Method for detecting defects in a 3D printer

Abstract

Computer-implemented method for detecting defects in a 3D printer, wherein the method comprises the steps of: a) capturing a first image of a build space of the 3D printer, wherein the build space comprises a 3D printed part shown in the first image; b) generating a second image having a higher spatial resolution than the first image from the first image by using a spatial resolution increasing artificial neural network.

Description

[0001] The present invention relates to a computer-implemented method for detecting defects in a 3D printer, a data processing unit, a 3D printer, a computer program, and a computer-readable data carrier.

[0002] In 3D printers, parts are manufactured layer by layer by selectively melting or sintering a powder bed. The powder in the powder bed can be metallic powder, and the powder can be melted or sintered using a laser or electron beam. Defects may occur in the part during manufacturing. For example, defects may occur when the coater does not completely deposit the powder bed layer. Hot spots may form during part manufacturing, which can lead to oxidation of the part and / or the formation of a rough surface. Hot spots can also cause deformation of the part, which may then lead to the lifting and / or lowering of adjacent parts in the powder bed. On the other hand, cold spots may form during part manufacturing, where hot spots may cause incomplete melting of the powder, which may result in porosity in the part.

[0003] 3D printers may include camera devices suitable for monitoring part manufacturing. These camera devices may be able to detect some of the defects mentioned above, but may not be able to detect the others. For example, a camera device may be able to detect defects in the powder bed, but may not be able to detect defects in the part itself.

[0004] Therefore, the object of the present invention is to provide a method for detecting defects in a 3D printer, wherein the method achieves improved defect detection.

[0005] A computer-implemented method for detecting defects in a 3D printer includes the steps of: a) capturing a first image of the construction space of the 3D printer, wherein the construction space includes 3D printed parts shown in the first image; b) generating a second image from the first image with a higher spatial resolution than the first image by using an artificial neural network that increases spatial resolution.

[0006] A computer-implemented method for detecting defects in a 3D printer, according to a second invention, includes the following steps: a) capturing a sequence of images of the construction space of the 3D printer, wherein the construction space includes 3D printed parts shown in the images in the sequence; a1) combining the images in the sequence into a first image; b) generating a second image with a higher spatial resolution than the first image from the first image by using an artificial neural network that increases spatial resolution.

[0007] By increasing the spatial resolution, defects in components that may not be visible in the first image can be detected in the second image. Therefore, improved defect detection is provided.

[0008] The spatial resolution-enhancing artificial neural network (ALN) comprises an input layer and an output layer. A first image is arranged in the input layer, and the output layer produces a second image. The ANN can be trained by providing a first set of images with low spatial resolution and a second set of corresponding images with high spatial resolution. The first and second sets can be generated, for example, by capturing images of multiple objects, each captured at both low and high spatial resolutions. Alternatively, a low-resolution image can be computed from each image in the second set to generate the first set. Publicly available datasets can also be obtained for both the first and second sets. The ANN is then trained by adjusting the weights of the ANN such that each output layer approximates a corresponding image in the second set when the images in the first set are used as input layers.

[0009] In the second method, the sequence is preferably performed during part manufacturing. For example, images in the sequence can be captured at a repetition rate of at least 1 kHz. For this purpose, the 3D printer may include a camera device, for example, a CMOS camera device or an sCMOS camera device. Thus, the first image contains information about the progress of part manufacturing. In step a1) of the second method, the images in the sequence can be combined, for example, by adding the intensity values ​​of the same pixels in each of the images in the sequence for each pixel.

[0010] Preferably, the method includes the following steps: a2) increasing the contrast of a first image by using a contrast-enhancing artificial neural network, wherein the first image with increased contrast is used in step b). By increasing the contrast, defects in parts and / or powder beds that may not be visible in the first image can be detected in the second image. Therefore, improved defect detection is provided. By increasing the contrast before increasing the spatial resolution, fewer pixels must be processed by the contrast-enhancing artificial neural network, as is the case when increasing the contrast after increasing the spatial resolution. Therefore, the processing time is shorter.

