Vision system for identifying support structures of 3D printed parts

Abstract

This disclosure describes a system and method for post-processing 3D-printed parts. For example, a support structure for the 3D-printed part can be removed during post-processing. In this system and method, a first image of the part is stored in memory. A second image of the 3D-printed part corresponding to that part is also captured. One or more cutting paths between the 3D-printed part and the support structure are then determined based on the first and second images. The 3D-printed part can then autonomously separate from the support structure by cutting through the cutting paths.

Description

Background Technology

[0001] This invention generally relates to the 3D printing of parts, and more specifically, to support structures for identifying 3D printed parts.

[0002] 3D printing technology has enabled the creation of a wide variety of parts, pushing the boundaries of traditional manufacturing and enabling the production of parts that perfectly match application requirements. However, the level of autonomy in 3D printing remains very limited because many pre- and post-processing steps must be performed manually. Therefore, mass production of custom parts can be very costly, primarily due to rigid production systems, process design, and the significant amount of manual labor required to execute processing steps. A key element in improving the level of autonomy and self-control in additive manufacturing is enhancing the information flow between the actual printing of 3D-printed parts and the manufacturing cells that perform further processing of the parts.

[0003] 3D printing technology (for metals and plastics) allows for the manufacture of a wide variety of parts with great flexibility. However, these technologies still involve many manual pre- and post-processing steps. The three most significant drawbacks of current processes include long processing times, inconsistent and unreliable quality, and high cost of produced parts. Furthermore, to obtain a fully functional product, the combination of different materials with parts and 3D-printed components still requires numerous design and organization-intensive tasks involving substantial manual labor. For example, it remains difficult to use different materials in the 3D printing process for a specific part. Additionally, many manual steps are still required to assemble parts that were previously produced more competitively using other manufacturing methods (such as electroforming, casting, or milling). Therefore, 3D-printed parts are often moved to another production line or even another location to assemble various 3D-printed and machined parts into the final product.

[0004] It should also be understood that 3D printing of parts requires some type of support structure to hold the part in place as it is formed by the 3D printing system. After the part is formed by the 3D printing system, various post-processing steps are required to prepare the part. For example, the part must be separated from the support structure before further processing and / or use. Inspection may also be necessary. Several methods can be used to separate the part from the support structure. For example, a fluid can be used to dissolve the support structure, thus separating it from the part. Alternatively, if the support structure is made of a poorly soluble material, it can be physically cut to separate the part from the support structure. In this case, autonomous removal of the support structure for a 3D-printed part using a machine (e.g., CNC, robot, etc.) typically relies on a pre-programmed path trajectory to determine where the cutting tool will cut through the support structure or part. However, this method can lead to cuts at undesirable locations and inconsistent cuts between parts due to unpredictable physical deformations (bending, lifting, cracking, etc.) occurring at various geometric locations on the part and support structure. Additional problems arise during autonomous removal of the support structure due to the lack of quality checks on the support structure removal process. Due to these issues, post-processing of 3D printed parts is currently a slow and expensive process. Therefore, it is desirable to be able to remove the support structure precisely and consistently from part to part. Summary of the Invention

[0005] This invention provides a system and method for post-processing 3D-printed parts, which may include removing a support structure from the 3D-printed part. The system and method include storing a first image of the component in a memory and capturing a second image of the 3D-printed component corresponding to the part. A cutting path between the 3D-printed part and the support structure is then determined based on the first and second images. The support structure and the 3D-printed part can then autonomously separate from each other by cutting through the cutting path. This invention may also include any other aspects or any combination thereof described in the following written description or the accompanying drawings. Attached Figure Description

[0006] The invention can be more fully understood by reading the following description in conjunction with the accompanying drawings, in which:

[0007] Figure 1 This is a schematic diagram of a 3D printing system;

[0008] Figure 2 This is a close-up stereoscopic view of a part of a 3D printed component;

[0009] Figure 3 This is a close-up stereoscopic view of another part of the 3D printed component;

[0010] Figure 4 Yes Figure 2Another magnified stereoscopic view of a portion of the 3D printed part;

[0011] Figure 5 This is a schematic diagram of another 3D printed part with identification markings;

[0012] Figure 6 This is a flowchart of one embodiment of the support structure removal system;

[0013] Figure 7 This is a flowchart of another embodiment of the support structure removal system;

[0014] Figure 8 This is a schematic diagram of an autonomous method for post-processing 3D printed parts, illustrating information input and learning;

[0015] Figure 9 This is a schematic diagram of a unit device for autonomous post-processing 3D printed parts; and

[0016] Figure 10 This is a flowchart of a method for post-processing 3D printed parts. Detailed Implementation

[0017] One problem with 3D printing technology is that, due to undesirable events during the printing process, such as thermal stress, impurities, and other disturbances, the construction of 3D-printed parts is difficult to precisely match the specifications provided in the corresponding CAD drawings. Therefore, a single CAD drawing may be insufficient to generate machine cutting paths for removing support structures from the part. Thus, the embodiments described herein can be used to record 3D-printed parts and generate cutting paths for separating the part from the support structure using an autonomous vision system. The autonomous vision system can also utilize machine learning or artificial intelligence algorithms.

