Recoater automated monitoring system and method for additive manufacturing machines

Through automatic defect identification and online monitoring of the automated monitoring system, the recoater errors in additive manufacturing are accurately identified and classified, and the problems of inaccurate error identification and insufficient cumulative impact assessment capabilities in the existing technology are solved, and the construction quality and production are improved efficiency.

CN111723453BActive Publication Date: 2024-11-01GENERAL ELECTRIC CO
View PDF 9 Cites 0 Cited by

Patent Information

Application Number
CN202010200603.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Priority Date
2019-03-20
Filing Date
2020-03-20
Publication Date
2024-11-01
Estimated Expiration
2040-11-01

AI Technical Summary

Technical Problem

The prior art has inaccuracy in the identification and classification of recoater errors in the additive manufacturing process and insufficient cumulative impact assessment capabilities, which makes it difficult to ensure the construction quality, which leads to material loss and low production efficiency.

Method used

Adopt an automated monitoring system, including an automatic defect identification subsystem and an online monitoring system, build a prediction model through offline learning and semi-automated learning stages, identify and classify the error indication of the recoating device, and combine user feedback optimization model to achieve the accuracy of recoating errors. Identify and classify.

Benefits of technology

It improves the accurate identification and classification of recoater errors, can promptly warn users and terminate failed construction, reduce material losses, save costs and time, and improve the output and productivity of additive manufacturing machines.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN111723453B_ABST
    Figure CN111723453B_ABST
Patent Text Reader

Abstract

A system for monitoring an additive manufacturing (AM) machine recoating operation includes an automatic defect recognition subsystem having a catalog of predictive models, each predictive model being applicable to a product and a recoating error indication, the recoating error indication having domain-relevant features; the predictive model representing the recoating error indication as it appears at the pixel level of an image captured during the recoating operation. The system includes an online monitoring subsystem having an image classifier unit that classifies the recoating error indication at the pixel level based on a selected predictive model, the predictive model being selected based on metadata of the predictive model; a virtual depiction unit that creates a virtual depiction of an ongoing AM build from continuously captured images; and a processor unit that monitors the build for recoating error indications, classifies detected indications, and provides a determination to the ongoing build regarding the severity of the detected indications. A method and non-transitory computer-readable medium are also disclosed.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure relates to a recoater automated monitoring system and method for an additive manufacturing machine. Background Art

[0002] The additive manufacturing process, which uses laser energy to melt a bed of powder material, builds up a part by successive layers of melted powder. Laser melting follows a CAD file that represents the part itself. For each layer to be built, a new layer of material to be melted is recoated over the previous layer. After the layer buildup is complete, the part can be removed from the powder that has accumulated around it. During these recoating steps, a number of errors can occur. Recoating issues that can occur during the layer buildup process can include underfeeding, powder lumps, blade lines, drops, and flicks.

[0003] Table I is a representative listing of recoater error indication types and possible causes.

[0004]

[0005] Table I

[0006] Conventional methods provide limited ability to quantify the severity of recoater errors during the layup process and the cumulative impact of the errors on the build. Likewise, the art is limited in its ability to assess the impact and cumulative impact of recoater errors to determine if the error is benign, tolerable, or fatal to the ongoing build. Summary of the invention

[0007] The embodied systems and methods provide a recoater automation monitoring system for monitoring an additive manufacturing machine to automatically identify recoater errors. In some implementations, a user may be alerted if the magnitude of the recoater error significantly affects the quality of the build.

[0008] The embodied systems and methods provide accurate identification and classification of recoater error indications. Embodiments can use these indications to determine if an ongoing build has failed (e.g., an out-of-tolerance or low-quality part has been built). Terminating a failed build can reduce material loss, save cost and time, and increase the yield and productivity of an additive manufacturing machine. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 schematically depicts a recoater automation monitoring system according to an embodiment;

[0010] Figure 2 depicts a process for developing a recoater error indication model according to an embodiment;

[0011] Figure 3 depicts a process for automated recoater monitoring according to an embodiment; and

[0012] Figure 4 A system for implementing a recoater automated monitoring system according to an embodiment is depicted. DETAILED DESCRIPTION

[0013] Figure 1 is a schematic diagram of a recoater automated monitoring system 100 according to an embodiment. The embodied recoater monitoring system and method includes two interdependent subsystems - an automatic defect identification system 110 and an online monitoring system 130 - running in parallel.

