Defect detection device, defect detection method, and program

Abstract

To provide a defect detection apparatus, a defect detection method and a program which are capable of reducing the false detection of defects in lamination molding.SOLUTION: A defect detection apparatus comprises an acquisition unit of, in the production of a three-dimensional molded object by a lamination molding device, acquiring a molded region image obtained by capturing a molded region molded by curing a powder material laid on a powder floor and a powder floor image captured with the powder floor having the powder material laid after the molding of the molded region for each layer, a defect detection unit of detecting a defect(s) generated at the powder floor as a candidate for a defect(s) generated at the molded region based on the powder floor image, a region detection unit of detecting the molded region in the powder floor based on the molded region image and a defect determination unit of determining the candidate for the defect(s) in the detected molded region as the defect(s) generated at the molded region based on the detection result of the candidate of the defect(s) and the detection result of the molded region.SELECTED DRAWING: Figure 2

Description

Technical Field

[0001] The present invention relates to a defect detection device, a defect detection method, and a program.

Background Art

[0002] Conventionally, in the production of three-dimensional shaped objects, additive manufacturing (AM) technology has been used. The additive manufacturing technology is a technique for manufacturing a three-dimensional shaped object by stacking two-dimensional layers (such as metal powder or resin) that have been sliced one by one based on three-dimensional design data (STL data) of 3D-CAD. In the additive manufacturing technology, for quality assurance, defects are detected by visually inspecting images captured during the shaping of the shaped object or the manufactured shaped object. In order to further improve the quality, it is required to detect defects that occur in the shaped object during the shaping of the shaped object.

[0003] As a technique for detecting defects during the shaping of a shaped object, for example, in Patent Document 1 below, a technique for detecting defects related to the powder laying state and the surface state of the workpiece based on image data obtained by imaging a powder bed platform during shaping is disclosed.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] However, in the technique of Patent Document 1 above, the region of the shaped object and the region outside the shaped object are not distinguished. Therefore, when detecting a defect that occurs in the shaped object, even a defect that occurs outside the shaped object may be erroneously detected as a defect that occurs in the shaped object.

[0006] In view of the above-mentioned problems, the object of the present invention is to provide a defect detection device, a defect detection method, and a program that can reduce false detection of defects in additive manufacturing. [Means for solving the problem]

[0007] To solve the above-mentioned problems, a defect detection device according to one aspect of the present invention includes, in the manufacturing of a three-dimensional object by an additive manufacturing device, an acquisition unit that acquires for each layer a manufacturing area image captured of a manufacturing area formed by the hardening of powder material laid on a powder bed, and a powder bed image captured of the powder bed on which the powder material was laid after the manufacturing of the manufacturing area; a defect detection unit that detects defects occurring on the powder bed as candidates for defects occurring in the manufacturing area based on the powder bed image; a region detection unit that detects the manufacturing area on the powder bed based on the manufacturing area image; and a defect determination unit that determines, based on the detection results of the candidate defects and the detection results of the manufacturing area, that the candidate defects within the detected manufacturing area are defects occurring in the manufacturing area.

[0008] A defect detection method according to one aspect of the present invention includes an acquisition process in which an acquisition unit acquires, for each layer, a molded region image captured of a molded region formed by the hardening of powder material laid on a powder bed during the manufacturing of a three-dimensional molded object by an additive manufacturing apparatus, and a powder bed image captured of the powder bed on which the powder material was laid after the molded region was formed; a defect detection process in which a defect detection unit detects defects occurring on the powder bed as candidates for defects occurring in the molded region based on the powder bed image; a region detection process in which a region detection unit detects the molded region on the powder bed based on the molded region image; and a defect determination process in which a defect determination unit determines, based on the detection results of the candidate defects and the detection results of the molded region, that the candidate defects located within the detected molded region are defects occurring in the molded region.

[0009] A program according to one aspect of the present invention causes a computer to function as: an acquisition means for acquiring, for each layer, a build region image captured of a build region formed by the hardening of powder material laid on a powder bed in the manufacturing of a three-dimensional object by an additive manufacturing apparatus, and a powder bed image captured of the powder bed on which the powder material was laid after the build region was formed; a defect detection means for detecting defects occurring on the powder bed as candidates for defects occurring in the build region based on the powder bed image; an area detection means for detecting the build region on the powder bed based on the build region image; and a defect determination means for determining, based on the detection results of the candidate defects and the detection results of the build region, that the candidate defects within the detected build region are defects occurring in the build region. [Effects of the Invention]

[0010] According to the present invention, it is possible to reduce false detection of defects in additive manufacturing. [Brief explanation of the drawing]

[0011] [Figure 1] This figure shows the schematic configuration of the additive manufacturing system according to this embodiment. [Figure 2] This block diagram shows an example of the functional configuration of the defect detection device according to this embodiment. [Figure 3] This figure shows an example of a defect-free image of the fabrication region according to this embodiment. [Figure 4] This figure shows an example of a powder bed image free of defects according to this embodiment. [Figure 5] This figure shows an example of a fabrication region image containing defects according to this embodiment. [Figure 6] This figure shows an example of a powder bed image containing defects according to this embodiment. [Figure 7] This figure shows an example of a powder bed image to which information output from the defect detection unit according to this embodiment has been added. [Figure 8] This figure shows an example of a build region image to which information output from the region detection unit according to this embodiment has been added. [Figure 9] This figure shows an example of an output image according to this embodiment. [Figure 10] This is a sequence diagram showing an example of the processing flow in the additive manufacturing system according to this embodiment. [Modes for carrying out the invention]

[0012] Embodiments of the present invention will be described in detail below with reference to the drawings.