[0018] The described embodiments further include an information marking solution for autonomous marking and manufacturing of 3D printed parts. Additive manufacturing technology is capable of producing highly diverse parts based on CAD designs. Furthermore, several parts with different designs can be printed on the same substrate. Therefore, it is desirable to be able to autonomously identify 3D printed parts in order to automate post-processing of the parts. Post-processing of 3D printed parts may require a large amount of part-related information, such as materials, CAD models, support structure designs, and quality specifications. Therefore, it is desirable to incorporate markings on the 3D printed parts, support structures, or substrates during the 3D printing process to encode information about the parts within the markings to support post-processing activities.

[0019] like Figure 1As shown, the system may have a base 10 on which a 3D printer 14 forms a 3D printed part 12. The 3D printed part is formed on the base 10 and attached to the base 10 by one or more support structures 18, which are attached to the base 10 and the 3D printed part 12 on opposite sides. Preferably, the support structure 18 is formed by the 3D printer 14 when the 3D printed part 12 is formed. Thus, for example, in the case where the 3D printed part 12 is formed of metal by a 3D metal printer 14, the support structure 18 is also metallic. In any case, preferably, the 3D printed part 12 and the support structure 18 are formed of the same material and have the same curing (e.g., plastic part). The support structure 20 may also be printed to support the internal structure of the 3D printed part 12. That is, the holes or cavities 22 of the 3D printed part 12 may have support structures 20 connected to different parts on opposite sides to support the holes or cavities 22 during forming. The bottom support structure 18 is located at... Figure 2 The internal support structure 20 is shown in more detail in the diagram. Figure 3 The details are shown in more detail below. For example... Figure 4 As shown, the support structure can be composed of multiple support members 22, which are spaced apart from each other and parallel to each other. Preferably, at least five support members 20 are equidistant from each other and parallel to each other, or even more preferably, at least ten support members 20 are equidistant from each other and parallel to each other.

[0020] A vision system 24 with a camera 26 (e.g., a 3D camera, laser sensor, and ordinary RGB camera) mounted on the robotic arm 28 can also be provided to capture images of the 3D printed part 12. The vision system 24 can also be mounted in a stationary position or elsewhere on another movable structure if desired. After the 3D printed part 12 has been formed by the 3D printer 14, the vision system 24 captures one or more images of the 3D printed part 12 and the support structures 18, 20. As further described below, the controller 30 then determines the cutting path 56 between the 3D printed part 12 and the support structures 18, 20. Figure 5 A cutting system 32 with a cutting head 34 and a robotic arm 36 is also provided for cutting along a cutting path 56 through the 3D printed part 12 or support structure 18, 20 to separate the 3D printed part 12 from the support structure 18, 20. The cutting system 32 is considered a post-processing system 32, and it should be understood that other post-processing systems 32, such as an inspection system 32, are also possible. Preferably, the cutting of the 3D printed part 12 or support structure 18, 20 involves using a rotary grinder, laser, or reciprocating saw on the cutting head 34. It should be understood that the methods and systems herein can also be used with other processing systems, such as an inspection system, which learns and tracks the quality of multiple 3D printed parts 12.

[0021] In one embodiment, the determination of the cutting path 56 can be accomplished in two steps. In the first step, the system can utilize offline learning from the CAD model. In the second step, the system can utilize online learning (“continuing education”) from quality assessment. It should be understood that offline learning refers to updates that occur before or after the actual use of the 3D printed part 12, while online learning refers to updates that occur during the actual use of the 3D printed part 12 or as a result of the actual use of the 3D printed part 12.

[0022] Offline learning may include generating synthetic images from CAD models (e.g., RGB or point clouds) to construct a representative dataset. It should be understood that the synthetic images are digital images defined by data recognizable by controller 30 and can be modified to update the synthetic images based on discrepancies identified between the CAD model and captured images of the 3D printed part 12, or discrepancies identified between two or more 3D printed parts 12. It may be desirable to generate multiple images of part 12 and support structures 18, 20 from different perspectives. It may also be desirable to generate images of part 12 with partially removed support structures 18, 20 from different perspectives. It may also be desirable to generate images of part 12 and / or support structures 18, 20 that deviate from the reference CAD model, for example, where part 12 partially separates from support structures 18, 20 due to deformation caused by thermal stress, resulting in undesirable pores or other defects that may be 3D printed on part 12 and support structures 18, 20. The CAD model can also be used to generate cutting paths 56 visible on the synthesized image, and / or can generate regions to define parts 12 and support structures 18, 20, which can then be used to generate cutting paths 56. The system is preferably trained using machine learning or artificial intelligence techniques to generate cutting paths 56 from the synthesized image. When the system is used online to separate support structures 18, 20 from multiple 3D printed parts 12, the synthesized model can be updated continuously (i.e., based on each 3D printed part 12 and its captured image) or periodically (i.e., based on images captured from multiple 3D printed parts 12). Therefore, the cutting paths 56 determined by the system are adjusted over time as the system updates the synthesized image based on learning that occurs during online use.

