal 3D printing.

[0033] It should be noted that the terms "comprising" and "having" are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to these processes, methods, products, or devices.

[0034] Example 1, as Figure 1 As shown in the embodiments of this application, an AI-driven method for automatic identification of metal 3D printing support types is provided, the method comprising:

[0035] S100: Obtain the part type and material type of the metal 3D printed part as printing information.

[0036] This step involves obtaining the geometric category and material grade of the part beforehand and establishing identification constraint boundaries using prior process knowledge. Compared to existing technologies that directly perform blind searches of the entire feature library or pure visual comparisons on unknown blanks, this step introduces metadata from the manufacturing end, narrowing the subsequent identification scope from the entire set of support forms to a subset of commonly used supports under specific material-structure combinations. This effectively suppresses cross-category misjudgments caused by differences in material thermal properties, such as the different requirements for support forms between aluminum alloys and stainless steel. Furthermore, the compression of the search space fundamentally improves the execution efficiency and accuracy of the identification algorithm.

[0037] Step S100 provided in this application embodiment includes:

[0038] The part type and material type of the metal 3D printed parts are collected and integrated to form printing information. The part type is defined as the structural category of the part to be formed in the metal 3D printing, and the material type is defined as the metal material category used in the metal 3D printing.

[0039] In this step, the first step is to collect the part type and material type of the metal 3D printed part. The part type is defined by the structural category of the part to be printed, based on its geometric topological features or main functional characteristics, such as a casing or a thin-walled rotating body. The material type is defined by the type of metal material used in the metal 3D printing, such as TC4 titanium alloy.

[0040] Furthermore, the collected part type and material type are integrated to form printing information. This involves combining the internal codes of the part type and material type into a single printing information data package, such as [Part type: casing type, Material type: TC4 titanium alloy]. This step clearly communicates the key operating condition of the current target being a thin-walled titanium alloy shell to downstream modules, enabling the subsequent system to prioritize and match the appropriate support feature library for thin-walled titanium alloy parts.

[0041] It should be noted that the above examples are for illustrative purposes only and do not constitute a limitation on the present invention.

[0042] In summary, this step establishes prior constraints for the entire subsequent process. By acquiring the part type and material type beforehand, the identification task is narrowed from a blind search of the entire support library to a finite subset search under specific working conditions. This not only compresses the algorithm's search space and reduces the unnecessary computational overhead of subsequent feature comparisons, but also introduces process correlation, preventing the inclusion of support types incompatible with the current material in the candidate range from the outset, thus reducing the risk of fundamental misjudgment.

[0043] S200: Based on the printed information, retrieve a set of commonly used support types from the support type database.

[0044] After obtaining the printing information, the corresponding support type is identified by building a support type database. Compared with the existing technology, which tends to output random results when dealing with fuzzy features between two types of support, this step is guided by the database and will prioritize matching to support types that are more reasonable in terms of process, thus improving the rationality of the identified process logic.

[0045] Step S200 provided in this embodiment includes:

[0046] Build a database of supporting types;

[0047] Based on the printed information, a matching process is performed in the support type database to obtain a set of commonly used support types.

[0048] In this embodiment, a support type database is constructed, including:

[0049] Collect metal 3D printing support types corresponding to different part types and different material types, and obtain standard shape feature parameters corresponding to each support type. The standard shape feature parameters include standard aspect ratio range, standard contour point cloud density, and standard branch angle range.

[0050] The part type, the material type, the corresponding support type, and the standard shape feature parameters are associated and stored to construct a support type database.

[0051] Specifically, firstly, prior data collection is conducted for combinations of different part types and material types. The collected data includes commonly used support structure categories under these combined working conditions, as well as the corresponding standard geometric characteristic parameters. Standard shape characteristic parameters include: standard aspect ratio range, i.e., the ratio range of the length to the width of the main support structure or cell on the horizontal projection plane; standard contour point cloud density, i.e., the order of magnitude of discrete points on the outer surface of the support within a unit volume or area; and standard branch angle range, i.e., the angle range between the main trunk and branches during the growth of the support structure.

[0052] For example, it is necessary to print a conformal cooling channel part of an injection mold. The part type and material type are identified as conformal runner mold and TC4 titanium alloy, respectively. Based on the identification results, typical supports used in printing are collected and combined. Due to the high precision requirements of the inner wall of the runner and the narrow internal space, point column supports are usually selected to facilitate powder discharge and post-processing. Subsequently, feature extraction is performed on the selected point column supports to obtain standard shape feature parameters: standard aspect ratio range is 0.8:1 or 1.2:1; standard contour point cloud density is 200 pts / cm³; standard branch angle range is 0°.

[0053] Furthermore, the part type, material type, support type, and standard shape feature parameters are associated and stored. Under the same part and material combination, multiple commonly used support types can be stored to form a commonly used support dataset, which is then entered into the support type database. When identifying unknown blanks in the future, the support data corresponding to the part type and material type in the support type database can be matched preferentially.