[0014] The automatic defect recognition subsystem (ADR) 110 may include an offline learning unit 112. The model catalog 114 includes recoater error indication models representing different types of recoater errors occurring. Table I is not an exhaustive list of recoater errors and indications. It should be readily understood that the embodied systems and methods are not limited to the indications in Table I.

[0015] Different recoater error indication models in the model catalog 114 can be optimized for different domains (i.e., machine models) of additive machines and for different machine units within the same domain. Likewise, different recoater error indication models can be optimized for different product types produced by additive machines. The development of each model is completed in a first self-learning phase. After model development, the model can be enhanced in a second phase.

[0016] First, a fully automated self-learning phase is performed to build a set of initial recoater error indication models. The first phase can be performed in an offline learning unit 112 as a fully automated self-learning process that accesses synthetic images 118 from training examples 116. The synthetic images can be generated from existing good constructions with no indications or with a limited number of indications.

[0017] Second, a semi-automated learning phase is performed, which may be user and / or event driven. In this phase, the catalog model may be optimized by supplementing synthetic images trained with manually annotated / labeled examples 120, which include selection images 122 and domain specific data 124. The model is optimized for different parts, product assemblies, additive manufacturing machine domain types, and units. The selection images 122 may include examples of recoater error indications that appear on other machines or product builds. The domain specific data 124 may include tolerances for additive manufacturing machines, historical wear degradation for a certain domain, etc.

[0018] The ADR 110 can create predictive models that can identify recoater defects without the need for a labeled training set compared to build images. These models are based on self-learning that applies synthetically generated recoater error indications and / or data from successful builds. Each predictive model can represent a recoater error indication that may occur on a particular production part. According to an embodiment, the ADR 110 can use a model to identify a recoater error indication using a set of predefined features of some other indication type.

[0019] According to an embodiment, feedback can be used to improve the performance of the recoater automated monitoring system 100. For example, when a user determines whether a marked indication is true (i.e., a recoater error indication in a captured image is identified) or false, the model can be updated to incorporate the user determination. In some embodiments, the location of the error indication within the build can be used to classify the automatically identified indication. In some embodiments, if the recoater error occurs outside the manufactured structure (i.e., a support area for production purposes); or occurs within the structure itself and is considered acceptable for the build, the recoater error can be tolerated (or not tolerated). By including indication classification based on powder bed position within the model, false positives can be reduced. By aligning the indication location with the CAD file of the part, position sensitivity can be determined using recoater errors. Embodiments can measure key indicators of quality, such as indication size, penetration layer, distance from the edge of the part, etc.

[0020] According to an embodiment, a composite image example of a recoater error indication may be generated by the following process. Given a set of K exposed images obtained from a CAD model showing the part and support areas and repaint the image and the corresponding labeled images We generate a set of K repainted images, which include synthesis instructions, denoted as and the corresponding labeled images The indications are marked based on their location (eg, part, support, powder).

[0021] For each labeled image Randomly extract a set of n points (e.g., n=9), the points are evenly distributed between the part, support structure, and powder (i.e., for n=9, three points are distributed on each).

[0022] Center each point extracted in the previous step from each and Extract an M×M square area. The two regions are represented as and In this work, M is set empirically for each type of indication.

[0023] The composite indication for each extracted region is defined as follows: where G is Gaussian noise and w is a weighting factor set empirically for each indication type. The set of all indications for an image is defined as

[0024] By adding the extracted indication to the repainted image I r Create a composite indicator image More specifically, Additionally, create an extracted instruction A marked image of where each indicated marker indicates its location (part, support, powder), such as Given in .

[0025] After creating the synthetic index image, calculate the reference repaint image It is defined as The pixel-level median of all repainted images in .

[0026] Then, for each image, use the exposed image Synthetic instruction image and reference repaint image To calculate the list of pixel-level intensity feature vectors F. The marked indicator image Used to create a label Y list so that each pixel is assigned one of the two labels {indication, not indication}. Finally, the features (F) and labels (Y) are used to train a supervised learning algorithm executed in the offline learning unit 112 to build a prediction model.