[0023] While offline generation of synthetic images may provide reliable results in many cases, it can present challenges with unexpected deformations in some 3D printed parts 12. Therefore, a system that learns from existing experience is preferred. Online learning can begin with the use of offline-generated synthetic images on a new 3D printed part 12. Actual captured images of the 3D printed part acquired during the cutting process (after the removal of each support structure 18, 20) can then be stored. The most recent image can then be compared with the original CAD model (i.e., the initial synthetic image) as a quality control action. If a difference from the reference CAD model is detected to exceed a threshold, the synthetic image can be updated to adjust the image and the correct cutting path 56. Quality control actions can also be used to confirm that the support structures 18, 20 have been completely and adequately removed, or to analyze whether cracks have appeared on the 3D printed part 12 during the removal of the support structures 18, 20. Retraining of the system and updating of the synthetic images can be performed continuously or periodically. After sufficient updates, the system can be expected to become increasingly reliable and eventually be used solely based on visual input from the 3D printed part 12 (i.e., without referencing the CAD model).

[0024] The system acquires input from camera sensor 26 and stores the captured images in computer memory. Using an autonomous computer vision algorithm, a first determination of the cutting path 56 between the 3D printed part 12 and the support structures 18, 20 can be performed. In a second step, the algorithm can group different regions of the support structures 18, 20 into independent regions. For example, it may be desirable to group the base support structure 18 separately from the inner support structure 20, because cutting different support structures 18, 20 may affect the 3D printed part 12 differently. Therefore, it is desirable that each region consists of closely positioned support structures 18, 20 that will be affected by the cutting operation in that region. In each region, the algorithm detects the cutting path 56 between the support structures 18, 20 separated from other regions and the 3D printed part 12. Optionally, the autonomous computer vision algorithm can compare the determined cutting path 56 in each region with the original CAD model. Therefore, when capturing images of the 3D printed part 12, it may be desirable to capture at least two different images of at least two different regions of the 3D printed part 12, where each image has a different region. Then, the two different regions in the synthesized image corresponding to the two different regions of the 3D printed part 12 can be updated based on the two different captured images. The system can also identify and quantify deviations (differences) between the captured images and the CAD model, or between captured images of different 3D printed parts 12. Preferably, as Figure 5As shown, the autonomously generated cutting path 56 is positioned through the support structures 18 and 20, slightly spaced from the connection point 54 of the 3D printed part 12 to provide a safety margin with the 3D printed part 12. Finally, this set of cutting paths 56 is sent to the cutting system 34 for the actual cutting and separation of the support structures 18 and 20 from the 3D printed part 12.

[0025] The system comprises a set of steps including capturing images of the CAD model and / or the actual 3D printed part 12. The connection points 54 between the support structures 18, 20 and the 3D printed part 12 can then be determined autonomously using an algorithm. The calculated connection points 54 can then be compared with the connection points in the original CAD model. Deviations in the captured images can then be identified and quantified. A cutting path 56, passing through the support structures 18, 20 slightly away from the connection points 54 to provide a safety margin, can then be autonomously generated to provide the actual cutting path 56. The cutting path 56 can then be connected to the cutting system 32 that performs the actual cutting. It is also desirable for the vision system to capture additional images of the region of interest by zooming (e.g., optical or digital zoom) within the region of interest where the support structures 18, 20 are attached to the 3D printed part 12. These images can then be used in a learning algorithm to improve and enhance the accuracy of the generated cutting path 56.

[0026] An exemplary flowchart of the system and method described herein is shown in Figure 6As shown in the figure, a synthetic image is initially generated from the CAD model of the component (38). It should also be understood that an initial image can be captured from the actual 3D printed component 12. The initial image of the component is then stored in memory (40). Multiple images can also be used for the initial image. Machine learning or artificial intelligence can also be used with multiple initial images to learn the position of the connection point 54 between the 3D printed component 12 and the support structures 18, 20. In this case, at least some of the multiple initial images will include the 3D printed component 12 and the support structures 18, 20 attached to the 3D printed component 12. A synthetic image can also be formed from multiple initial images and learning. An image of the actual 3D printed component 12 is then captured, where the 3D printed component 12 corresponds to the component (42) of the initial image. Preferably, the captured image is an RGB image, a depth image, a point cloud, or a line scan. If desired, the initial image can also be an RGB image, a depth image, a point cloud, or a line scan. Multiple images of different actual 3D printed components 12 can also be captured. Machine learning or artificial intelligence can also be used in conjunction with multiple captured images of the actual part 12 to learn the location of the connection point 54 between the 3D printed part 12 and the support structures 18, 20. In this case, each of the multiple captured images will include the 3D printed part 12 and the support structures 18, 20 attached to the 3D printed part 12. A synthetic image can also be formed from the multiple captured images and the learned images. Based on the initial image (44) of the 3D printed part 12 and the captured images, one or more cutting paths 56 are then determined between the 3D printed part 12 and one or more support structures 18, 20 attached to the 3D printed part 12 (46). The cutting paths 56 can be determined using a database of multiple initial images and multiple captured images, as well as machine learning or artificial intelligence of the database. The synthetic image (initial image) can then be updated based on the difference between the synthetic image and the captured image (48). Then, the cutting system 32 can use the generated cutting path 56 to cut through the 3D printed part 12 or the support structures 18, 20 to separate the 3D printed part 12 from the support structures 18, 20 (50). It should be understood that the described system and method can be implemented in the controller 30 in the form of a non-transitory computer-readable medium, which includes program code controlling the 3D printer 14, the vision system 24, and / or the cutting system 32 to autonomously execute the described method.