[0013] <1. Configuration of the additive manufacturing system> Referring to Figure 1, the schematic configuration of the additive manufacturing system according to this embodiment will be described. Figure 1 is a diagram showing the schematic configuration of the additive manufacturing system according to this embodiment. The additive manufacturing system 1 shown in Figure 1 is a system for detecting defects in three-dimensional objects manufactured using additive manufacturing technology. In this embodiment, an example of detecting defects while the object is being manufactured will be described below. As shown in Figure 1, the additive manufacturing system 1 includes an additive manufacturing device 10, a defect detection device 20, and a user terminal 30. In the additive manufacturing system 1, the additive manufacturing device 10, the defect detection device 20, and the user terminal 30 are connected to a network NW.

[0014] (1) Additive manufacturing device 10 The additive manufacturing apparatus 10 is a device that manufactures three-dimensional objects using additive manufacturing technology. For example, the additive manufacturing apparatus 10 is a 3D printer that manufactures three-dimensional objects using powder bed fusion (PBF). Powder bed fusion is a fabrication method that involves layering powder material laid on a powder bed, curing each layer as it is built up. Powder bed fusion is classified into several methods depending on the type of powder material and curing method, including MJF (Multi Jet Fusion), SLS (Selective Laser Sintering), SLM (Selective Laser Melting), and EBM (Electron Beam Melting).

[0015] The MJF method and the SLS method are methods that use a powder material of resin. The MJF method is a method of melting and curing the powder material to form a shape by applying a melting agent and a fixing agent to the powder material laid on the powder bed and heating it. The SLS method is a method of melting and curing the powder material to form a shape by irradiating an infrared laser to the powder material laid on the powder bed. The SLM method and the EBM method are methods that use a powder material of metal. The SLM method is a method of melting and curing the powder material to form a shape by irradiating a laser to the powder material laid on the powder bed. The EBM method is a method of melting and curing the powder material to form a shape by irradiating an electron beam to the powder material laid on the powder bed. Hereinafter, an example of manufacturing a shaped object by the SLM method in which the additive manufacturing apparatus 10 melts and cures a powder material of metal by laser irradiation will be described. Note that the additive manufacturing apparatus 10 may manufacture a shaped object by the MJF method, the SLS method, or the EBM method.

[0016] As shown in FIG. 1, the additive manufacturing apparatus 10 includes an imaging device 11. The imaging device 11 is provided in the additive manufacturing apparatus 10 so as to be able to photograph the powder bed of the additive manufacturing apparatus 10. Note that the imaging device 11 may be built in the additive manufacturing apparatus 10 or may be attached outside the additive manufacturing apparatus 10. The imaging device 11 images the powder bed after the powder material laid on the powder bed is irradiated with a laser in the manufacture of a three-dimensional shaped object by the additive manufacturing apparatus 10. Thereby, the imaging device 11 images an image (hereinafter, also referred to as a "shaped area image") including the shaped area formed by curing the powder material in the powder bed. In addition, the imaging device 11 images the powder bed on which the powder material is laid after the shaping of the shaping area by laser irradiation in the manufacture of a three-dimensional shaped object by the additive manufacturing apparatus 10. Thereby, the imaging device 11 images an image (hereinafter, also referred to as a "powder bed image") including the powder bed on which the powder material is laid after the shaping of the shaping area. Furthermore, the imaging device 11 captures an image of the fabrication area in each layer being stacked, and then captures an image of the powder bed.

[0017] The additive manufacturing apparatus 10 is connected to the defect detection apparatus 20 via a network NW in a communicable manner. In communication with the defect detection apparatus 20, the additive manufacturing apparatus 10 transmits the manufacturing area image and the powder bed image captured by the imaging apparatus 11.

[0018] (2) Defect detection device 20 The defect detection device 20 is a device for detecting defects in objects manufactured by the additive manufacturing device 10. The defect detection device 20 can be implemented, for example, by a PC (Personal Computer) or a server device.

[0019] The defect detection device 20 is connected to the additive manufacturing apparatus 10 via a network NW in a communicable manner. In communication with the additive manufacturing apparatus 10, the defect detection device 20 receives a build area image and a powder bed image.

[0020] The defect detection device 20 determines whether or not defects have occurred in the build area of ​​each layer based on the received build area image and powder bed image. For example, the defect detection device 20 detects defects occurring in the powder bed as candidates for defects occurring in the build area based on the powder bed image. The defect detection device 20 also detects the build area in the powder bed based on the build area image. The defect detection device 20 determines, based on the detection results of candidate defects and the detection results of the build area, that candidate defects within the detected build area are defects that occurred in the build area. As a result, the defect detection device 20 does not detect defects that occur outside the build object as defects that occur in the build object, thereby reducing false detections in defect detection.

[0021] The defect detection device 20 then outputs an image as an output image, which includes information indicating the detection result or judgment result. At this time, the defect detection device 20 classifies the output images into images that do not require user confirmation (hereinafter also referred to as "OK images") and images that do require confirmation (hereinafter also referred to as "NG images").

[0022] Furthermore, the defect detection device 20 is connected to the user terminal 30 via a network NW so as to be able to communicate with it. The defect detection device 20 transmits output images when communicating with the user terminal 30. When the defect detection device 20 transmits an output image that has been classified as an NG image, it may also transmit a notification indicating that an NG image has been transmitted (hereinafter also referred to as an "NG judgment notification") along with the NG image.