[0027] Additive manufacturing machine 160 represents a field of machines. The embodied systems and methods are not limited to this field of additive manufacturing machines. Laser power head 162 provides a focused laser beam to melt the material in powder bed 166 to form an object represented in CAD file 170. Powder material is spread by recoater unit 164. Image capture device 168 obtains an image of the build. Control / motor unit 172 coordinates these operations.

[0028] Figure 2 A process 200 for developing a recoater error indication model according to an embodiment is depicted. At step 205, synthetic image data for recoater error indications is provided to the offline learning unit 112. At step 210, an indication prediction model is constructed. The prediction model is used to classify pixel-level indications of the image acquired during the construction process.

[0029] In step 215, the prediction model is applied to the image acquired during the construction process to classify the indications at the individual pixel level (i.e., semantic segmentation) of the synthesized image data. Semantics are assigned to individual pixels based on the probability of the class to which the pixel-level image most likely belongs by comparing the pixel-level image to the starter model.

[0030] Based on the determination at step 220 to enhance self-learning, process 200 may proceed to steps 225 and / or 230 , user feedback may be introduced (step 225 ), manually annotated / labeled examples 120 may be introduced (step 230 ), or both may be introduced.

[0031] User feedback may include a value representing a user's subjective assessment of the classification generated by the predictive model. For example, a user may indicate satisfaction or dissatisfaction with the classification; acceptance or rejection; a numerical value indicating agreement with the classification (e.g., "1" for very agreement; "10" for very disagreement); or other indications. It should be readily appreciated that the numerical representation is not limited to any particular scale or representation; for example, "1" may indicate very disagreement.

[0032] After introducing user feedback and / or manually annotated / labeled examples, process 200 returns to step 210 to fine-tune the prediction model. Such fine-tuning can incorporate feedback into the semantic segmentation classification. If it is determined at step 220 not to enhance the self-learning process, the prediction model is added to the model catalog 114 at step 222. Based on the determination at step 224 that another prediction model is to be created, process 200 returns to step 205. If no more prediction models are to be created, process 200 ends.

[0033] In some embodiments, the prediction model can be trained for sub-regions of the powder bed. The placement of the imaging device at different locations relative to the powder bed may change the captured image due to off-axis imaging, the field of view or depth of field of the imaging device, lighting conditions, perspective distortion, direction of laser travel, and other factors. These non-recoater artifacts in the image may cause inaccuracies in the determination. The selection of the prediction model for a specific sub-region can be based on metadata associated with the image.

[0034] return Figure 1, a second subsystem operating in parallel with the ADR 110 is an online monitoring system 130, which can automatically identify recoater errors by applying predictive models developed by the ADR and stored in the model catalog 114. The online monitoring system operates in parallel with the ongoing continuous learning operation of the ADR 110, which improves the predictive model using input from the online monitoring system. According to an embodiment, when a recoater error is identified, a user can be alerted if it is determined that the error may have a significant impact on the quality of the component. In some embodiments, the system 130 can analyze certain recoater error indications to make this impact determination. The online monitoring system can be trained (e.g., initialized and / or configured) prior to use by providing examples from production builds of the specific field unit being monitored. Training can include calculating a reference recoat image as described above and calculates the exposure image containing the user-selected part Affine transformation to the corresponding exposed image from the reference machine.

[0035] Figure 3 A process 300 for automated recoater monitoring according to an embodiment is depicted. At step 305, a recoat image is received. During the additive manufacturing process, the image capture device 168 captures the image. The new image may result in the generation of a post-exposure image and repaint the image Both.

[0036] Preprocessing techniques are applied to both images at step 310. Preprocessing may include applying the affine transformation computed in initialization / configuration, normalizing the image intensities using a reference repaint image, assigning labels to pixels using a labeled image obtained from a reference machine, and registering the image coordinates to coordinates in the CAD file for the part.