[0027] like Figure 5As shown, one or more markers 52A-C can also be used to determine the cutting path 56 between the 3D printed part 12 and the support structures 18, 20 attached to the 3D printed part 12. As shown, marker 52A can be embedded in the 3D printed part 12, encoding information about the part 12, substrate 10, support structures 18, 20, and / or required post-processing steps and tools. Marker 52B encoding this information can also be embedded in one or more support structures 18, 20. Marker 52C containing this information can also be embedded in the substrate 10. Preferably, when markers 52A and 52B are embedded in the 3D printed part 12 and / or support structures 18, 20, preferably, when the 3D printed part 12 and support structures 18, 20 are printed by the 3D printer 14, the markers 52A and 52B are printed on the 3D printed part 12 and / or support structures 18, 20 by the 3D printer 14. Although the mark 52C on the base plate 10 can be embedded or placed separately from the 3D printing process, the mark 52C can also be 3D printed onto the base plate 10 by the 3D printer 14 during printing, or the base plate 10 and the mark 52C can be 3D printed together by the 3D printer 14 during printing. The marks 52A-C can also be laser engraved. The marks 52A-C enable flexible, reliable, and independent devices to autonomously identify the 3D printed part 12 to fully formulate the entire cascade of post-processing activities, thereby enabling cost-competitive production of customized 3D printed parts 12.

[0028] By embedding markings 52A-C on the 3D printed part 12, support structures 18, 20, or substrate 10 of the 3D printed part 12, information related to the 3D printed part 12, required post-processing steps, and the part 12 itself can be made available. Therefore, machines and equipment used for post-processing can read reliable information about the actual part 12 to be processed from the markings 52A-C to achieve autonomous post-processing activities. Furthermore, the necessary equipment, tools, and machines required for post-processing a given part 12 may be part-specific. Therefore, the markings 52A-C for the 3D printed part 12 can be used to contain all the information required to achieve unit autonomy.

[0029] To complete one or more post-processing steps, such as powder removal, removal of support structures 18, 20, quality inspection, or heat treatment for stress relief, a wide range of relevant information about the 3D printed part 12 may be required, such as material, mechanical and electrical properties, quality specifications, support structures 18, 20, and CAD design. Using markings 52A-C to encode the information associated with each 3D printed part 12 allows information to flow from the actual 3D printed part 12 to other machines performing post-processing operations.

[0030] The types of information that can be provided by markings 52A-C include (but are not limited to): the number and location of parts 12 printed on substrate 10; specifications of the 3D printed parts 12, such as material, mechanical and electrical properties, and mass limitations; cutting tools required for cutting through support structures 18, 20 and their parameterization (e.g., cutting speed); dimensions and location of support structures 18, 20; limitations on post-processing steps, such as mechanical limitations of the 3D printed parts 12; information on how to remove powder left from 3D printing and support structures 18, 20; information on heat treatment parameters used in the furnace; and information on how to assemble the 3D printed parts 12 into the final product (e.g., assembly steps). Where markings 52B are provided on support structures 18, 20 for removal, separate markings 52B may be required on individual support structures 18, 20 to provide independent location information for support structures 18, 20, thereby improving accuracy when removing multiple support structures 18, 20.