[0023] (3) User terminal 30 The user terminal 30 is a terminal used by the user. The user is, for example, a manufacturer who is operating the additive manufacturing apparatus 10 to produce an object. The user terminal 30 may be implemented as, for example, a PC, smartphone, or tablet device.

[0024] The user terminal 30 is connected to the defect detection device 20 via a network NW, enabling communication. In communication with the defect detection device 20, the user terminal 30 receives at least the output image and, in some cases, also receives an NG judgment notification. The user terminal 30 displays the received output image on its screen. This allows the user to check whether or not there are defects in the printed object (printing area) by examining the displayed output image. Depending on the results of the output image check (whether or not there are defects, the extent of the defects, etc.), the user can stop the manufacturing process in the additive manufacturing apparatus 10.

[0025] Furthermore, when the user terminal 30 receives an NG (Not Finished) notification, it notifies the user of the notification through a screen display or audio output. This allows the user to detect abnormalities in the manufacturing of the printed object, such as defects in the printed object, at an early stage. In addition, the user only needs to check the output image when the user terminal 30 receives an NG notification, eliminating the need to check all output images and reducing the time spent checking output images.

[0026] In this way, users can detect defects in the printed object early and interrupt the printing process early, preventing wasted printing time, inspection time, powder materials, and other resources. In other words, users can reduce costs in additive manufacturing.

[0027] <2. Functional Configuration of the Defect Detection Device> The schematic configuration of the additive manufacturing system 1 according to this embodiment has been described above. Next, the functional configuration of the defect detection device 20 according to this embodiment will be described with reference to Figures 2 to 9. Figure 2 is a block diagram showing an example of the functional configuration of the defect detection device 20 according to this embodiment. As shown in Figure 2, the defect detection device 20 comprises a communication unit 210, a storage unit 220, and a control unit 230.

[0028] (1) Communications Section 210 The communication unit 210 has the function of sending and receiving various types of information. For example, the communication unit 210 receives a build area image and a powder bed image from the additive manufacturing apparatus 10 via the network NW. The communication unit 210 also sends an output image or an NG judgment notification to the user terminal 30 via the network NW. The communication unit 210 may receive learning data from the additive manufacturing device 10 or from the user terminal 30.

[0029] (2) Storage section 220 The storage unit 220 has the function of storing various types of information. The storage unit 220 is composed of storage media provided as hardware by the defect detection device 20, such as HDD (Hard Disk Drive), SSD (Solid State Drive), flash memory, EEPROM (Electrically Erasable Programmable Read Only Memory), RAM (Random Access read / write Memory), ROM (Read Only Memory), or any combination of these storage media. As shown in Figure 2, the storage unit 220 stores the defect detection model 221 and the region detection model 222.

[0030] (2-1) Defect detection model 221 The defect detection model 221 is a pre-trained model that was created using machine learning with images showing defects in the powder bed as training data. Examples of defects include holes, lumps, vertical lines, horizontal lines, insufficient powder, edge deformation, and welts. Holes are defects where holes appear in the powder bed. Lumps are defects where clump-like protrusions form in the powder bed. Vertical lines are defects where lines run vertically in the powder bed. Horizontal lines are defects where lines run horizontally in the powder bed. Insufficient powder is a defect where there is insufficient powder material laid on the powder bed. Edge deformation is a defect where the edges of the printed object are deformed. Welts are defects where welt-like protrusions form in the powder bed. These defects become apparent when powder material is laid on the powder bed after the build area has been constructed.

[0031] (2-2) Region detection model 222 The region detection model 222 is a pre-trained model that has been machine-trained using images representing the shape of the build region as training data.

[0032] (3) Control unit 230 The control unit 230 has the function of controlling the overall operation of the defect detection device 20. The control unit 230 is implemented, for example, by causing the CPU (Central Processing Unit) provided as hardware in the defect detection device 20 to execute a program. As shown in Figure 2, the control unit 230 includes an acquisition unit 231, a learning unit 232, a defect detection unit 233, a region detection unit 234, a defect determination unit 235, and an output control unit 236.

[0033] (3-1) Acquisition section 231 The acquisition unit 231 has the function of acquiring a build area image and a powder bed image for each layer. For example, the acquisition unit 231 acquires the build area image and powder bed image received by the communication unit 210 from the additive manufacturing apparatus 10.

[0034] Here, with reference to Figures 3 to 6, the build area image and powder bed image according to this embodiment will be described. Figure 3 is a diagram showing an example of a build area image without defects according to this embodiment. Figure 4 is a diagram showing an example of a powder bed image without defects according to this embodiment. Figure 5 is a diagram showing an example of a build area image containing defects according to this embodiment. Figure 6 is a diagram showing an example of a powder bed image containing defects according to this embodiment.

[0035] First, let's explain the defect-free image with reference to Figures 3 and 4. Figure 3 shows a defect-free build area image 40. The build area image 40 is an image of the powder bed 41 after the build area 42 and build area 43 have been formed by irradiating the laid powder material with a laser. Figure 4 shows a powder bed image 50 free of defects. The powder bed image 50 is an image of the powder bed 51 after the powder material has been laid, compared to the powder bed 41 shown in Figure 3. In the powder bed 51, the build area 42 and build area 43, which are hidden by the laid powder material, are located at the positions indicated by the dashed lines.