[0011] Step a2) may also include adjusting the brightness of the first images. Some first images may not be adequately illuminated, so the contrast-increasing artificial neural network increases the brightness. Other first images may be over-illuminated, so the contrast-increasing artificial neural network decreases the brightness.

[0012] The method preferably includes the following steps: c) increasing the contrast of the second image by using a contrast-enhancing artificial neural network. Alternatively, in step c), the contrast in the first image is first determined, and the contrast of the second image is increased only if the determined contrast is below a threshold. In this way, processing time can be reduced.

[0013] Preferably, the contrast-enhancing artificial neural network has an input layer and an output layer, and is trained by providing a first set of images with low contrast and a second set of corresponding images with high contrast, and by adjusting the weights of the contrast-enhancing artificial neural network, such that when the images in the first set are used as input layers, each histogram of the output layer approximates the histogram of the corresponding image in the second set. By training the histograms instead of the images themselves, training the contrast-enhancing artificial neural network requires less processing time. A histogram counts how many pixels in an image are measured to have intensities arranged within a specific intensity range. A histogram may, for example, include a graph with a horizontal axis and a vertical axis, plotting intensities divided into several intensity ranges on the horizontal axis and the number of pixels for each intensity range on the vertical axis. The first set of images and the second set of images may also differ in brightness. For example, the first set of images may be under-illuminated or over-illuminated.

[0014] Alternatively, the contrast-enhancing artificial neural network has an input layer and an output layer, and is trained by providing a first set of images with original images and a second set of corresponding images as gamma-corrected original images, and by adjusting the weights of the contrast-enhancing artificial neural network, such that when the images in the first set are used as input layers, each image of the output layer approximates the gamma-corrected image of the corresponding image in the second set.

[0015] The method preferably includes the following steps: d1) detecting potential defects in the powder bed surrounding the part in a first image. Typically, defects in the powder bed can be detected in images with relatively low resolution. By detecting defects in the first image instead of the second image, fewer pixels must be processed, resulting in a shorter processing time.

[0016] Preferably, the method includes the following steps: d2) detecting defects that may occur in the component and optionally in the powder bed surrounding the component in the second image.

[0017] Preferably, in step d1) and / or step d2), defects are detected by image processing methods, and in particular by machine learning methods.

[0018] Image processing and machine learning methods may include the following steps: (i) providing an image dataset and applying principal component analysis to the image data to compute a large number of image clusters; (ii) applying a clustering algorithm to the analyzed image data and computing the corresponding cluster centroids; (iii) comparing the computed cluster centroids with a set of reference anomaly centroids, wherein, based on the matching of the cluster centroids with the references, the image data is segmented layer by layer into clustered images with specific anomalies; (iv) transforming the segmented images into a defined color space, such as the Lab color space or grayscale spectrum; and (v) integrating the pixel information of the transformed segmented clustered images to compute a threshold for the image dataset, thereby identifying the corresponding anomalies.

[0019] The method preferably includes the following step: e) classifying defects. This includes, for example, distinguishing between powder bed defects, hot spots, cold spots, at least partially porous components, components with at least one void, and / or components with at least one non-molten region.

[0020] Preferably, steps a) and b) and optional steps a1), a), c), d1), d2) and / or e) are performed during component manufacturing. Alternatively, preferably, steps a) and b) and optional steps a1), a), c), d1), d2) and / or e) are performed after component manufacturing.

[0021] The data processing apparatus according to the invention includes a processor adapted to perform the steps of the method.

[0022] The 3D printer according to the invention includes a data processing device, a construction space, and an imaging device adapted to capture a first image and / or a sequence of images. For the first inventive method, the imaging device may be adapted, for example, to capture images at a repetition rate of 1 Hz to 60 Hz. For this purpose, the 3D printer may include an imaging device, for example, a CCD imaging device. The 3D printer is preferably adapted to perform the method steps of the first inventive method or its preferred embodiment on each of the images captured by the imaging device. For the second inventive method, the imaging device may be adapted, for example, to capture images at a repetition rate of at least 1 kHz.

[0023] The computer program according to the present invention includes instructions that, when executed by a computer, cause the computer to perform the steps of the method.

[0024] A computer program is stored on a computer-readable data carrier according to the present invention.