[0031] Markings 52A-C can be sensed by vision system 24, and controller 30 can determine the position of cutting path 56 based on the sensed markings 52A-C. Cutting system 32 can then cut along cutting path 56 through 3D printed part 12 or support structures 18, 20 to separate 3D printed part 12 from support structures 18, 20. Markings 52A-C can also encode information associated with a specific 3D printed part 12, support structure 18, 20, and post-processing steps. During the 3D printing process, markings 52A-C are preferably created using the same printer 14 used to print 3D printed parts 12 and support structures 18, 20. Markings 52A-C can include various features. For example, markings 52A-C can be printed on 3D printed parts 12, support structures 18, 20, or on a substrate 10 containing multiple, optionally different, 3D printed parts 12. Markings 52A-C can encode information in a machine-readable format (e.g., QR code, barcode, notch, a series of notches, engraving, or relief). Markings 52A-C can encode information about other markings 52A-C or the 3D printed part 12, such as relative distances to other markings 52A-C, relative distances to the 3D printed part 12, relative distances to support structures 18, 20, relative distances to the connection point 54 between the 3D printed part 12 and support structures 18, 20, relative distances to the cutting path 56 between the 3D printed part 12 and support structures 18, 20, or may include the type of connection between the 3D printed part 12 and support structures 18, 20 to support post-processing of the part. Information about the dimensions of the connection point 54 between support structures 18, 20 and the 3D printed part 12 can also be used to calculate the forces required for cutting. Markings 52A-C can be used to indicate critical points on support structures 18, 20. Markings 52A-C can indicate geometric points requiring quality checks and how such checks should be performed. Markers 52A-C can encode the relative position of markers 52A-C to one or more connection points 54 or to one or more cutting paths 56. Encoded positional information or other component 12 information can be explicitly encoded in markers 52A-C, allowing direct reading of said information without reference to another data source. Alternatively, markers 52A-C can encode pointers to information about the 3D printed part 12, such as the relative position of connection points 54 or cutting paths 56. For example, markers 52A-C can encode a unique identifier or URL that allows access to desired information stored in a database (e.g., a website, web service, and the cloud). The database may also include other post-processing information, such as powder removal, assembly, and polishing. Markers 52A-C (especially marker 52A on the 3D printed part 12) can also encode the identity of the 3D printed part 12 (e.g., a generic part number or a specific serial number).The markings 52A-C can also encode information for end-of-life handling of the 3D printed part 12, such as dismantling, recycling, or ownership instructions.

[0032] When the mark 52A is placed on the 3D printed part 12 itself, preferably, the function of the 3D printed part 12 is not affected by the mark 52A. In this case, the information encoded in the mark 52A can be used not only during manufacturing and assembly, but also during disassembly and recycling at the end of the product's life. This can be used to improve recycling efficiency. For example, if the component 12 or the assembled product has direct information stored thereon about how to best disassemble, reuse, and / or recycle the component, recycling facilities can recycle in a more efficient manner. If necessary, the information stored on the marks 52A-C can also be encrypted when confidentiality is required.

[0033] Figure 7 An exemplary flowchart of the system and method described herein is shown. As shown, the 3D printer 14 can be used to simultaneously 3D print part 12, support structures 18, 20, and one or more markers 52A-C (58). The markers 52A-C can then be sensed by the vision system 24 or other sensing system (60). Based on the information obtained from the markers 52A-C, one or more cutting paths 56 are then determined between the 3D printed part 12 and one or more support structures 18, 20 attached to the 3D printed part 12 (62). The cutting system 32 can then use the resulting cutting paths 56 to cut through the 3D printed part 12 or support structures 18, 20 to separate the 3D printed part 12 from the support structures 18, 20 (64). It should be understood that the described system and method can be implemented in a controller 30 in the form of a non-transitory computer-readable medium, which includes program code controlling the 3D printer 14, vision system 24, and / or cutting system 32 to autonomously perform the described methods.

[0034] Providing autonomous manufacturing cell facilities for realizing digital designs into functional products is also useful. Autonomous manufacturing cell facilities can include two elements: on the one hand, a set of physical equipment and fixtures, such as, but not limited to, 3D printers, lasers, printers, robotic systems, vision and sensor systems, storage systems, quality inspection, conveyors, fixtures, milling or CNC machines. The primary purpose of these devices includes performing the required set of operations, such as printing, polishing, material removal, inspection, or assembly, to physically transform raw materials and components into functional products; on the other hand, a set of intelligent control systems that determine and learn processing steps based on existing knowledge of the product (e.g., CAD of parts, assembly plans, etc.) and / or input data from the physical systems. The control system converts available data into useful information to perform an optimal set of tasks for creating functional products based on user-defined performance metrics such as cost, production time, or number of steps.

[0035] Manufacturing cell units will also be advantageous, autonomously designing production processes and coordinating, executing, controlling, monitoring, and improving processing steps to obtain fully functional products. Such technology will enable the flexible, reliable, and cost-competitive production of customized parts in large volumes with short delivery cycles. One such manufacturing cell unit is described herein, autonomously designing, implementing, controlling, monitoring, and improving these processing steps to obtain a fully functional component, part, component, or device that needs to be manufactured and has specific characteristics (e.g., material, mechanical and electrical properties, functional characteristics, quality specifications, and attached components or surface finishes). On one hand, manufacturing cell units include hardware such as (but not limited to) production, assembly, and processing machines and equipment, as well as robotic systems, transport systems, storage systems, and quality control systems. On the other hand, the equipment and machines within the manufacturing cell unit need to collaborate and work together autonomously to achieve customer-defined goals, such as minimum production time or minimum production cost. The available information in the system can include prior knowledge of components, parts, components, or equipment and their requirements (CAD models, assembly plans, etc.), input data from the physical systems within the manufacturing unit, and process data generated by the unit. Decision-making tasks include (but are not limited to): manufacturing process design, i.e., the process steps required to obtain the product; allocating production steps to different hardware; production sequence; production activities within each hardware; coordinated control of production activities; monitoring processing steps; learning strategies for production steps; and learning optimal process parameters to maximize, for example, final quality.