[0054] For example, continuing with the previous example, the associated stored data record might be: [Part type: conformal flow channel mold, material type: TC4 titanium alloy, support type: point column support, standard aspect ratio: 0.8-1.2, standard contour point cloud density: 200pts / cm³, standard branch angle: none]; or [Part type: conformal flow channel mold, material type: TC4 titanium alloy, support type: thin-walled sheet support, standard aspect ratio: greater than 3.0, standard contour point cloud density: 500pts / cm³, standard branch angle: none]; other types could be [Part type: impeller, material type: Al-Si-10Mg, support type: tree-like branch support, standard aspect ratio: 0.5-2.0, standard contour point cloud density: 50pts / cm³, standard branch angle: 30°-60°]. The above associated stored data records together constitute the support type database.

[0055] In this step, firstly, the aforementioned support type database is constructed. Further, based on the obtained print information data package, the two key fields of part type and material type are extracted and combined using a logical AND operation to generate the composite primary key for this database query. Next, the corresponding query command is executed in the support type database to obtain the matching results, i.e., the corresponding set of commonly used support types.

[0056] For example, continuing with the previous example, the printed information is [Part type: conformal flow channel mold, material type: TC4 titanium alloy]. A query command is executed in the support type database to retrieve the corresponding support type and standard shape feature parameters for the part type being conformal flow channel mold and the material type being TC4 titanium alloy. The commonly used support type set is obtained as follows: [point column support, standard aspect ratio 0.8-1.2, standard contour point cloud density 200pts / cm³]; and [thin-walled sheet support, standard aspect ratio: greater than 3.0, standard contour point cloud density: 500pts / cm³].

[0057] It should be noted that the above values ​​are for illustrative purposes only and do not constitute a limitation on the present invention.

[0058] In summary, by leveraging database relationships, the comparison focus of the recognition algorithm is narrowed down to candidate types with reasonable manufacturing processes. This endows the recognition system with the ability to be guided by prior process knowledge. For example, for conformal flow channel parts, priority is given to the parameter range of point columnar or thin-walled sheet supports, while actively reducing the matching weight of tree-like and block-like supports. This ensures that the recognition results not only conform to geometric features but also align with actual manufacturing process practices, thereby improving the engineering practicality of the recognition output.

[0059] S300: Extract the support shape feature parameters in metal 3D printing, and perform support type identification based on the support shape feature parameters to obtain the support type identification result.

[0060] After obtaining the support shape feature parameters, the support shape feature parameters in the actual metal 3D printing are extracted, and the similarity is calculated and compared. Finally, a weighted comprehensive similarity is calculated as the support type identification result, so that when dealing with the transitional form support between sheet and block, a smoother and more reasonable matching result can be given.

[0061] Step S300 provided in this embodiment includes:

[0062] Based on the part type, extraction parameters are obtained, including point cloud sampling step size, point cloud denoising threshold, and spatial coordinate filtering range.

[0063] Using the extraction parameters, three-dimensional point cloud data of the metal 3D printed parts and supports are collected as support shape feature parameters, wherein the support shape feature parameters include the aspect ratio range, contour point cloud density and branch angle range of the support.

[0064] Calculate the deviation between the aspect ratio range and the standard aspect ratio range of multiple commonly used support types in the common support type set, and obtain multiple aspect ratio similarities;

[0065] Calculate the deviation between the outline point cloud density and the standard outline point cloud density of multiple common support types in the common support type set, and obtain multiple outline similarities;

[0066] Calculate the deviation between the branch angle range and the standard branch angle range of multiple common support types in the common support type set, and obtain the similarity of multiple branch angles;

[0067] The aspect ratio similarity, contour similarity, and branch angle similarity of each commonly used support type are weighted and summed to obtain a comprehensive similarity.

[0068] The most commonly used support type with the highest overall similarity is selected as the support type identification result.

[0069] In this step, due to the significant differences in geometric dimensions among different types of parts, it is necessary to dynamically adjust the point cloud accuracy range according to the part type. First, based on the part type, the corresponding point cloud sampling step size, point cloud denoising threshold, and spatial coordinate filtering range parameters are extracted. The point cloud sampling step size, point cloud denoising threshold, and spatial coordinate filtering range parameters mentioned above are prior parameters, and their meaning and function will not be explained in detail here.

[0070] For example, the input part type is conformal flow channel mold, and the extraction parameters are configured as follows: [point cloud sampling step size: 0.05mm, point cloud denoising threshold: 0.1mm, spatial coordinate filtering range: 10mm-15mm].

[0071] Furthermore, using the adaptive parameters generated in the previous step, the 3D scanning device is driven to acquire the actual point cloud of the printed blank, the support area is segmented from the point cloud, and the support shape feature parameters that match the standard parameter definitions in the support type database are extracted: the aspect ratio range of the support, the density of the contour point cloud, and the range of the branch angle.

[0072] Furthermore, the extracted actual support shape feature parameters are compared with the support shape feature parameters of all corresponding part types in the common support type set to obtain similarity, and a comprehensive similarity is calculated through weighted summation. The similarity includes: standard aspect ratio similarity (deviation from the aspect ratio range); contour similarity (deviation from the standard contour point cloud density); and branch angle similarity (deviation from the standard branch angle range). Deviation is calculated by dividing the actual data by the standard data. For example, aspect ratio similarity is calculated as the ratio of the actual aspect ratio to the median of the standard aspect ratio range. The weights for the weighted summation can be preset by those skilled in the art based on the significance of the features' impact on the support's mechanical properties and manufacturing process; for example, aspect ratio 0.4, contour density 0.4, and branch angle 0.2.