[0037] In some embodiments, these images may be provided to the ADR 110 at step 315 for training of a predictive model. At step 320, each captured image may be added to a virtual depiction of the build created by the virtual depiction unit 134. The virtual depiction is a representation of the build in progress. Analysis of the virtual depiction may provide information on the number of layers penetrated by the recoater error, relative to the three-dimensional location of the portion of the build product.

[0038] At step 325, the captured image is compared to a slice of the CAD file representing the specific layer being evaluated to locate the part, support structures, and powder. At step 330, an indication prediction model is applied to the captured image and the positioning information to identify the location of each indication (if any).

[0039] In some embodiments, pixels may be classified with a binary label (e.g., indication = 1, no indication = 0). Indications outside the powder bed are not included. In step 335, the impact of the defect classification labeled pixels is analyzed in the context of their location using the virtual depiction to determine the impact of the defect. Actions are taken based on a catalog of predefined rules applied to the virtual depiction. Rules may specify the number, size, and penetration of indications based on the part being built.

[0040] If the defect is determined to be fatal to part production in step 340, the build is terminated. The termination determination may be made using input from a user or by following a decision tree applied by the algorithm.

[0041] At step 345, an output report is generated containing information (fatal or non-fatal) regarding the results of the build monitoring. The output may include an indication map showing the location and degree of penetration of the indication. The output report may be in a table format containing a unique identifier for each defect detected indicating the type, location (x, y), number of penetration layers, area (mm 2 ) and / or volume (mm 3 The output report can provide the defect location on the powder bed relative to the CAD file (i.e., part of the part or not related to the part), as well as its distance to the part boundary. These factors can be used to manually assess the severity of the accumulated defects.

[0042] Figure 4 A system 400 for implementing the recoater automation monitoring system 100 according to an embodiment is shown. The control processor 410 may include a processor unit 412 and a memory unit 414. The control processor may communicate with elements of the system 100 across a local control / data network and / or an electronic communication network as desired. The processor unit 412 may execute executable instructions 422 that cause the processor to perform queries to a federated data repository according to the embodiments disclosed above. The memory unit 414 may provide a local cache memory to the control processor. The data repository 420 may include a model catalog 424, a captured image repository 426, a virtual depiction 428, a CAD file repository 430, and a rule repository 434 to support the operation of the recoater monitoring system as described above.

[0043] The embodied systems and methods provide classification of recoater error indications with high accuracy and low positive rates. Accurate identification and classification of recoater error indications can result in a decision to terminate the build when the build fails (e.g., poor quality parts) rather than during post-production part inspection. Thus, material loss is reduced and production throughput is increased, saving cost and time. The automated detection and monitoring implemented by the embodied systems and methods is quantitative, with greater accuracy and repeatability than relying on human intervention.

[0044] In accordance with some embodiments, a computer program application stored in a non-volatile memory or computer-readable medium (e.g., register memory, processor cache, RAM, ROM, hard drive, flash memory, CD ROM, magnetic media, etc.) may include code or executable program instructions that, when executed, may instruct and / or cause a controller or processor to perform the methods discussed herein, such as the methods for providing accurate identification and classification of repaint error indications as described above.

[0045] The computer readable medium may be a non-transitory computer readable medium, which includes all forms and types of memory and all computer readable media except transitory propagating signals. In one embodiment, the non-volatile memory or computer readable medium may be an external memory.

[0046] Although specific hardware and methods have been depicted herein, it is noted that any number of other configurations may be provided in accordance with embodiments of the present invention. Therefore, although the basic novel features of the present invention have been shown, depicted, and pointed out, it should be understood that various omissions, substitutions, and changes may be made to the form and details of the illustrated embodiments and their operation. It will be understood by those skilled in the art without departing from the spirit and scope of the present invention. The substitution of elements from one embodiment to another is also fully anticipated and contemplated. The present invention is limited only by the appended claims and their equivalents.