[0036] Autonomous manufacturing cell units rely on production hardware (i.e., post-processing tools) as described above, such as 3D printers, lasers, robotic systems, CNC machines, storage racks, and conveyors. The autonomous system considers various available production information within a set of intelligent learning control systems, including prior knowledge of the product (e.g., CAD of parts, assembly plans, and material properties) and data generated by the arrangement of the manufacturing cell, to optimally achieve customer-defined goals. This provides manufacturing autonomy generated by the intelligent control systems to design and coordinate manufacturing cell activities and improve the production process through learning. The result of these control systems is a set of processing steps and parameters that produce a fully functional product. A set of learning algorithms and methods encoded in the control systems is responsible for: achieving coordination and communication between different machines and equipment; designing and implementing process steps and learning the most suitable process parameters for each processing step; coordinating data from machines and equipment; integrating available information into a set of processing steps; and the autonomous execution of the production process. The autonomous post-processing of 3D printed parts described herein can also be extended to other manufacturing operations, such as the autonomous assembly of 3D printed parts and many other operations.

[0037] The intelligent control system embedded in the autonomous manufacturing cell unit is one of two main software blocks: (i) the process design and optimization block, or (ii) the process execution and control block. The former (i) analyzes the manufacturing cell components, performs calculations, and refines steps and parameters to optimally produce the desired product using a learning strategy based on selected performance indicators (KPIs) (this learning strategy depends on the manufacturing cell description), including available machines and equipment and their specifications, product information (such as the CAD design of the final product), and process measurements generated by process execution. The second block (ii) controls the execution of manufacturing steps and captures data from sensors and equipment to create new information shared with the process design and optimization block. To create a feasible processing strategy and learn the optimal strategy, the following steps can be used: A list of available machines and equipment in the manufacturing cell unit can be read. Performance indicators can also be selected. Product information (CAD files, parts, materials, attributes, and quality requirements, etc.) can also be read. An initial processing strategy based on the available information can be obtained, satisfying cell boundaries and machine / equipment capabilities, including the required processing steps, sequencing, and machine / equipment parameters. The learning algorithm can be invoked to set up the components to be improved and create a representative set of production sequences (training set) to be run in a manufacturing cell layout. Based on the training set provided by the learning algorithm, processing steps can be executed multiple times to collect process measurements from the manufacturing cell apparatus. Feedback from a continuous quality control system that monitors quality during the build process can also be included. Self-correcting actions can also be included during process execution through the control system to avoid accidents. Taking into account the production boundaries of the manufacturing cell, CAD constraints, product specifications, and selected performance metrics (cost, energy, aging, and weight, etc.), the learning algorithm can be solved to generalize from observed measurements regarding the optimal processing sequence and processing parameters. The results can be refined by repeating the above steps using the learning parameters obtained during the corresponding invocation of the learning algorithm. In subsequent runs of the processing strategy, the learning algorithm can employ different processing tools to better meet the set performance metrics. The learning algorithm can also be invoked to understand the variability caused by varying input parameters by (a) using the solution information obtained from the last invocation of the execution, and (b) by making multiple invocations of a generalized strategy using classification and regression techniques. Figure 8 The steps of using feedback input information, learning, execution, and adaptation are illustrated.

[0038] The task of the 3D printer in the arrangement is to produce most (e.g., 90%) of the individual parts for the final product. Once a part is 3D printed, this information is sent to the production module, which processes the information and sends instructions to the robotic system on how to remove the substrate with the 3D printed part on it. The system then autonomously performs post-processing tasks on the part, such as removing support structures, machining surface edges, polishing, heat treatment, and quality inspection.

[0039] The post-processing can begin with a camera recording the 3D-printed part on the substrate. These images are processed by a production block algorithm that distinguishes the support structure from the main 3D-printed part. Based on this information, the connection points between the support and the 3D-printed part can be generated, and the trajectory the robot will follow later when removing the support structure can be represented. To achieve this, the production block algorithm plans the removal and sends instructions to the robot system regarding the selection of tools (e.g., laser cutting and / or milling tools depending on the material to be removed) and the type of brush used later for fine surface polishing. Initially, the production block can have presets with different options, and then over time it can learn to select tools based on experience with support structure removal. During this process, the robot system can use its vision and laser scanner capabilities to record the 3D-printed part and compare the dimensions of the actual 3D-printed part with the CAD drawing to verify the quality of the printed part.

[0040] Following post-processing, the same or another robotic system, or a combination of both, can replace the components and begin assembling the different parts (e.g., made of different materials) into the final product. Here, an assembly block algorithm can be used to control and plan the assembly process. The first input can be taken from the CAD design of the components and the product. The assembly block algorithm can then calculate the steps and identify the tools required for each step of the assembly. The robot can receive this information and begin execution. Furthermore, the robotic system can use its vision capabilities to receive information about the product and its working environment in real time and send this information to the assembly block to update the assembly plan. Input can be provided by cameras and other sensors, such as acoustic measurements, laser measurements, X-rays, etc. The assembly block preferably autonomously determines the order in which the individual components are processed to obtain the final product.