[0036] Next, we will explain images containing defects with reference to Figures 5 and 6. Figure 5 shows a build area image 40a containing defects. The powder bed 41a shown in build area image 40a is a defect in the powder bed 41 shown in build area image 40 of Figure 3. As shown in Figure 5, defects 60, 61, and 62 have occurred in the powder bed 41a. Note that in build area image 40a, the powder material has not yet been laid, and the defects are not yet exposed, so defects 60, 61, and 62 are shown with dashed lines. Figure 6 shows a powder bed image 50a containing defects. The powder bed image 50a is an image of the powder bed 51a after the powder material has been laid on the powder bed 41a in Figure 5. As shown in Figure 6, in the powder bed image 50a, defects 60, 61, and 62, which were not exposed in the build area image 40a in Figure 5, are exposed due to the laying of the powder material. In addition, a defect 63 has occurred in the powder bed 51a. In defect 63, a part of the build area 43 is exposed due to insufficient laid powder material.

[0037] Furthermore, the acquisition unit 231 also has the function of acquiring training data. For example, the acquisition unit 231 acquires training data received by the communication unit 210 from the additive manufacturing apparatus 10 or the user terminal 30.

[0038] (3-2) Learning Department 232 The learning unit 232 has the function of generating a trained model. For example, the learning unit 232 generates a trained model by machine learning using the training data acquired by the acquisition unit 231.

[0039] Specifically, the learning unit 232 uses images showing defects in the powder bed, acquired as training data by the acquisition unit 231, as training data to learn how to detect defects in the powder bed. The learning unit 232 trains a model that performs detection processing using, for example, the Faster R-CNN (Convolutional Neural Network) algorithm. Through this machine learning, when a powder bed image is input, the learning unit 232 detects the defects in the powder bed shown in the powder bed image as candidates for defects in the fabrication area and generates a defect detection model 221 that outputs information about the detected candidate defects. The defect detection model 221 outputs location information (coordinates) indicating the location where the defect candidate occurs, and confidence information indicating the confidence level that the defect candidate is a defect, as information about the defect candidate.

[0040] Furthermore, the learning unit 232 uses images showing the shape of the build area, acquired as training data by the acquisition unit 231, as training data to perform machine learning on the build area in the powder bed. The learning unit 232 performs machine learning on a model that performs detection processing using, for example, the Mask R-CNN algorithm. Through this machine learning, the learning unit 232 generates a region detection model 222 that, when a build area image is input, detects the build area in the powder bed indicated by the build area image and outputs information indicating the detected build area. The region detection model 222 outputs positional information (coordinates) indicating the position of the contour of the detected build region, as information indicating the detected build region.

[0041] (3-3) Defect detection unit 233 The defect detection unit 233 has the function of detecting defects occurring in the powder bed as candidate defects occurring in the fabrication area based on the powder bed image. For example, the defect detection unit 233 detects candidate defects using the defect detection model 221. Specifically, the defect detection unit 233 inputs the powder bed image acquired by the acquisition unit 231 to the defect detection model 221 and outputs information about candidate defects output from the defect detection model 221 as the detection result for candidate defects.

[0042] The defect detection results output by the defect detection unit 233 include, for example, location information and confidence information for each defect candidate. This location information and confidence information can be used to add information to the powder bed image. For example, the location information is used to add an indication to the powder bed image that clearly shows the location of the defect (for example, a frame surrounding the defect). The confidence information is used to add an indication to the powder bed image that clearly shows the reliability of the defect (for example, a numerical value indicating the reliability). Note that the addition of information to the powder bed image is performed not by the defect detection unit 233, but by the defect determination unit 235, which will be described later. For this reason, the defect detection unit 233 does not output the powder bed image.

[0043] Here, with reference to Figure 7, we will describe a powder bed image to which information output from the defect detection unit 233 according to this embodiment has been added. Figure 7 is a diagram showing an example of a powder bed image to which information output from the defect detection unit 233 has been added according to this embodiment. Here, we will describe a powder bed image to which the detection results of candidate defects output from the defect detection unit 233 have been added as an example. Figure 7 shows a powder bed image 70 with the detection results for potential defects added. The powder bed image 70 is the powder bed image 50a from Figure 6 with the detection results for potential defects added. As shown in the powder bed image 70, for defect 60, a frame 71 indicating its location is displayed, and the confidence level of defect 60 is displayed near frame 71 as "98%". For defect 61, a frame 72 indicating its location is displayed, and the confidence level of defect 61 is displayed near frame 72 as "92%". For defect 62, a frame 73 indicating its location is displayed, and the confidence level of defect 62 is displayed near frame 73 as "88%". For defect 63, a frame 74 indicating its location is displayed, and the confidence level of defect 63 is displayed near frame 74 as "93%". In addition, the defect detection unit 235, which will be described later, does not assign detection results for candidate defects to the powder bed image; only the detection results for defects selected from the candidate defects are assigned. Therefore, the powder bed image 70 shown in Figure 7 is not actually output, but for the sake of explanation, the example shown in Figure 7 has been described.

[0044] (3-4) Region detection unit 234 The region detection unit 234 has the function of detecting the build area in the powder bed based on the build area image. For example, the region detection unit 234 detects the build area using the region detection model 222. Specifically, the region detection unit 234 receives the build area image acquired by the acquisition unit 231 as input to the region detection model 222 and outputs information indicating the build area output from the region detection model 222 as the build area detection result.

[0045] The detection result of the build area output by the region detection unit 234 is, for example, position information indicating the position of the outline of the build area. This position information can be used to add information to the powder bed image. For example, the position information indicating the position of the outline of the build area is used to add an indication (e.g., the outline of the build area) to the powder bed image to clearly show the position of the build area. Note that the addition of information to the powder bed image is performed not by the region detection unit 234, but by the defect determination unit 235, which will be described later. For this reason, the region detection unit 234 does not output the powder bed image.

