Apparatus and method for classifying a metal powder

Abstract

The invention relates to a method (29) and an apparatus (1) for classifying a metal powder (2). The aim of the invention of improving a characterization and classification of the metal powder (2) is achieved by means of an apparatus (1) which comprises: - an illumination unit (8) for illuminating the metal powder (2), the illumination unit (8) having a plurality of illuminants (10) located around the metal powder (2) so that the metal powder (2) can be illuminated from various horizontal angles and with various light wavelengths; - an image-capturing unit (9) for capturing image data relating to the metal powder (2) illuminated by the illumination unit (8) from the various horizontal angles and the various light wavelengths; and - a processing device (13) for processing the image data, the processing device (13) being operatively connected to the image-capturing unit (9) for signal communication, so that the processing device (13) can classify the metal powder (2) on the basis of the image data transmitted by the image-capturing unit (9).

Description

[0001] Apparatus and method for classifying a metal powder

[0002] The invention relates to a device and a method for classifying a metal powder.

[0003] Practical applications in metal 3D printing include processes and devices that can be used to characterize the properties of metal powders. The use of such devices and processes has become common in various industries and sectors, both in series production and prototype production. In an additive manufacturing process, the metal powder is combined into a complex component in a selective melting process.

[0004] Additive manufacturing processes encompass a variety of technologies in which components are successively built layer by layer based on a pre-programmed 3D data model. The advantage of these processes is that CAD data can be converted into a component directly, quickly, and without the need for special tools. The entire component is manufactured in a single step, unlike conventional manufacturing processes, which usually require several consecutive manufacturing steps. At the same time, this allows for a very high degree of geometric and design freedom.

[0005] In all powder-based metal 3D printing processes, in addition to process parameters such as melting power, scanning speed, and build chamber temperature, the properties of the metal powder used play a decisive role in the properties and quality of the manufactured component. The most important parameters of the metal powder include particle size distribution within a build chamber, flowability of the metal powder, specific surface area, and solid density of the metal powder. Knowing the quality of the metal powder is particularly crucial when using previously used but unmelted metal powder. So-called metal powder recycling is playing an increasingly important role in metal 3D printing, particularly due to rising raw material costs, supply bottlenecks, and sustainability considerations.Typically, during a metal 3D printing process, only a certain amount of the metal powder in the build chamber is used to produce the finished component. A large amount of metal powder is therefore unused after the component is completed and could be reused for another metal 3D printing process.

[0006] The problem here, however, is that the metal powder to be recycled is affected in unknown ways and to varying degrees by the previous metal 3D printing process. The energy input in and around a melting point of the metal powder during a metal 3D printing process is significantly higher than in areas farther away from the melting point. Accordingly, the metal powder located close to the component has to absorb more energy and heat than metal powder farther away from the component.

[0007] Other causes of aging, degradation, and contamination of the metal powder can include an oxidation reaction, which changes the properties of the metal powder, and an evaporation reaction, in which the evaporation of the more volatile components can change the composition of the metal powder. Accordingly, a metal 3D printing process may result in a non-homogeneous mixture of metal powder, making it unsuitable for use without further processing.

[0008] In practice, therefore, leftover metal powder from a metal 3D printing process is either disposed of, processed, and / or analyzed before a new metal 3D printing process. A common procedure here is to sort out larger metal lumps and particles by sieving, vibrating, and other methods, so that only the finer metal powder is reused in a new metal 3D printing process. In the prior art, WO 2020 / 229838 A1 discloses a method and apparatus for analyzing a metal powder for use in an additive manufacturing process. US 2018 / 200957 A1 discloses powder bed-based additive manufacturing of a component in a powder bed. Zhang, Jiahui & Habibnejadkorayem, Mahdi & Liu, Zhiying & Lyu, Tianyi & Sun, Qiang & Zou, Yu. (2021). A Computer Vision Approach to Evaluate Powder Flowability for Metal Additive Manufacturing. Integrating Materials and Manufacturing Innovation. 10. 101007 / s40192-021-00226-3 discloses a rapidly deployable, cost-effective, and reliable computer vision approach for evaluating the flowability of powders based on scanning electron microscopy images.

[0009] It is an object of the invention to provide a device and a method with which a characterization and classification of the metal powder can be carried out.

[0010] The problem is solved according to the independent patent claims. Embodiments are specified in the dependent claims.

[0011] Disclosed is an apparatus for classifying a metal powder, comprising:

[0012] - a lighting unit for illuminating the metal powder, wherein the lighting unit has at least one illuminating means, preferably a plurality of illuminating means, which are arranged in particular around the metal powder, with which the metal powder can be illuminated from different illumination angles, preferably horizontal angles, and with different light wavelengths;

[0013] - an image recording unit for recording image data of the metal powder illuminated by the illumination unit from the different illumination angles, preferably horizontal angles, and the different light wavelengths, and

[0014] - a processing device for processing the image data, wherein the processing device is operatively connected to the image recording unit by signal technology, so that the processing device can classify the metal powder based on the image data transmitted by the image recording unit.

[0015] Embodiments of the invention could have the advantage of allowing metal powders to be characterized in a simple but reliable manner. Compared to common analytical methods for determining powder quality, such as determining the angle of repose, Hall flow meter measurements, determining the flowability of the metal powder, determining the particle size of the metal powder, or scanning electron microscopy measurements, the device in question could significantly simplify the determination. For example, in-situ characterization could be possible without requiring special powder preparation (such as, for example, for performing scanning electron microscopy measurements).