[0041] Therefore, the entire production line can include machines organized into a cell structure. Each cell is equipped with a set of tools (robots, printers, or furnaces with processing tools such as lasers, grinders, etc., or assembly tools such as clamps and wrenches) and sensors. This organization provides greater flexibility, priority-based production, and scalability, and is robust to unplanned downtime for individual cells.

[0042] Additive manufacturing cells can perform the tasks required to build a product in the following ways: They can be used to process different materials that can be used anywhere in the product. Conventional production equipment can be used, enabling the use of production tools created as needed using tool printers. Inventory of commodity parts (e.g., screws) can be used; these parts are not product-specific and are used in any component or assembly. Storage systems can be used for intermediate parts produced. Systems can be used for transporting parts (e.g., mobile robots, cantilever robots, and drones). Robotic cells can be used to assemble all parts together into a finished product. Quality inspection cells can be used to perform various inspections (e.g., surface inspection, tomography, and functional testing) on ​​individual parts or assembled products. Figure 9 An example of battery arrangement is shown. As shown in one unit assembly 70, a production unit 78 performing 3D printing, processing, or assembly may have robotic units 72 distributed therein. The robotic units 72 can perform the assembly, processing, inspection, or movement of parts between different production units 78 for multiple 3D printed parts. In another unit assembly 74, a robotic conveyor 76 may also be used to move parts between production units 7 if needed.

[0043] Figure 10An exemplary flowchart of the system and method described herein is shown. In the first two steps, a list of user-defined performance metrics (80) and available post-processing tools (82) is read from one or more computer memories. The user-defined performance metrics may be the performance of post-processing multiple 3D printed parts 12 into finished or partially finished parts, and may include cost, production time, or number of steps. Available post-processing tools may have different processing capabilities and may include reduction manufacturing methods, such as lasers or grinders, wrenches, or any tools, machines, etc., described above. The list of available post-processing tools may also include post-processing tools from different production cells 78. For example, the cell assembly may have production cells 78 distributed by robotic cells 72. Robotic cells 72 may perform post-processing steps, such as assembling multiple 3D printed parts 12 together, or may move 3D printed parts between different production cells 78. Conveyors 76 may also move 3D printed parts 12 between production cells 78. Production cells 78 may include one or more 3D printers 14 for printing 3D printed parts 12, and may also include CNC machines or other production equipment as described above. After reading the user-defined performance metrics and the list of available post-processing tools, a first post-processing strategy (84) can be generated in one or more computer processors using the user-defined performance metrics and the list of available post-processing tools. The first post-processing strategy may include computer instructions for operating a first subset of the post-processing tools. The first 3D printed part 12 can then be printed using one or more 3D printers 14 (86). It should be understood that the post-processing strategy can be generated after printing the 3D printed part 12 if necessary. The first 3D printed part 12 is then post-processed using the first post-processing strategy and a first subset of the post-processing tools (88). Post-processing may include removing support structures 18, 20 printed with the 3D printed part 12 or any other post-processing processes described above. For example, removing support structures 18, 20 may include milling, laser cutting, or sawing through the 3D printed part 12 or support structures 18, 20. Support structure 18 may be attached to base 10, on which the 3D printed part 12 is formed. Then, in response to the post-processing of the first 3D printed part 12 (90), a first performance metric is determined using one or more sensors corresponding to a user-defined performance metric. The system can then learn from the difference between the user-defined performance metric and the first performance metric to improve future post-processing strategies (92). For example, a second post-processing strategy can then be generated in one or more computer processors using the user-defined performance metric, the first performance metric, and a list of available post-processing tools (92, 84). The second post-processing strategy may include computer instructions for operating a first or second subset of the post-processing tools, and the second post-processing strategy may differ from the first post-processing strategy.The second post-processing strategy may include computer instructions for operating a second subset of post-processing tools, wherein the first and second subsets of post-processing tools include different post-processing tools. The first and second post-processing strategies may also include quality checks on the first and second 3D printed parts 12. For example, the quality check may compare the 3D printed part 12 with a CAD drawing. The second 3D printed part 12 can then be printed using one or more 3D printers 14 (86), and post-processing can be performed using the second post-processing strategy and a first or second subset of the post-processing tools (88). Then, in response to the post-processing of the second 3D printed part 12 (90), a second performance metric may be determined using one or more sensors corresponding to user-defined performance indicators. The first and second performance metrics are then compared in one or more computer processors to determine the level of improvement between the first and second performance metrics, if necessary (92), such learning and improvement occurring with each iteration. Preferably, the entire method operates autonomously.

[0044] While preferred embodiments of the invention have been described, it should be understood that the invention is not limited thereto and modifications can be made without departing from the invention. Although each embodiment described herein may relate to only certain features and may not specifically relate to every feature described with respect to other embodiments, it should be recognized that, unless otherwise described, the features described herein are interchangeable even without specific mention of specific features. It should also be understood that the foregoing advantages are not necessarily the only advantages of the invention, and it is not necessarily expected that each embodiment of the invention will achieve all of the described advantages. The scope of the invention is defined by the appended claims, and all apparatuses and methods falling within the scope of the claims, whether literally or equivalently, are protected by this invention.