[0046] Here, with reference to Figure 8, we will describe a build region image to which information output from the region detection unit 234 according to this embodiment has been added. Figure 8 is a diagram showing an example of a build region image to which information output from the region detection unit 234 has been added according to this embodiment. Here, we will describe a build region image to which the build region detection result output from the region detection unit 234 has been added as an example. Figure 8 shows a build area image 80 with the build area detection results added. The build area image 80 is the build area image 40a in Figure 5 with the build area detection results added. As shown in the build area image 80, the outline 81 of the build area 42 is displayed. Also, the outline 82 of the build area 43 is displayed. In addition, the defect detection unit 235, which will be described later, does not assign the detection result of the build area to the build area image, but rather assigns the detection result of the build area to the powder bed image. For this reason, the build area image 80 shown in Figure 8 is not actually output, but for the sake of explanation, the example shown in Figure 8 has been described.

[0047] (3-5) Defect detection unit 235 The defect determination unit 235 has the function of determining that a candidate defect located within the detected build area is a defect that occurred in the build area, based on the detection results of candidate defects and the detection results of the build area. For example, the defect determination unit 235 detects candidate defects within the build area by performing calculations based on the detection results of candidate defects output by the defect detection unit 233 and the detection results of the build area output by the area detection unit 234. In this calculation, for example, it is determined whether or not each candidate defect is located within the build area by comparing the position information of each candidate defect indicated by the candidate defect detection results with the position information indicating the position of the outline of the build area indicated by the build area detection results. Thus, in the defect detection device 20 according to this embodiment, by using a combination of the defect detection model 221 and the region detection model 222, based on the output from each model, only defects detected within the build area are ultimately detected as defects, and defects detected outside the build area are ultimately not detected as defects. As a result, the defect detection device 20 can reduce false detections in defect detection.

[0048] The defect detection unit 235 outputs an output image that includes information about defects occurring in the build area. For example, the defect detection unit 235 adds location information indicating the location of the defect and confidence information indicating the confidence level of the defect to the powder bed image used to detect defect candidates, and outputs an output image. Specifically, the defect detection unit 235 adds an indication (e.g., a frame surrounding the defect) to the powder bed image based on the location information of the defect candidate determined to be a defect. The defect detection unit 235 also adds an indication (e.g., a numerical value indicating the confidence level) to the powder bed image based on the confidence level information of the defect candidate determined to be a defect. The defect detection unit 235 may also add information indicating the build area to the output image.

[0049] The defect detection unit 235 classifies and outputs output images based on reliability information into two categories: OK images (where no unacceptable defects are present) and NG images (where at least one unacceptable defect is present). Unacceptable defects are those within the build area whose degree of defect has a significant impact on the quality of the build object and cannot be ignored. On the other hand, acceptable defects are those within the build area whose degree of defect has a small impact on the quality of the build object and can be ignored, or those detected as defects but are not actually defects (e.g., shadows). Specifically, the defect determination unit 235 compares a threshold value, which is set as the boundary between the reliability of acceptable defects and the reliability of unacceptable defects, with a numerical value indicating the reliability of each defect assigned to the output image, to determine whether each defect is an acceptable or unacceptable defect. The threshold value is set, for example, so that output images with unacceptable defects are reliably classified as NG images. If the judgment determines that all defects shown in the output image are acceptable defects, the defect judgment unit 235 classifies the output image as an OK image. On the other hand, if even one defect in the output image is unacceptable, the defect judgment unit 235 classifies the output image as an NG image. This allows the user to easily identify output images that require visual inspection (NG images) from the output images transmitted from the defect detection device 20 to the user terminal 30. Furthermore, by visually inspecting only the output images classified as NG images, the user does not need to inspect all output images, thus reducing the time required for visual inspection. The defect detection unit 235 may output the information to be added to the NG image without adding it to the NG image.

[0050] Now, with reference to Figure 9, the output image according to this embodiment will be described. Figure 9 is a diagram showing an example of the output image according to this embodiment. Figure 9 shows an output image 90 with information about defects in the build area and information indicating the build area added. Output image 90 is an image of the powder bed image 50a in Figure 6 with information about defects in the build area and information indicating the build area added. As shown in output image 90, contour 81 (contour of build area 42) and contour 82 (contour of build area 43) are displayed as information indicating the build area. In addition, for defect 60 within contour 82, a frame 71 indicating its location and a confidence level of "98%" are displayed. In addition, for defect 63 within contour 82, a frame 74 indicating its location and a confidence level of "93%" are displayed.

[0051] (3-6) Output control unit 236 The output control unit 236 has the function of controlling various outputs. For example, the output control unit 236 transmits the output image output by the defect determination unit 235 to the user terminal 30 via the communication unit 210. Furthermore, when the output control unit 236 outputs an output image classified as an NG image by the defect determination unit 235 to the user terminal 30, it also sends an NG determination notification to the user terminal 30 along with the output image.

[0052] <3. Processing Flow> The functional configuration of the defect detection device 20 according to this embodiment has been described above. Next, the processing flow in the additive manufacturing system 1 according to this embodiment will be described with reference to Figure 10. Figure 10 is a sequence diagram showing an example of the processing flow in the additive manufacturing system 1 according to this embodiment. In the sequence diagram shown in Figure 10, it is assumed that the defect detection model 221 and the region detection model 222 have already been generated and are stored in the storage unit 220 of the defect detection device 20. Therefore, the explanation of the generation process for each model will be omitted.