[0025] The invention will be described below based on flowcharts.

[0026] Figure 1 A first exemplary flowchart is shown.

[0027] Figure 2 A second exemplary flowchart is shown.

[0028] Figure 3 A third exemplary flowchart is shown.

[0029] according to Figure 1 and Figure 2 A computer-implemented method for detecting defects in a 3D printer includes the following steps: a) capturing a first image 1 of the construction space of the 3D printer, wherein the construction space includes 3D printed parts shown in the first image; b) generating a second image 2 from the first image with a higher spatial resolution than the first image by using an artificial neural network that increases spatial resolution. The 3D printer may include a camera device.

[0030] according to Figure 3 A computer-implemented method for detecting defects in a 3D printer includes the following steps: a) capturing a sequence 11 of images of the construction space of the 3D printer, wherein the construction space includes 3D printed parts shown in the images in the sequence; a1) combining the images 12 in the sequence into a first image; b) generating a second image 2 from the first image with a higher spatial resolution than the first image by using an artificial neural network that increases spatial resolution.

[0031] 3D printers may include camera devices. For those based on Figure 1 and Figure 2 The method allows the camera device to capture a first image at a repetition rate of 1 Hz to 60 Hz, and step b) can be performed on each of the first images captured in step a). Figure 1 and Figure 2 The camera device can be, for example, a CCD camera device. For those based on... Figure 3 The method allows the camera device to capture a first image at a repetition rate of at least 1 kHz to 60 Hz, and step b) can be performed on each of the first images generated in step a1). Figure 3 The camera device can be, for example, a CMOS camera device or an sCMOS camera device.

[0032] Prior to step a), the method may include a pre-calibration step 7 for the camera device, such as... Figure 1 and Figure 2 As exemplarily indicated in the diagram. In pre-calibration step 7, each pixel of both the first and second images is assigned to the corresponding position in the 3D printer.

[0033] Figure 1The method is shown to include the following steps: c) increasing the contrast of the second image by using a contrast-enhancing artificial neural network. The contrast-enhancing artificial neural network may have an input layer and an output layer, and is trained by providing a first set of images with low contrast and a second set of corresponding images with high contrast, and by adjusting the weights of the contrast-enhancing artificial neural network such that when the images in the first set are used as input layers, each histogram of the output layer approximates the histogram of the corresponding image in the second set.

[0034] from Figure 1 As can be seen from the image, the method may include the following steps: d2) detecting potential defects 5 in the part and / or the powder bed surrounding the part in the second image. Step d2) can be performed, for example, by visual inspection performed by the operator of the 3D printer. Alternatively, defects can be detected in step d2) by image processing methods, and particularly by machine learning methods.

[0035] Figure 1 The method is also shown to include the following steps: e) classifying defects 6. This includes, for example, distinguishing between powder bed defects, hot spots, cold spots, at least partially porous components, components with at least one void, and / or components with at least one non-molten region.

[0036] from Figure 1 As can be seen from the image, the method may include the following steps: d1) detecting possible defects 4 in the powder bed surrounding the part in the first image. Then, the information obtained from steps d1), d2), and / or e) can be used in step f), in particular by using the assignment performed in the pre-calibration step to locate and visualize the defects 8.

[0037] Figure 2 The method is shown to include the following steps: a2) increasing the contrast of a first image 3 by using a contrast-enhancing artificial neural network, wherein the first image with increased contrast is used in step b). Additionally, the method may include the following step: d2) detecting defects 5 that may appear in the part and / or the powder bed surrounding the part in a second image. Step d2) can be performed, for example, by visual inspection performed by the operator of the 3D printer. Alternatively, defects can be detected in step d2) by image processing methods, and particularly by machine learning methods.

[0038] Steps a) and b) and optional steps c), d) and / or e) can be performed during component manufacturing. Alternatively, steps a) and b) and optional steps c), d) and / or e) can be performed after component manufacturing.