[0073] Furthermore, after obtaining the comprehensive similarity score, multiple comprehensive similarity scores are compared to obtain the most commonly used support type with the highest comprehensive similarity score, which is then used as the support type identification result.

[0074] For example, continuing with the previous example, the part type is still a conformal flow channel mold. The actual data collected includes: measured aspect ratio: 1.0, measured contour point cloud density: 180pts / cm³, and measured branch angle: 0°. The above actual data is compared with the data in the common support type set in the support type database to obtain similarity. The common support type set includes two items: candidate A [support type: point column support, standard aspect ratio: 0.8-1.2, standard contour point cloud density: 200pts / cm³, standard branch angle: none] and candidate B [support type: thin-walled sheet support, standard aspect ratio: greater than 3.0, standard contour point cloud density: 500pts / cm³, standard branch angle: none]. Compared to candidate A, the measured aspect ratio is 1 compared to the median of the standard aspect ratio range, therefore the aspect ratio similarity is 1.0; the contour point cloud density is 180 pts / cm³, which differs from the standard 200 pts / cm³ by 10%, therefore the contour similarity is 180 / 200 = 0.9; all branch angles are unbranched, therefore the branch angle similarity is 1. Using the same method, the aspect ratio of candidate B is 0.2, the contour similarity is 0.3, and the branch angle similarity is 1. The preset weights are aspect ratio 0.4, contour density 0.4, and branch angle 0.2. After performing a weighted summation calculation, the comprehensive similarity of candidate A is (1.0×0.4)+(0.9×0.4)+(1.0×0.2)=0.40+0.36+0.20=0.96, and the comprehensive similarity of candidate B is (0.2×0.4)+(0.3×0.4)+(1.0×0.2)=0.08+0.12+0.20=0.40. Since the comprehensive similarity of candidate A is greater than that of candidate B, candidate A is output, i.e., the point column support is the support type identification result.

[0075] It should be noted that the above values ​​are for illustrative purposes only and do not constitute a limitation on the present invention.

[0076] In summary, this step uses 3D point cloud data as input to extract multi-dimensional stereo features such as the aspect ratio, contour density, and branch angle of the support, and calculates the comprehensive similarity for classification. Compared with existing methods that rely on 2D image edge detection or theoretical slice data, this method achieves a complete perception of the full-dimensional stereo topology of the support structure and can realistically reflect the forming deviation of the printed object, ensuring that the recognition results are consistent with the actual physical state of the blank, rather than remaining at the design theoretical value.

[0077] S400: Combine the support type database to obtain the recognition credibility of the support type recognition result.

[0078] Existing identification methods typically output only a classification label, lacking a quantitative assessment of the label's certainty. Even with extremely low confidence levels, the system may still force an output, leading downstream automated equipment to perform erroneous operations based on low-quality data. This step introduces identification confidence as an intermediate monitoring variable. This confidence level not only includes the geometric similarity calculated in step three but also incorporates common part and material parameters extracted from the database. Even with high geometric similarity, if the support type has historically been rarely used in the current part-material combination, the overall confidence level will be moderately lowered. This adds a process logic verification barrier to the identification results and provides a decision-making basis for downstream automated support removal systems.

[0079] Step S400 provided in this embodiment includes:

[0080] The frequency of occurrence of the support type identification results corresponding to the part type and the material type in the support type database is obtained as the part commonness parameter and the material commonness parameter.

[0081] The recognition credibility of the support type identification result is obtained by calculating the weighted sum of the commonness parameters of the parts, the commonness parameters of the materials, and the comprehensive similarity.

[0082] In this step, during the construction and operation of the support type database, not only are standard shape feature parameters stored, but the historical usage frequency of various supports under specific working conditions is also accumulated and recorded. Based on the currently input part type and material type, the frequency of occurrence of the identification result in the corresponding dimension is retrieved from the database as a priori reliability basis. This includes two parameters: part commonality parameter, which is the probability of the currently identified support type appearing in all historical records of the current part type; and material commonality parameter, which is the probability of the currently identified support type appearing in all historical records of the current material type.

[0083] Furthermore, the common parameters of the aforementioned parts and materials are weighted and summed with the obtained comprehensive similarity to calculate the recognition credibility of the support type identification result. The weights in the weighting calculation can be configured by those skilled in the art according to actual needs and experience, ensuring that the sum of the weights equals 1.

[0084] For example, in the case where the part type is a conformal flow channel mold and the material type is TC4 titanium alloy, the previous step identifies the support type as point columnar support. First, the frequency of occurrence of point columnar support is obtained. With 200 records of the part type being conformal flow channel mold, point columnar support appears 150 times, resulting in a part commonality parameter of 0.75. With 500 records of the material type being TC4 titanium alloy, point columnar support appears 350 times, resulting in a material commonality parameter of 0.7. The weights for the part commonality parameter and the material commonality parameter are set to 0.2, and the overall similarity is 0.6, indicating a greater emphasis on the geometric matching degree of the actual point cloud. Using the same similarity result as the previous example, which is 0.96, the confidence level of the identification is calculated as (0.75×0.2)+(0.70×0.2)+(0.96×0.6)=0.15+0.14+0.576=0.866.