[0047] Further aspects of the invention are provided by the subject matter of the following clauses:

[0048] 1. A system for monitoring recoating in an additive manufacturing machine, the system comprising: an automatic defect recognition subsystem, the automatic defect recognition subsystem comprising: a catalog of a plurality of predictive models, each predictive model being applicable to one of a plurality of products and one of a plurality of recoating error indications, the plurality of recoating error indications having domain-related features; each of the plurality of predictive models representing a recoating error indication occurring at a pixel level of an image captured during a recoating operation on the additive manufacturing machine; an online monitoring subsystem, the online monitoring subsystem comprising: an image classifier unit, the image classifier unit configured to classify the recoating error indication at the pixel level based on one of the selected predictive models, the predictive model being selected based on metadata of the predictive model; a virtual depiction unit, the virtual depiction unit configured to create a virtual depiction of an ongoing build from continuously captured images; and a processor unit, the processor unit configured to execute executable instructions, the executable instructions causing the processor unit to perform a method of monitoring an ongoing additive manufacturing build for recoating error indications, classifying detected indications, and providing a determination regarding the severity of the detected indications to the ongoing additive manufacturing build.

[0049] 2. A system according to any of the preceding clauses, wherein the executable instructions cause the processor to perform the method, comprising: constructing a set of initial recoating indication models in an offline learning unit, the offline learning unit being configured to implement self-learning techniques to process synthetic images of indications; optimizing the model set construction by supplementing the models with manually annotated / labeled examples; and storing the constructed set as predictive models in the model catalog, each predictive model representing a specific type of recoating error indication that may occur in a specific additive manufacturing machine field when producing a specific part.

[0050] 3. A system according to any preceding clause, wherein examples of the manual annotation / marking include selected images of known overpainting error indications selected by a user or domain specific data provided by a user.

[0051] 4. A system according to any of the preceding clauses, wherein the executable instructions cause the processor to perform the method, comprising: applying semantic segmentation at a pixel level of the composite image to classify overpaint error indications represented in the composite image.

[0052] 5. A system according to any of the preceding clauses, wherein the executable instructions cause the processor to perform the method comprising: receiving a layer image captured during the ongoing build; adding the image to the virtual depiction; comparing the layer image to a CAD file to identify a build portion from support structures and powder; predicting a pixel-level binary label for a detected indication; identifying a location of the detected indication by applying a CAD file comparison to the captured image; and using the virtual depiction to analyze the impact of a defect classification in the context of its location.

[0053] 6. A method for monitoring recoating in an additive manufacturing machine, the method comprising: creating a catalog of a plurality of predictive models, each predictive model being applicable to one of a plurality of products and one of a plurality of recoating error indications, the plurality of recoating error indications having domain-related characteristics; representing, in each of the plurality of predictive models, recoating error indications occurring at a pixel level in an image captured during a recoating operation on the additive manufacturing machine; classifying, at the pixel level, the recoating error indications based on one of the selected predictive models, the predictive model being selected based on metadata of the predictive model; and creating a virtual depiction of an ongoing build from continuously captured images.

[0054] 7. A method according to any of the preceding clauses, comprising: constructing a set of initial recoating indication models in an offline learning unit, the offline learning unit being configured to implement a self-learning technique to process synthetic images of indications; optimizing the model set construction by supplementing the models with manually annotated / labeled examples; and storing the constructed set as predictive models in the model catalog, each predictive model representing a specific type of recoating error indication that may occur in a specific additive manufacturing machine field when producing a specific part.

[0055] 8. According to any of the preceding clauses, examples of the manual annotation / marking include selected images of known overpainting error indications selected by a user or domain specific data provided by a user.

[0056] 9. A method according to any preceding clause, comprising applying semantic segmentation at a pixel level of the composite image to classify indications of repainting errors represented in the composite image.

[0057] 10. A method according to any of the preceding clauses, comprising: receiving a layer image captured during the ongoing build; adding the image to the virtual depiction; comparing the layer image to a CAD file to identify the build portion from support structures and powder; predicting pixel-level binary labels for detected indications; identifying the location of the detected indications by applying a CAD file comparison to the captured image; and using the virtual depiction to analyze the impact of defect classification in the context of its location.

[0058] 11. A non-transitory computer-readable medium having executable instructions stored thereon, which, when executed by a processor unit, cause the processor unit to perform a method for monitoring recoating in an additive manufacturing machine, the method comprising: creating a catalog of multiple predictive models, each predictive model being applicable to one of a plurality of products and one of a plurality of recoating error indications, the plurality of recoating error indications having domain-related characteristics; representing, in each of the multiple predictive models, recoating error indications occurring at a pixel level in an image captured during a recoating operation on the additive manufacturing machine; classifying, at the pixel level, the recoating error indications based on one of the selected predictive models, the predictive model being selected based on metadata of the predictive model; and creating a virtual depiction of an ongoing build from continuously captured images.