[0053] As shown in Figure 10, first, the additive manufacturing apparatus 10 irradiates a laser onto the position of the powder material to be manufactured, which is laid on the powder bed (step S101). Next, the additive manufacturing apparatus 10 captures an image of the manufacturing area formed by the melting and hardening of the powder material by laser irradiation using the imaging device 11 before laying the powder material (step S102). Next, the additive manufacturing apparatus 10 lays powder material on the powder bed where the manufacturing area has been created by laser irradiation (step S103). Next, the additive manufacturing apparatus 10 captures an image of the powder bed on which the powder material is laid using the imaging device 11 before irradiating it with a laser (step S104). Next, the additive manufacturing apparatus 10 transmits the manufacturing area image and powder bed image captured by the imaging device 11 to the defect detection device 20 (step S105). After each image is transmitted, the additive manufacturing apparatus 10 checks whether the fabricated area is the final layer (step S106). If it is not the final layer (step S106 / NO), the additive manufacturing apparatus 10 repeats the process from step S101. In this way, the additive manufacturing apparatus 10 captures an image of the fabricated area and an image of the powder bed for each layer being added and transmits them to the defect detection device 20 for each layer. On the other hand, if it is the final layer (step S106 / YES), the additive manufacturing apparatus 10 proceeds to step S107. If the process proceeds to step S107, the additive manufacturing apparatus 10 terminates the manufacturing of the object (step S107).

[0054] The defect detection device 20 starts processing triggered by the reception of each image from the additive manufacturing device 10. Therefore, if the communication unit 210 receives each image from the additive manufacturing device 10 (step S108 / YES), the defect detection device 20 proceeds to step S109. On the other hand, if the communication unit 210 has not received each image from the additive manufacturing device 10 (step S108 / NO), the defect detection device 20 does not proceed with processing.

[0055] If the process proceeds to step S109, the acquisition unit 231 acquires the powder floor image from the images received by the communication unit 210 (step S109). The defect detection unit 233 performs defect detection processing based on the powder bed image acquired by the acquisition unit 231 (step S110). Specifically, the defect detection unit 233 outputs information about candidate defects, which is output by inputting the powder bed image to the defect detection model 221, as the detection result for candidate defects.

[0056] Next, the acquisition unit 231 acquires the build area image from the image received by the communication unit 210 (step S111). The region detection unit 234 performs region detection processing based on the build region image acquired by the acquisition unit 231 (step S112). Specifically, the region detection unit 234 outputs information indicating the build region as the build region detection result, which is output by inputting the build region image to the region detection model 222.

[0057] Next, the defect determination unit 235 performs defect determination processing (step S113). Specifically, based on the detection results of candidate defects output from the defect detection unit 233 and the detection results of the build region output from the region detection unit 234, the defect determination unit 235 determines that the candidate defects within the detected build region are defects that have occurred in the build region. After determination, the defect determination unit 235 outputs an output image with information about the defects that it determined to be defects that have occurred in the build region. Furthermore, based on the confidence information, the defect determination unit 235 classifies the output image into an OK image or an NG image and outputs it.

[0058] Next, the output control unit 236 determines whether the output image output by the defect determination unit 235 is an NG image (step S114). If the output image is an NG image (step S114 / YES), the output control unit 236 proceeds to step S115. On the other hand, if the output image is an OK image (step S114 / NO), the output control unit 236 proceeds to step S116.

[0059] If the process proceeds to step S115, the output control unit 236 sends an NG judgment notification to the user terminal 30 via the communication unit 210 (step S115). After sending, the output control unit 236 proceeds the process to step S116. If the process proceeds to step S116, the output control unit 236 transmits the output image to the user terminal 30 via the communication unit 210 (step S116). The output control unit 236 transmits the output image to the user terminal 30 regardless of whether the output image is an OK image or an NG image.

[0060] After the output image is transmitted, the defect detection device 20 repeats the process from step S108. This allows the defect detection device 20 to determine whether a defect has occurred in the build area each time it receives a build area image and a powder bed image from the additive manufacturing device 10. As described above, the defect detection device 20 receives build area images and powder bed images for each layer from the additive manufacturing device 10. Therefore, the defect detection device 20 can determine whether a defect has occurred in each layer of the object being manufactured by the additive manufacturing device 10.

[0061] The user terminal 30 performs output processing triggered by the receipt of an NG judgment notification or an output image from the defect detection device 20 (step S117). If an NG judgment notification is received from the defect detection device 20, the user terminal 30 outputs information related to the NG judgment notification. If an output image is received from the defect detection device 20, the user terminal 30 displays the output image regardless of whether it is an OK image or an NG image.

[0062] As described above, the defect detection device 20 according to this embodiment acquires a build area image taken after laser irradiation and a powder bed image taken after laser irradiation but before the powder material is laid, for each layer, during the manufacturing of a three-dimensional object by the additive manufacturing device 10. The defect detection device 20 also detects defects occurring in the powder bed as candidates for defects occurring in the build area based on the acquired powder bed image, and detects the build area in the powder bed based on the acquired build area image. Furthermore, the defect detection device 20 determines that the candidate defects within the detected build area are defects that have occurred in the build area, based on the detection results of the candidate defects and the build area.

[0063] With this configuration, the defect detection device 20 according to this embodiment ultimately detects only defects detected within the build area as defects, and does not ultimately detect defects detected outside the build area as defects. Therefore, the defect detection device 20 according to this embodiment makes it possible to reduce false detection of defects in additive manufacturing.

[0064] <4. Variation> Embodiments of the present invention have been described above. Next, modifications of the embodiments of the present invention will be described. Each modification described below may be applied to the embodiments of the present invention individually or in combination. Furthermore, each modification may be applied in place of the configuration described in the embodiments of the present invention, or it may be applied in addition to the configuration described in the embodiments of the present invention.