[0085] It should be noted that the above values ​​are for illustrative purposes only and do not constitute a limitation on the present invention.

[0086] In summary, this step, based on geometric similarity, introduces two prior frequency parameters: part commonality and material commonality, and calculates a weighted average to generate a comprehensive recognition confidence level. A quantifiable confidence level label is attached to each recognition result: on the one hand, it verifies whether the geometric features are highly matched; on the other hand, it verifies whether the matching result frequently occurs in similar process scenarios. This provides precise numerical basis for subsequent discrimination steps, possesses the ability to self-evaluate output quality, and overcomes the deficiency of existing technologies that only have classification labels and lack deterministic metrics.

[0087] S500: Based on the recognition confidence level, a judgment is made. When the recognition confidence level is greater than or equal to the confidence threshold, the support type recognition result is output. When the recognition confidence level is less than or equal to the confidence threshold, the extraction parameters of the support shape feature parameters are optimized, and the recognition is iterated until a support type recognition result with a recognition confidence level greater than or equal to the confidence threshold is obtained.

[0088] Existing technologies use preset single acquisition parameters, which directly lead to recognition failure for support areas with poor surface conditions. Upon failure, an error message is displayed or an empty result is output, requiring manual parameter resetting and process restart. This step constructs a closed-loop iterative mechanism based on credibility feedback. When the recognition credibility falls below the credibility threshold, the system automatically calculates the credibility deviation and generates optimized extraction parameters accordingly, then automatically jumps to the feature extraction step for re-execution.

[0089] Step S500 provided in this embodiment includes:

[0090] Based on the statistical results of the recognition credibility in the historical support type recognition structure, obtain the credibility threshold;

[0091] When the recognition confidence level is greater than or equal to the confidence threshold, the support type recognition result is output.

[0092] In this step, a confidence threshold is first obtained based on the statistical results of the recognition confidence in the historical support type identification structure. The confidence threshold is dynamically determined after statistical analysis of the recognition confidence distribution of historical support type identification results. After each periodic or individual identification task is completed, historical records that have been confirmed as correctly identified through manual review or subsequent process verification are extracted. The mean and standard deviation of the recognition confidence of these records at the time of identification are calculated, and the confidence threshold for the current task is set accordingly. If historical data is insufficient in the initial stage of operation, a manually preset conservative threshold, such as 0.85, can be used as the initial value.

[0093] Specifically, one way to set a confidence threshold is as follows: First, data is collected, including historical records of the last N correct recognitions and their recognition confidence values; the lower limit of recognition confidence for the sample set is calculated, that is, the mean and standard deviation of the recognition confidence of the N records are calculated, and the difference between the recognition confidence and the standard deviation is the lower limit of recognition confidence, which serves as a reference benchmark; a small tolerance margin is added to the reference benchmark to form the confidence threshold for the current task.

[0094] Furthermore, the recognition confidence level is compared with the confidence threshold. When the recognition confidence level is greater than the confidence threshold, the current recognition result is determined to have a sufficient confidence level, and the result is directly entered into the result output environment to output the supporting type recognition result.

[0095] For example, the historical recognition record database is queried, and 500 correctly recognized records within the past 30 days are selected, specifying the part type as a conformal flow channel mold, the material type as TC4 titanium alloy, and the records have been confirmed as correct by post-processing. A list of recognition confidence scores is then compiled, with a mean confidence score of 0.92 and a standard deviation of 0.05. Therefore, the lower limit of the confidence score is 0.92 - 0.05 = 0.87. Adding a tolerance margin of 0.02, the final confidence threshold is 0.85. Using the above example, the actual recognition confidence score is 0.866. Since this result is greater than the confidence threshold of 0.85, the output support type recognition result is determined to be a point column support.

[0096] It should be noted that the above values ​​are for illustrative purposes only and do not constitute a limitation on the present invention.

[0097] Step S500 provided in this embodiment of the application further includes:

[0098] When the recognition confidence is less than or equal to the confidence threshold, the difference between the confidence threshold and the recognition confidence is calculated as the confidence deviation;

[0099] Calculate 1 plus the aforementioned confidence deviation, and use it as the point cloud optimization coefficient;

[0100] The point cloud sampling step size is calculated and divided by the point cloud optimization coefficient to obtain the optimized point cloud step size.

[0101] The point cloud denoising threshold is calculated by dividing the point cloud optimization coefficient, and the result is the optimized point cloud denoising threshold.

[0102] The lower limit of the spatial coordinate filtering range is calculated and divided by the point cloud optimization coefficient to obtain the lower limit of the optimized filtering range. The upper limit of the spatial coordinate filtering range is calculated and multiplied by the point cloud optimization coefficient to obtain the upper limit of the optimized filtering range.

[0103] The upper limit of the optimized filtering range and the lower limit of the optimized filtering range are combined to form the optimized spatial coordinate filtering range;

[0104] The optimized point cloud step size, the optimized point cloud denoising threshold, and the optimized spatial coordinate filtering range are integrated to form optimized extraction parameters;

[0105] The optimized extraction parameters are used to iterate the identification process until a support type identification result with a confidence level greater than or equal to the confidence threshold is obtained, or the number of iterations is reached.