[0059] 12. A non-transitory computer-readable medium according to any of the preceding clauses, wherein the executable instructions are further configured to cause the processor unit to perform the method, comprising: constructing a set of initial recoating indication models in an offline learning unit, wherein the offline learning unit is configured to implement a self-learning technique to process synthetic images of indications; optimizing the model set construction by supplementing the models with manually annotated / labeled examples; and storing the constructed set as predictive models in the model catalog, each predictive model representing a specific type of recoating error indication, which may occur in a specific additive manufacturing machine field when producing a specific part.

[0060] 13. A non-transitory computer-readable medium according to any of the preceding clauses, wherein the executable instructions are further configured to cause the processor unit to perform the method, wherein examples of the manual annotations / markings include selected images of known overpainting error indications selected by a user or domain-specific data provided by a user.

[0061] 14. According to any of the preceding clauses, the non-transitory computer-readable medium, the executable instructions further configured to cause the processor unit to perform the method, comprising: applying semantic segmentation at a pixel level in the composite image to classify overpaint error indications represented in the composite image.

[0062] 15. A non-transitory computer-readable medium according to any of the preceding clauses, wherein the executable instructions are further configured to cause the processor unit to perform the method, comprising: receiving a layer image captured during the ongoing build; adding the image to the virtual depiction; comparing the layer image to a CAD file to identify a build portion from support structures and powder; predicting a pixel-level binary label for a detected indication; identifying a location of the detected indication by applying a CAD file comparison to the captured image; and using the virtual depiction to analyze the impact of a defect classification in the context of its location.

Claims

1. A system for monitoring recoating in an additive manufacturing machine, characterized in that The system comprises: An automatic defect recognition subsystem, the automatic defect recognition subsystem comprising: a catalog of a plurality of prediction models, each prediction model being applicable to one of a plurality of products and one of a plurality of recoating error indicators, the plurality of recoating error indicators having domain-related characteristics; each of the plurality of predictive models represents an indication of recoating errors occurring at a pixel level of an image captured during a recoating operation on the additive manufacturing machine; An online monitoring subsystem, the online monitoring subsystem comprising: an image classifier unit configured to classify the repaint error indication at the pixel level based on one of the selected prediction models, the prediction model being selected based on metadata of the prediction model; a virtual rendering unit configured to create a virtual rendering of the ongoing construction from the continuously captured images; and A processor unit configured to execute executable instructions that cause the processor unit to perform a method of monitoring an ongoing additive manufacturing build for recoating error indications, classifying the detected indications, and providing a determination to the ongoing additive manufacturing build regarding the severity of the detected indications; the executable instructions causing the processor to perform the method, comprising: constructing a set of initial repaint instruction models in an offline learning unit configured to implement a self-learning technique to process the synthetic images of the instructions; Supplementing the model with manually annotated / labeled examples; and The build sets are stored as predictive models in the model catalog, each predictive model representing a particular type of recoating error indication that is likely to occur in a particular field of additive manufacturing machines when producing a particular part.

2. The system according to claim 1, characterized in that Examples of such manual annotations / labeling include selected images of known overpainting error indications selected by the user or domain specific data provided by the user.

3. The system according to claim 1, characterized in that The executable instructions cause the processor to perform the method, including applying semantic segmentation at a pixel level of the composite image to classify overpaint error indications represented in the composite image.

4. The system according to claim 1, characterized in that The executable instructions cause the processor to perform the method, comprising: receiving a layer image captured during the ongoing build; adding the image to the virtual depiction; comparing the layer image to a CAD file to identify the build portion from support structures and powder; For detected indications, predict pixel-level binary labels; identifying the location of said detected indication by applying a CAD file comparison to the captured image; and The virtual depiction is used to analyze the impact of defect classification in the context of its location.