Claims

1. A method for post-processing a 3D printed part (12), comprising: Store the first image of the component in memory (40); Based on the first image (44), a cutting path is determined between the 3D printed part (12) and the support structure (18, 20) attached to the 3D printed part (12), wherein the cutting path (56) is visible, or wherein a region is generated to define the part (12) and the support structure (18, 20). Capture a second image of the actual 3D printed part (12), which corresponds to the part (42) in the first image. If the difference between the second image (46) and the first image exceeds a threshold, the first image is updated to adjust the image and the cutting path (56), wherein the first image is updated based on the difference between the first image and the second image; as well as The adjusted cutting path is output to the post-processing system (32) for further processing (50) of the 3D printed part (12).

2. The method according to claim 1, wherein the post-processing system (32) is a cutting system (32), and further comprises cutting through the 3D printed part (12) or the support structure (18, 20) along the cutting path to separate the 3D printed part (12) from the support structure (18, 20) (50).

3. The method according to claim 1 or 2, wherein a plurality of first images are stored in the memory, at least some of the plurality of first images including the 3D printed part (12) and the support structure (18, 20) attached to the 3D printed part (12), and further comprising learning the position of the connection (54) between the 3D printed part (12) and the support structure (18, 20) from the plurality of first images using machine learning or artificial intelligence.

4. The method according to claim 1 or 2, wherein a plurality of second images are captured from different 3D printed parts (12), each of the plurality of second images including the 3D printed part (12) and the support structure (18, 20) attached to the 3D printed part (12), and further comprising learning the position of the connection (54) between the 3D printed part (12) and the support structure (18, 20) from the plurality of second images using machine learning or artificial intelligence.

5. The method according to claim 1 or 2, wherein the first image is a composite image (38) and / or wherein the second image is an RGB image, a depth image, a point cloud, or a line scan.

6. The method of claim 1 or 2, further comprising a database of a plurality of first images and a plurality of second images, wherein the cutting path is determined using the database based on machine learning or artificial intelligence.

7. The method according to claim 1 or 2, wherein the first image is generated from a CAD model of the part (38); and / or wherein the first image is captured from another 3D printed part (12) and wherein the first image is updated (48) based on the difference between the first image and the second image.

8. The method according to claim 7, wherein two different second images of the 3D printed part (12) are captured from two different regions of the 3D printed part (12), each second image having a different region, and two different regions in the first image corresponding to the two different regions of the 3D printed part (12) are updated based on the two different second images.

9. The method according to claim 1 or 2, wherein the cutting path passes through the support structure (18, 20) and is spaced apart from the 3D printed part (12).

10. The method according to claim 1 or 2, wherein the support structure (18) is attached to the base (10) and the 3D printed part (12) is formed on the base (10); or wherein the support structure (20) is attached to different portions of the 3D printed part (12) on opposite sides.

11. The method of claim 2, wherein cutting includes milling, laser cutting, or sawing through the 3D printed part (12) or the support structure (18, 20).

12. The method according to claim 1 or 2, wherein the 3D printed part (12) and the support structure (18, 20) are both metallic; and / or wherein the 3D printed part (12) and the support structure (18, 20) are both made of the same material having the same curing properties.

13. The method according to claim 1 or 2, wherein the support structure (18, 20) comprises a plurality of support members (22) spaced apart from and parallel to each other.

14. A method for removing a support structure (18, 20) from a 3D printed part (12) according to any one of claims 1 to 10, wherein after each of the plurality of 3D printed parts (12), the first image is updated based on the difference between the first image and the second image, and wherein the support structure (18, 20) is removed from each of the plurality of 3D printed parts (12).

15. The method of claim 14, wherein cutting includes milling, laser cutting, or sawing through the 3D printed part (12) or the support structure (18, 20).

16. The method according to claim 14 or 15, wherein the 3D printed part (12) and the support structure (18, 20) are both metallic; and / or wherein the 3D printed part (12) and the support structure (18, 20) are both made of the same material having the same curing properties.

17. The method according to claim 14 or 15, wherein the support structure (18, 20) comprises a plurality of support members (22) spaced apart from and parallel to each other.

18. A method for removing a support structure (18, 20) from a plurality of 3D printed parts (12) according to any one of claims 1 to 10, wherein a first image is periodically updated based on a difference between a first image and a second image of the plurality of 3D printed parts (12), and wherein the support structure (18, 20) is removed from each of the plurality of 3D printed parts (12).

19. The method of claim 18, wherein cutting includes milling, laser cutting, or sawing through the 3D printed part (12) or the support structure (18, 20).

20. The method according to claim 18 or 19, wherein the 3D printed part (12) and the support structure (18, 20) are both metallic; and / or wherein the 3D printed part (12) and the support structure (18, 20) are both made of the same material having the same cured state.

21. The method according to claim 18 or 19, wherein the support structure (18, 20) comprises a plurality of support members (22) spaced apart from and parallel to each other.

Images