[0106] In this step, when the obtained recognition confidence is less than the confidence threshold, the recognition method needs to be iterated. The iteration method is to calculate an optimized extraction parameter and iterate according to the optimized extraction parameter until the recognition confidence is greater than or equal to the confidence threshold or the number of iterations is reached.

[0107] The optimization of extraction parameters is as follows: First, the difference between the confidence threshold and the recognition confidence is calculated as the confidence deviation. Further, based on the extraction parameters obtained from the part type: point cloud sampling step size, point cloud denoising threshold, and spatial coordinate filtering range, the confidence deviation is calculated and incremented by 1 to obtain the point cloud optimization coefficient. The point cloud sampling step size is divided by the point cloud optimization coefficient to obtain the optimized point cloud step size. The point cloud denoising threshold is divided by the point cloud optimization coefficient to obtain the optimized point cloud denoising threshold. The lower limit of the spatial coordinate filtering range is divided by the point cloud optimization coefficient to obtain the point cloud denoising threshold. Specifically, the lower limit of the spatial coordinate filtering range divided by the point cloud optimization coefficient yields the lower limit of the optimized filtering range, and the upper limit of the spatial coordinate filtering range multiplied by the point cloud optimization coefficient yields the upper limit of the optimized filtering range. The optimized point cloud step size, optimized point cloud denoising threshold, and optimized spatial coordinate filtering range are integrated to form the optimized extraction parameters.

[0108] For example, when the set confidence threshold is 0.7, and the recognition confidence is 0.866 as in the previous example, the confidence deviation is 0.7-0.866=0.004; the point cloud optimization coefficient is 1+0.004=1.004; the original point cloud sampling step size obtained in the previous example is 0.5mm, so the optimized point cloud step size is 0.5 / 1.004≈0.0498mm; the original point cloud denoising threshold is 0.1mm, so the optimized point cloud denoising threshold is 0.1 / 1.004≈0.0996mm; the original spatial coordinate filtering range is 10mm-15mm, so the upper and lower limits of the optimized spatial coordinate filtering range are 10 / 1.004≈9.96mm and 15×1.004=15.06mm respectively, so the calculated optimized spatial coordinate filtering range is 9.96mm-15.06mm. Therefore, the final optimized extraction parameters are: [Optimized point cloud step size: 0.0498mm, optimized point cloud denoising threshold: 0.0996mm, optimized spatial coordinate filtering range: 9.96mm-15.06mm].

[0109] The process then jumps back to the feature extraction step, re-collects point cloud data using the optimized parameters, and repeats the subsequent shape feature calculation, similarity comparison, and credibility assessment. The iteration count is set to 5 to prevent infinite loops. The final calculated recognition credibility is 0.872, which is greater than the credibility threshold of 0.87, indicating successful iteration and output of the result.

[0110] It should be noted that the above values ​​are for illustrative purposes only and do not constitute a limitation on the present invention.

[0111] Step S500 provided in this embodiment of the application further includes:

[0112] When the number of iterations reaches the preset number of iterations, and the recognition credibility obtained in each iteration is less than the credibility threshold, the comprehensive similarity of the supporting shape feature parameters in each iteration is calculated.

[0113] If the overall similarity is less than or equal to the preset similarity threshold, it is determined that the current metal 3D printing support has a chaotic shape due to printing failure and does not have the characteristics of the standard support type, and an abnormal support shape prompt message is output.

[0114] If the overall similarity is greater than the preset similarity threshold, it is determined that the current recognition process has not matched the optimal extraction parameters, and a recognition incomplete prompt message is output.

[0115] In this step, during the parameter optimization iterative loop, two metrics are continuously monitored: the current iteration count and the recognition credibility of each iteration. Once the cumulative iteration count reaches the preset maximum, such as 5 iterations, and in these 5 attempts, the calculated recognition credibility fails to meet the credibility threshold, the system will trigger the iteration termination protection mechanism, exit the optimization loop, and enter anomaly analysis. First, the comprehensive similarity of the support shape feature parameters in each iteration is calculated, and a similarity threshold is preset for the comprehensive similarity. When the comprehensive similarity is less than or equal to the preset similarity threshold, it is determined that the current metal 3D printing support has a chaotic shape due to printing failure and does not possess the characteristics of the standard support type, and an abnormal support shape prompt message is output. When the comprehensive similarity is greater than the preset similarity threshold, it is determined that the current recognition process has not matched the optimal extraction parameters, and an recognition incomplete prompt message is output.

[0116] The similarity threshold is set based on the highest similarity score ever recorded in known failed cases, and must be lower than the lower tolerance limit of the parameter for the standard support type in the database. For example, historical scan samples manually labeled as printing failures or morphological abnormalities are retrieved from the support type database. The overall similarity score between these failed samples and their theoretically designed support type is calculated, and the maximum value is extracted. A distinction buffer margin is set, and the sum of the maximum overall similarity score and the buffer margin is the preset similarity threshold. For instance, historical data shows that the collapsed block support, which most resembles a point column support, had a highest overall similarity score of 0.58. With a buffer margin of 0.05, the preset similarity threshold is 0.63.