[0065] In the above-described embodiment, an example was explained in which processing using each model is performed in the defect detection device 20 based on each image received from the additive manufacturing device 10, but the system is not limited to this example. For example, the additive manufacturing device 10 may be provided with a configuration having the same functions as the storage unit 220 and control unit 230 of the defect detection device 20, so that the additive manufacturing device 10 can perform the processing performed by the defect detection device 20. This allows the additive manufacturing device 10 to determine whether or not a defect has occurred in the manufacturing area using only its own device. In this case, the additive manufacturing system 1 can be configured to consist only of the additive manufacturing device 10 and the user terminal 30. Furthermore, by providing the additive manufacturing apparatus 10 with a configuration that allows output of output images and NG judgment notifications, the additive manufacturing system 1 may be configured to consist only of the additive manufacturing apparatus 10.

[0066] Furthermore, although the above-described embodiment explains an example in which the defect detection device 20 generates the defect detection model 221 and the region detection model 222 using the functions of the learning unit 232, the invention is not limited to such an example. For example, the functions of the learning unit 232 may be implemented by a device other than the defect detection device 20 (such as a PC or server device).

[0067] Furthermore, in the above-described embodiment, a defect detection device 20 performs defect detection in each layer while the additive manufacturing device 10 is manufacturing the object, and the user visually checks each time a defect is detected to decide whether to continue or interrupt manufacturing. However, the embodiment is not limited to this example. For example, the defect detection device 20 performs defect detection in each layer while the additive manufacturing device 10 is manufacturing the object, but the user may perform the visual check after the manufacturing of the object is complete. In this case, the user can visually check only the NG images from the output images output from the defect detection device 20 all at once. Therefore, compared to the case where a visual check is performed each time an NG image is output, as in the above-described embodiment, the time and effort required for visual check can be reduced.

[0068] Furthermore, in the above-described embodiment, an example was explained in which the learning unit 232 generates a defect detection model 221 by machine learning using images showing defects in the powder bed as training data, but the invention is not limited to such an example. For example, the learning unit 232 may generate a defect detection model 221 by machine learning using training data in which information indicating the type of defect is attached to images showing defects in the powder bed. This allows the defect detection model 221 to also output information indicating the type of defect detected. The defect detection unit 233 can use the defect detection model 221 to output information about candidate defects, including information indicating the type of defect, as the detection result for candidate defects. This allows the defect determination unit 235 to also attach information indicating the type of defect to the output image. The user can then decide whether to continue or stop manufacturing the molded object, taking into account the type of defect attached to the output image. Furthermore, the learning unit 232 may generate a defect detection model 221 by machine learning using training data in which images showing defects in the powder bed are assigned information indicating the type of defect and information indicating a further classification of that type. This enables the defect detection model 221 to also output information indicating the type and category of the detected defect. The defect detection unit 233 can use the defect detection model 221 to output information about candidate defects, including information indicating the type and category of the defect, as the detection result for candidate defects. This enables the defect determination unit 235 to also assign information indicating the type and category of the defect to the output image. The user can then decide whether to continue or discontinue the manufacturing of the molded object, taking into account the type and category of the defect assigned to the output image. In this way, by increasing the amount of information about defects in the training data used to generate the defect detection model 221, the amount of information about defects attached to the output image can be increased, enabling users to make more accurate judgments.

[0069] Furthermore, the defect detection unit 235 described in the above embodiment may have a function to group similar defects in multiple layers and output an output image with grouping information attached when similar defects are detected in the fabrication area of ​​multiple layers. The definition (annotation) of the grouping is set by the user. The grouping information may include, for example, information indicating annotation, information indicating whether the detected defect was detected in any other layer, or information indicating that similar defects are continuously being detected. As a result, the user can determine that for defects in the NG image with grouping information attached, similar defects have already been visually confirmed in other layers. As a result, the user can omit visual confirmation of defects with grouping information attached, and the time required for visual confirmation of the NG image can be reduced.

[0070] Furthermore, the learning unit 232 described in the above-described embodiment may also have a retraining function. For example, when a new annotation is set by the user, the learning unit 232 regenerates the defect detection model 221 by retraining based on the said annotation. This improves the accuracy of defect candidate detection by the defect detection model 221.

[0071] Embodiments of the present invention have been described above. It should be noted that some or all of the defect detection device in the above-described embodiments may be implemented using a computer. In that case, the program for implementing this function may be recorded on a computer-readable recording medium, and the program recorded on this recording medium may be loaded into a computer system and executed. Here, "computer system" includes hardware such as an OS and peripheral devices. "Computer-readable recording medium" refers to portable media such as flexible disks, magneto-optical disks, ROMs, CD-ROMs, and storage devices such as hard disks built into a computer system. Furthermore, "computer-readable recording medium" may also include those that dynamically hold programs for a short period, such as communication lines used when transmitting programs via networks such as the Internet or communication lines such as telephone lines, and those that hold programs for a certain period, such as volatile memory inside a computer system acting as a server or client. The program may also be for implementing some of the functions described above, or it may be a program that can implement the functions described above in combination with a program already recorded in the computer system, or it may be implemented using a programmable logic device such as an FPGA (Field Programmable Gate Array).

[0072] Although embodiments of this invention have been described in detail above with reference to the drawings, the specific configuration is not limited to those described above, and various design changes can be made without departing from the spirit of this invention.