Claims

1. A computer-implemented method for detecting defects in a 3D printer, wherein, The method includes the following steps: a) Capture a first image (1) of the construction space of the 3D printer, wherein the construction space includes the 3D printed parts shown in the first image; b) Generating a second image (2) from the first image with a higher spatial resolution than the first image by using an artificial neural network that increases spatial resolution, and wherein the method includes the following steps: d1) Detect possible defects in the powder bed surrounding the component in the first image (4). The method further includes the following steps: a1) Increase the contrast of the first image by using a contrast-enhancing artificial neural network (3), wherein the first image with increased contrast is used in step b), and The contrast-enhancing artificial neural network has an input layer and an output layer. It is trained by providing a first set of images with low contrast and a second set of corresponding images with high contrast, and by adjusting the weights of the contrast-enhancing artificial neural network, such that when the images in the first set are used as the input layer, each histogram of the output layer approximates the histogram of the corresponding image in the second set.

2. The method according to claim 1, wherein, The method further includes the following steps: c) Increase the contrast of the second image by using the contrast-enhancing artificial neural network (3).

3. The method according to claim 1 or 2, wherein, The method further includes the following steps: d2) Detect defects that may occur in the component and / or in the powder bed surrounding the component in the second image (5).

4. The method according to claim 3, wherein, In step d1) and / or step d2), the defect is detected by image processing methods.

5. The method according to claim 3, wherein, In step d1) and / or step d2), the defect is detected by machine learning methods.

6. The method according to claim 3, wherein, The method further includes the following steps: e) Classify the defects (6).

7. The method according to claim 6, wherein, During the manufacture of the component, steps a) and b) are performed, as well as at least one of steps a1), c), d1), d2), and e).

8. The method according to claim 3, wherein, After the component is manufactured, perform at least one of steps a) and b), and steps a1), c), d1), and d2).

9. A computer-implemented method for detecting defects in a 3D printer, wherein, The method includes the following steps: a) A sequence (11) of images capturing the construction space of the 3D printer, wherein the construction space includes 3D printed parts shown in the images in the sequence; a1) Combine the images in the sequence into a first image (12); b) Generating a second image (2) from the first image with a higher spatial resolution than the first image by using an artificial neural network that increases spatial resolution, and wherein the method includes the following steps: d1) Detect possible defects in the powder bed surrounding the component in the first image (4). The method further includes the following steps: a2) Increase the contrast of the first image by using a contrast-enhancing artificial neural network (3), wherein the first image with increased contrast is used in step b), and The contrast-enhancing artificial neural network has an input layer and an output layer. It is trained by providing a first set of images with low contrast and a second set of corresponding images with high contrast, and by adjusting the weights of the contrast-enhancing artificial neural network, such that when the images in the first set are used as the input layer, each histogram of the output layer approximates the histogram of the corresponding image in the second set.

10. The method according to claim 9, wherein, The method further includes the following steps: c) Increase the contrast of the second image by using the contrast-enhancing artificial neural network (3).

11. The method according to claim 9 or 10, wherein, The method further includes the following steps: d2) Detect defects that may occur in the component and / or in the powder bed surrounding the component in the second image (5).

12. The method according to claim 11, wherein, In step d1) and / or step d2), the defect is detected by image processing methods.

13. The method according to claim 11, wherein, In step d1) and / or step d2), the defect is detected by machine learning methods.

14. The method according to claim 11, wherein, The method further includes the following steps: e) Classify the defects (6).

15. The method according to claim 14, wherein, During the manufacture of the component, steps a) and b) are performed, as well as at least one of steps a1), a2), c), d1), d2), and e).

16. The method according to claim 11, wherein, After the component is manufactured, perform steps a) and b), and at least one of steps a1), a2), c), d1), and d2).

17. A data processing apparatus comprising a processor adapted to perform the steps of the method according to any one of claims 1 to 16.

18. A 3D printer comprising a data processing apparatus according to claim 17, the construction space, and an imaging apparatus adapted to capture the first image and / or a sequence of images.

19. A computer program product comprising instructions that, when executed by a processor, cause a computer to perform the steps of the method according to any one of claims 1 to 16.

20. A computer-readable data carrier having a computer program storing instructions that, when executed by a processor, cause a computer to perform the steps of the method according to any one of claims 1 to 16.

Images