[0117] For example, in the anomaly analysis phase, the comprehensive similarity of the support shape feature parameters in three iterations is calculated to be 0.55, 0.58, and 0.52, respectively. The representative comprehensive similarity, i.e., the maximum value of 0.58, is taken, and a similarity threshold of 0.60 is set. The representative comprehensive similarity is then compared with the similarity threshold. If it is found to be less than the similarity threshold, it is determined that the current metal 3D printed support has a disordered shape due to printing failure and does not possess the characteristics of a standard support type. An abnormal support shape is output to the log system, suggesting manual intervention to check the printing quality or directly marking it as scrap. If the representative comprehensive similarity obtained after iteration is 0.63, which is greater than the similarity threshold, it is determined that the current identification process has not matched the optimal extraction parameters. An incomplete identification is output to the log system, suggesting manual assistance or triggering a higher-precision offline re-inspection process.

[0118] It should be noted that the above values ​​are for illustrative purposes only and do not constitute a limitation on the present invention.

[0119] In summary, this step constructs a closed-loop iterative control mechanism based on confidence feedback. When the recognition result fails to reach the confidence threshold, the system automatically triggers a parameter optimization process, reducing the sampling step size, lowering the denoising threshold, expanding the search window, and re-executing feature extraction and recognition. This endows the system with adaptive robustness to complex surface conditions, enabling it to overcome the feature ambiguity caused by powder adhesion and surface roughness by dynamically adjusting the acquisition parameters. Furthermore, it achieves online self-optimization of feature extraction parameters, automatically converging to a high-confidence result without manual intervention. While ensuring the accuracy of the final output, this significantly improves the automation level of the recognition process.

[0120] In summary, the embodiments of this application narrow the identification range by prior constraints of part information, and combine the dual mechanisms of feature extraction and credibility iteration verification. While ensuring high accuracy and robustness in the identification of metal 3D printing support types, it also achieves strong adaptability of the identification process to complex blank surface conditions. Ultimately, it provides highly reliable structured data support that can directly drive machine path planning for subsequent automated support removal processes.

[0121] Example 2, as Figure 2 As shown in the embodiment of this application, an AI-driven automatic identification system for metal 3D printing support types is provided. The system includes:

[0122] The part information acquisition module 11 is used to acquire the part type and material type of the metal 3D printed part as printing information;

[0123] Obtain the part type and material type of the metal 3D printed part as printing information, including:

[0124] The part type and material type of the metal 3D printed parts are collected and integrated to form printing information. The part type is defined as the structural category of the part to be formed in the metal 3D printing, and the material type is defined as the metal material category used in the metal 3D printing.

[0125] Support type data management module 12 is used to match and obtain a set of commonly used support types in the support type database based on the printed information;

[0126] Based on the printed information, a set of commonly used support types is retrieved from the support type database, including:

[0127] Build a database of supporting types;

[0128] Based on the printed information, a matching process is performed in the support type database to obtain a set of commonly used support types.

[0129] This includes building a supporting type database, including:

[0130] Collect metal 3D printing support types corresponding to different part types and different material types, and obtain standard shape feature parameters corresponding to each support type. The standard shape feature parameters include standard aspect ratio range, standard contour point cloud density, and standard branch angle range.

[0131] The part type, the material type, the corresponding support type, and the standard shape feature parameters are associated and stored to construct a support type database.

[0132] The feature extraction and recognition reasoning module 13 is used to extract the support shape feature parameters in metal 3D printing, and to perform support type recognition based on the support shape feature parameters to obtain the support type recognition result;

[0133] Extract the support shape feature parameters from metal 3D printing, and perform support type identification based on the support shape feature parameters to obtain the support type identification result, including:

[0134] Based on the part type, extraction parameters are obtained, including point cloud sampling step size, point cloud denoising threshold, and spatial coordinate filtering range.

[0135] Using the extraction parameters, three-dimensional point cloud data of the metal 3D printed parts and supports are collected as support shape feature parameters, wherein the support shape feature parameters include the aspect ratio range, contour point cloud density and branch angle range of the support.

[0136] Calculate the deviation between the aspect ratio range and the standard aspect ratio range of multiple commonly used support types in the common support type set, and obtain multiple aspect ratio similarities;

[0137] Calculate the deviation between the outline point cloud density and the standard outline point cloud density of multiple common support types in the common support type set, and obtain multiple outline similarities;

[0138] Calculate the deviation between the branch angle range and the standard branch angle range of multiple common support types in the common support type set, and obtain the similarity of multiple branch angles;

[0139] The aspect ratio similarity, contour similarity, and branch angle similarity of each commonly used support type are weighted and summed to obtain a comprehensive similarity.

[0140] The most commonly used support type with the highest overall similarity is selected as the support type identification result.

[0141] The method of obtaining the recognition credibility of the support type recognition result by combining the support type database includes:

[0142] The frequency of occurrence of the support type identification results corresponding to the part type and the material type in the support type database is obtained as the part commonness parameter and the material commonness parameter.

[0143] The recognition credibility of the support type identification result is obtained by calculating the weighted sum of the commonness parameters of the parts, the commonness parameters of the materials, and the comprehensive similarity.