[0073] (Note 1) In the manufacturing of a three-dimensional object using an additive manufacturing apparatus, an acquisition unit acquires, for each layer, a manufacturing area image captured of the manufacturing area formed by the hardening of powder material laid on the powder bed, and a powder bed image captured of the powder bed after the manufacturing of the manufacturing area and the laying of the powder material. A defect detection unit detects defects occurring in the powder bed as candidates for defects occurring in the molding area based on the powder bed image, A region detection unit detects the molding region in the powder bed based on the molding region image, A defect determination unit determines, based on the detection results of the candidate defects and the detection results of the molding region, that the candidate defects found within the detected molding region are defects that have occurred in the molding region. A defect detection device equipped with the following features.

[0074] (Note 2) The defect detection unit inputs the acquired powder bed image to a defect detection model that has been trained using images showing defects in the powder bed as training data, and obtains information about the candidate defects output from the defect detection model as the detection result of the candidate defects. The defect detection device described in Appendix 1.

[0075] (Note 3) The defect detection unit obtains the detection results of candidate defects using the defect detection model, which has been trained using the training data to which information indicating the type of defect has been added. The defect detection device described in Appendix 2.

[0076] (Note 4) The region detection unit inputs the acquired image of the build region to a region detection model that has been trained using images showing the shape of the build region as training data, and acquires the information indicating the build region output from the region detection model as the detection result of the build region. A defect detection device as described in any one of the appendices 1 to 3.

[0077] (Note 5) The defect detection unit outputs an output image that includes location information indicating the location where the defect is located and confidence information indicating the confidence level that the defect is indeed a defect. A defect detection device as described in any one of the appendices 1 through 4.

[0078] (Note 6) The defect determination unit classifies and outputs output images in which no unacceptable defects have occurred and output images in which at least one unacceptable defect has occurred, based on the reliability information. The defect detection device described in Appendix 5.

[0079] (Note 7) When the defect detection unit detects defects similar to those in the fabricated area in multiple layers, it groups the similar defects and outputs an output image with information about the grouping added. A defect detection device as described in any one of the appendices 1 through 6. [Explanation of Symbols]

[0080] 1…Additive manufacturing system, 10…Additive manufacturing device, 11…Imaging device, 20…Defect detection device, 30…User terminal, 210…Communication unit, 220…Storage unit, 221…Defect detection model, 222…Region detection model, 230…Control unit, 231…Acquisition unit, 232…Learning unit, 233…Defect detection unit, 234…Region detection unit, 235…Defect determination unit, 236…Output control unit, NW…Network

Claims

1. In the manufacturing of a three-dimensional object using an additive manufacturing apparatus, an acquisition unit acquires, for each layer, a manufacturing area image captured of the manufacturing area formed by the hardening of powder material laid on the powder bed, and a powder bed image captured of the powder bed after the manufacturing of the manufacturing area and the laying of the powder material. A defect detection unit detects defects occurring in the powder bed as candidates for defects occurring in the molding area based on the powder bed image, A region detection unit detects the molding region in the powder bed based on the molding region image, A defect determination unit determines, based on the detection results of the candidate defects and the detection results of the molding region, that the candidate defects found within the detected molding region are defects that have occurred in the molding region. A defect detection device equipped with the following features.

2. The defect detection unit inputs the acquired powder bed image to a defect detection model that has been trained using images showing defects in the powder bed as training data, and obtains information about the candidate defects output from the defect detection model as the detection result of the candidate defects. The defect detection device according to claim 1.

3. The defect detection unit obtains the detection results of candidate defects using the defect detection model, which has been trained using the training data to which information indicating the type of defect has been added. The defect detection device according to claim 2.

4. The region detection unit inputs the acquired image of the build region to a region detection model that has been trained using images showing the shape of the build region as training data, and obtains the information indicating the build region output from the region detection model as the detection result of the build region. The defect detection device according to claim 1.

5. The defect detection unit outputs an output image that includes location information indicating the location where the defect is located and confidence information indicating the confidence level that the defect is indeed a defect. The defect detection device according to claim 1.

6. The defect determination unit classifies and outputs output images based on the reliability information into those in which no unacceptable defects have occurred and those in which at least one unacceptable defect has occurred. The defect detection device according to claim 5.

7. When the defect detection unit detects defects similar to those in the fabricated area in multiple layers, it groups the similar defects and outputs an output image with information about the grouping added. A defect detection device according to claim 1 or claim 6.

8. The acquisition unit acquires, for each layer, an image of the fabricated area, which is captured when the powder material laid on the powder bed hardens during the manufacturing of a three-dimensional object by an additive manufacturing device, and an image of the powder bed, which is captured after the fabrication of the fabricated area and the subsequent laying of the powder material. A defect detection process in which the defect detection unit detects defects occurring in the powder bed as candidates for defects occurring in the molding area based on the powder bed image, The region detection unit performs a region detection process to detect the molding region in the powder bed based on the molding region image, A defect determination process in which the defect determination unit determines, based on the detection result of the candidate defect and the detection result of the molding area, that the candidate defect located within the detected molding area is a defect that occurred in the molding area. A defect detection method including the following.

9. Computers, In the manufacturing of a three-dimensional object using an additive manufacturing apparatus, an acquisition means is provided to acquire, for each layer, a manufacturing area image captured of the manufacturing area formed by the hardening of powder material laid on the powder bed, and a powder bed image captured of the powder bed after the manufacturing of the manufacturing area and the laying of the powder material. A defect detection means that detects defects occurring in the powder bed as candidates for defects occurring in the molding area based on the powder bed image, A region detection means for detecting the molding region in the powder bed based on the molding region image, A defect determination means that determines, based on the detection results of the candidate defects and the detection results of the molding region, that the candidate defects found within the detected molding region are defects that have occurred in the molding region. A program designed to function as such.

Images