5. A method for monitoring recoating in an additive manufacturing machine, characterized in that The method comprises: creating a catalog of a plurality of prediction models, each prediction model being applicable to one of a plurality of products and one of a plurality of recoating error indications, the plurality of recoating error indications having domain-related characteristics; representing, in each of the plurality of predictive models, an indication of recoating errors occurring at a pixel level of an image captured during a recoating operation on the additive manufacturing machine; classifying, at the pixel level, the repaint error indication based on one of the selected prediction models, the prediction model being selected based on metadata of the prediction model; and creating a virtual depiction of the ongoing construction from continuously captured images; constructing a set of initial repaint instruction models in an offline learning unit configured to implement a self-learning technique to process the synthetic images of the instructions; Supplementing the model with manually annotated / labeled examples; and The build sets are stored as predictive models in the model catalog, each predictive model representing a particular type of recoating error indication that is likely to occur in a particular field of additive manufacturing machines when producing a particular part.

6. The method according to claim 5, characterized in that Examples of such manual annotations / labeling include selected images of known overpainting error indications selected by the user or domain specific data provided by the user.

7. The method according to claim 5, characterized in that This includes applying semantic segmentation at a pixel level in the composite image to classify repaint error indications represented in the composite image.

8. The method according to claim 5, characterized in that include: receiving a layer image captured during the ongoing build; adding the image to the virtual depiction; comparing the layer image to a CAD file to identify the build portion from support structures and powder; For detected indications, predict pixel-level binary labels; identifying a location of the detected indication by applying a CAD file comparison to the captured image; and The virtual depiction is used to analyze the impact of defect classification in the context of its location.

9. A non-transitory computer-readable medium, characterized in that The non-transitory computer readable medium has executable instructions stored thereon, which, when executed by a processor unit, cause the processor unit to perform a method for monitoring recoating in an additive manufacturing machine, the method comprising: creating a catalog of a plurality of prediction models, each prediction model being applicable to one of a plurality of products and one of a plurality of recoating error indications, the plurality of recoating error indications having domain-related characteristics; representing, in each of the plurality of predictive models, an indication of recoating errors occurring at a pixel level of an image captured during a recoating operation on the additive manufacturing machine; classifying, at the pixel level, the repaint error indication based on one of the selected prediction models, the prediction model being selected based on metadata of the prediction model; and creating a virtual depiction of the ongoing construction from continuously captured images; The executable instructions are further configured to cause the processor unit to perform the method, including: constructing a set of initial repaint instruction models in an offline learning unit configured to implement a self-learning technique to process the synthetic images of the instructions; Supplementing the model with manually annotated / labeled examples; and The build sets are stored as predictive models in the model catalog, each predictive model representing a particular type of recoating error indication that is likely to occur in a particular field of additive manufacturing machines when producing a particular part.

10. The non-transitory computer readable medium of claim 9, wherein: The executable instructions are further configured to cause the processor unit to perform the method, examples of the manual annotations / markings including selected images of known overpainting error indications selected by a user or domain specific data provided by a user.

11. The non-transitory computer readable medium of claim 9, wherein: The executable instructions are further configured to cause the processor unit to perform the method, including applying semantic segmentation at a pixel level of the composite image to classify overpaint error indications represented in the composite image.

12. The non-transitory computer readable medium of claim 9, wherein: The executable instructions are further configured to cause the processor unit to perform the method, including: receiving a layer image captured during the ongoing build; adding the image to the virtual depiction; comparing the layer image to a CAD file to identify the build portion from support structures and powder; For detected indications, predict pixel-level binary labels; identifying the location of said detected indication by applying a CAD file comparison to the captured image; and The virtual depiction is used to analyze the impact of defect classification in the context of its location.

Citation Information

Patent Citations

  • Computer-aided simulation of multi-layer selective laser sintering and melting additive manufacturing processes

    CN107073818A

  • Scalable finite element simulation of additive manufacturing

    CN107403026A

  • System and method for optimizing tool paths based on thermal / structural simulations of a part being produced via a 3d-printer

    CN108292325A

  • Geometrical characteristic and dilution rate modeling method for cladding layer in laser additive remanufacturing

    CN108763801A

  • Traceable method of metal additive manufacturing product

    CN109047756A