[0144] The credibility assessment and iteration control module 14 is used to combine the support type database to obtain the recognition credibility of the support type recognition result, make a judgment based on the recognition credibility, and when the recognition credibility is less than or equal to the credibility threshold, optimize the extraction parameters of the support shape feature parameters and iterate the recognition until a support type recognition result with recognition credibility greater than or equal to the credibility threshold is obtained.

[0145] Specifically, the recognition credibility of the support type identification result is determined based on the recognition credibility obtained from the support type database, including:

[0146] Based on the statistical results of the recognition credibility in the historical support type recognition structure, the credibility threshold is obtained.

[0147] Specifically, when the recognition confidence level is less than or equal to the confidence threshold, the extraction parameters of the support shape feature parameters are optimized, and the recognition is iteratively performed until a support type recognition result with a recognition confidence level greater than or equal to the confidence threshold is obtained, including:

[0148] When the recognition confidence is less than or equal to the confidence threshold, the difference between the confidence threshold and the recognition confidence is calculated as the confidence deviation;

[0149] Calculate 1 plus the aforementioned confidence deviation, and use it as the point cloud optimization coefficient;

[0150] The point cloud sampling step size is calculated and divided by the point cloud optimization coefficient to obtain the optimized point cloud step size.

[0151] The point cloud denoising threshold is calculated by dividing the point cloud optimization coefficient, and the result is the optimized point cloud denoising threshold.

[0152] The lower limit of the spatial coordinate filtering range is calculated and divided by the point cloud optimization coefficient to obtain the lower limit of the optimized filtering range. The upper limit of the spatial coordinate filtering range is calculated and multiplied by the point cloud optimization coefficient to obtain the upper limit of the optimized filtering range.

[0153] The upper limit of the optimized filtering range and the lower limit of the optimized filtering range are combined to form the optimized spatial coordinate filtering range;

[0154] The optimized point cloud step size, the optimized point cloud denoising threshold, and the optimized spatial coordinate filtering range are integrated to form optimized extraction parameters;

[0155] The optimized extraction parameters are used to iterate the identification process until a support type identification result with a confidence level greater than or equal to the confidence threshold is obtained, or the number of iterations is reached.

[0156] Among them, when the number of iterations has been reached but no support type identification result with a recognition credibility greater than or equal to the credibility threshold has been obtained, including:

[0157] When the number of iterations reaches the preset number of iterations, and the recognition credibility obtained in each iteration is less than the credibility threshold, the comprehensive similarity of the supporting shape feature parameters in each iteration is calculated.

[0158] If the overall similarity is less than or equal to the preset similarity threshold, it is determined that the current metal 3D printing support has a chaotic shape due to printing failure and does not have the characteristics of the standard support type, and an abnormal support shape prompt message is output.

[0159] If the overall similarity is greater than the preset similarity threshold, it is determined that the current recognition process has not matched the optimal extraction parameters, and a recognition incomplete prompt message is output.

[0160] The result output interface module 15 is used to make a judgment based on the recognition credibility. When the recognition credibility is greater than or equal to the credibility threshold, the support type recognition result is output.

[0161] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0162] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. An AI-driven method for automatic identification of support types in metal 3D printing, characterized in that, include: Obtain the part type and material type of the metal 3D printed part as printing information; Based on the printed information, a set of commonly used support types is obtained by matching and retrieving the support type database; Extract the support shape feature parameters from metal 3D printing, and identify the support type based on the support shape feature parameters to obtain the support type identification result; Based on the support type database, the recognition credibility of the support type recognition result is obtained; Based on the recognition credibility, a judgment is made, and when the recognition credibility is greater than or equal to the credibility threshold, the support type recognition result is output; When the recognition confidence is less than or equal to the confidence threshold, the extraction parameters of the support shape feature parameters are optimized, and the recognition is iterated until a support type recognition result with a recognition confidence greater than or equal to the confidence threshold is obtained.

2. The AI-driven automatic identification method for metal 3D printing support type according to claim 1, characterized in that, Obtain the part type and material type of the metal 3D printed part as printing information, including: The part type and material type of the metal 3D printed parts are collected and integrated to form printing information. The part type is defined as the structural category of the part to be formed in the metal 3D printing, and the material type is defined as the metal material category used in the metal 3D printing.

3. The AI-driven automatic identification method for metal 3D printing support type according to claim 1, characterized in that, Based on the printed information, a set of commonly used support types is retrieved from the support type database, including: Build a database of supporting types; Based on the printed information, a matching process is performed in the support type database to obtain a set of commonly used support types.

4. The AI-driven automatic identification method for metal 3D printing support type according to claim 3, characterized in that, Build a supporting type database, including: Collect metal 3D printing support types corresponding to different part types and different material types, and obtain standard shape feature parameters corresponding to each support type. The standard shape feature parameters include standard aspect ratio range, standard contour point cloud density, and standard branch angle range. The part type, the material type, the corresponding support type, and the standard shape feature parameters are associated and stored to construct a support type database.

5. The AI-driven automatic identification method for metal 3D printing support type according to claim 1, characterized in that, Extract the support shape feature parameters from metal 3D printing, and perform support type identification based on the support shape feature parameters to obtain the support type identification result, including: Based on the part type, extraction parameters are obtained, including point cloud sampling step size, point cloud denoising threshold, and spatial coordinate filtering range. Using the extraction parameters, three-dimensional point cloud data of the metal 3D printed parts and supports are collected as support shape feature parameters, wherein the support shape feature parameters include the aspect ratio range, contour point cloud density and branch angle range of the support. Calculate the deviation between the aspect ratio range and the standard aspect ratio range of multiple commonly used support types in the common support type set, and obtain multiple aspect ratio similarities; Calculate the deviation between the outline point cloud density and the standard outline point cloud density of multiple common support types in the common support type set, and obtain multiple outline similarities; Calculate the deviation between the branch angle range and the standard branch angle range of multiple common support types in the common support type set, and obtain the similarity of multiple branch angles; The aspect ratio similarity, contour similarity, and branch angle similarity of each commonly used support type are weighted and summed to obtain a comprehensive similarity. The most commonly used support type with the highest overall similarity is selected as the support type identification result.

6. The AI-driven automatic identification method for metal 3D printing support type according to claim 5, characterized in that, Based on the aforementioned support type database, the recognition reliability of the support type identification result is obtained, including: The frequency of occurrence of the support type identification results corresponding to the part type and the material type in the support type database is obtained as the part commonness parameter and the material commonness parameter. The recognition credibility of the support type identification result is obtained by calculating the weighted sum of the commonness parameters of the parts, the commonness parameters of the materials, and the comprehensive similarity.

7. The AI-driven automatic identification method for metal 3D printing support type according to claim 1, characterized in that, Based on the recognition credibility, a judgment is made. When the recognition credibility is greater than or equal to the credibility threshold, the support type recognition result is output, including: Based on the statistical results of the recognition credibility in the historical support type recognition structure, obtain the credibility threshold; When the recognition confidence level is greater than or equal to the confidence threshold, the support type recognition result is output.

8. The AI-driven automatic identification method for metal 3D printing support type according to claim 5, characterized in that, When the recognition confidence level is less than or equal to the confidence threshold, the extraction parameters of the support shape feature parameters are optimized, and the recognition is iteratively performed until a support type recognition result with a recognition confidence level greater than or equal to the confidence threshold is obtained, including: When the recognition confidence is less than or equal to the confidence threshold, the difference between the confidence threshold and the recognition confidence is calculated as the confidence deviation; Calculate 1 plus the aforementioned confidence deviation, and use it as the point cloud optimization coefficient; The point cloud sampling step size is calculated and divided by the point cloud optimization coefficient to obtain the optimized point cloud step size. The point cloud denoising threshold is calculated by dividing the point cloud optimization coefficient, and the result is the optimized point cloud denoising threshold. The lower limit of the spatial coordinate filtering range is calculated and divided by the point cloud optimization coefficient to obtain the lower limit of the optimized filtering range. The upper limit of the spatial coordinate filtering range is calculated and multiplied by the point cloud optimization coefficient to obtain the upper limit of the optimized filtering range. The upper limit of the optimized filtering range and the lower limit of the optimized filtering range are combined to form the optimized spatial coordinate filtering range; The optimized point cloud step size, the optimized point cloud denoising threshold, and the optimized spatial coordinate filtering range are integrated to form optimized extraction parameters; The optimized extraction parameters are used to iterate the identification process until a support type identification result with a confidence level greater than or equal to the confidence threshold is obtained, or the number of iterations is reached.

9. The AI-driven automatic identification method for metal 3D printing support type according to claim 8, characterized in that, When the number of iterations has been reached but no supporting type identification result with a recognition confidence level greater than or equal to the confidence threshold has been obtained, including: When the number of iterations reaches the preset number of iterations, and the recognition credibility obtained in each iteration is less than the credibility threshold, the comprehensive similarity of the supporting shape feature parameters in each iteration is calculated. If the overall similarity is less than or equal to the preset similarity threshold, it is determined that the current metal 3D printing support has a chaotic shape due to printing failure and does not have the characteristics of the standard support type, and an abnormal support shape prompt message is output. If the overall similarity is greater than the preset similarity threshold, it is determined that the current recognition process has not matched the optimal extraction parameters, and a recognition incomplete prompt message is output.

10. An AI-driven automatic recognition system for metal 3D printing support types, characterized in that, For implementing the AI-driven automatic identification method for metal 3D printing support types according to any one of claims 1-9, the system comprises: The part information acquisition module is used to acquire the part type and material type of the metal 3D printed part as printing information; The support type data management module is used to match and obtain a set of commonly used support types from the support type database based on the printed information. The feature extraction and recognition reasoning module is used to extract the support shape feature parameters in metal 3D printing, and to perform support type recognition based on the support shape feature parameters to obtain the support type recognition result; The credibility assessment and iteration control module is used to combine the support type database to obtain the recognition credibility of the support type recognition result. When the recognition credibility is less than or equal to the credibility threshold, the extraction parameters of the support shape feature parameters are optimized and the recognition is iterated until a support type recognition result with recognition credibility greater than or equal to the credibility threshold is obtained. The result output and interface module is used to make a judgment based on the recognition credibility. When the recognition credibility is greater than or equal to the credibility threshold, the support type recognition result is output.

Images