[0016] Embodiments could further have the advantage that the plurality of illuminants arranged around the metal powder enables illumination of the metal powder from all possible horizontal angles between 0° and 360°. This could enable a particularly precise determination of the particle size and degree of degradation, since metal powder particles are illuminated from a variety of horizontal directions regardless of the viewing angle, ensuring optimal, representative illumination of the metal powder particles. In addition, illuminating the metal powder with different light wavelengths from all possible horizontal angles between 0° and 360° could enable precise characterization of specific properties of the illuminated metal powder, in particular oxide layers of the metal powder as an indication of degradation or particle size distributions.

[0017] The device according to the invention is, for example, in an optically operative connection to a powder feed stream of the metal powder supplying the device with metal powder, so that on the one hand the illumination unit can illuminate the powder feed stream of the metal powder, and on the other hand the image recording unit can record the image data from the powder feed stream of the metal powder thus illuminated. For example, the metal powder can be classified immediately before a metal 3D printing process or a powder metallurgical manufacturing process, whereby an unsuitable portion of the metal powder can be replaced and disposed of before processing in the respective metal 3D printing process or manufacturing process; preliminary examinations of samples of the metal powder are therefore no longer absolutely necessary. Advantageously, this aspect of the device according to the invention sustainably and efficiently increases the degree of utilization of the metal powder used.

[0018] For example, the illumination unit allows the metal powder to be illuminated from different vertical illumination angles. Different vertical illumination angles could further improve the shadows cast by the different particles of the metal powder and enable the creation of closed reflection points, which could enable improved detection of the particle size distribution of the metal powder in the image data acquired with the image acquisition unit.

[0019] For example, the metal powder can be illuminated from the same horizontal angle with different monochromatic or non-monochromatic light wavelengths. Examples could have the advantage that the metal powder illuminated with varying light spectra emits different light depending on its condition (degradation, particle size). This, in turn, can be detected by the camera used, thus enabling easier and more precise characterization of the specific properties of the illuminated metal powder, such as oxide layers as an indication of degradation or a particle size distribution.

[0020] Advantageously, the illumination unit comprises a plurality of illuminants arranged in the vicinity of the metal powder and secured to the illumination unit in such a way that the metal powder can be correctly illuminated, at least in part. The various illumination angles from which the metal powder can be illuminated result in additionally improved shadows cast by the various particles of the metal powder, particularly from flatter illumination angles, thereby improving the detection of a particle size distribution of the metal powder in the image data recorded with the image recording unit. The metal powder typically consists of particles of different sizes; the particle size distribution describes the distribution of the different sized particles within the total amount of metal powder.

[0021] The metal powder comes into contact with oxygen, for example, during the metal 3D printing process. The additional oxygen and the thermal energy, e.g., from a laser beam, cause an oxidation layer to form on the particle surface, whereby the oxygen content increases after each printing process. Titanium alloys Ti-5AI-5Mo-5V-1Cr-1Fe and Ti-6A1-2Mo-1,5Cr-2Zr-2Sn-2Nb form oxide layers of different colors at certain temperatures. For example, a yellowish / golden TiO layer forms at 500°C, and a bluish Ti2O3 layer at 600°C. Violet discoloration has also been observed at 550°C. With regard to mechanical properties, it is known that blue-colored oxide layers should be considered a critical color between acceptable and unacceptable tensile properties.

[0022] According to the invention, the at least one illuminant of the illumination unit can simultaneously illuminate the metal powder with different light wavelengths, wherein the image recording unit is configured to record the image data of the metal powder illuminated by the illuminants of the illumination unit from different illumination angles and with different light wavelengths. The image data can be real-time recordings in the form of video data or images recorded at time intervals, each of which is provided to the processing device as image data.

[0023] Accordingly, the image acquisition unit is designed, for example, to capture microscopic images and image recordings, so that resolutions in the pm range are possible in order to capture a sufficient number of particles of the metal powder. Tests have revealed advantageous resolutions of approximately 10 pm to over 200 pm. The processing device processes the image data received from the image acquisition unit to such an extent that, for example, a comparison can be made between the image data recorded in an operating state and the training image data recorded in a training state. Based on the image data of the metal powder recorded from different illumination angles and with different light wavelengths, the processing device is able to draw conclusions about the properties of the metal powder and to classify the metal powder on which the image data is based.

[0024] In this respect, the device according to the invention with the processing device can determine a particle size distribution of the metal powder from the image data and detect oxide layers of the metal powder, so that a classification of the metal powder is possible based on training image data recorded in the training state - tests showed good classification results for metal powders such as tin bronze, stainless steel, iron, hot-work steel and titanium.

[0025] The advantage of the invention could be a cost-effective classification of the metal powder during the manufacturing process of a component from the metal powder, whereby each layer of the metal powder can be qualified separately before processing.

[0026] For example, it can be provided that the device further comprises a control unit, wherein the control unit is designed to synchronously control the illumination unit and the image recording unit for illuminating the metal powder based on an illumination specification and recording the image data, wherein the illumination specification comprises the illumination from the different illumination angles and with the different light wavelengths.

[0027] The illumination specification can comprise a temporal sequence, preferably a temporal sequence of illumination with illumination steps from the various illumination angles and the various light wavelengths, and / or a combined use of the illuminants used in the illumination unit. On the one hand, the metal powder is illuminated simultaneously from a subset of the various possible illumination angles with various possible light wavelengths, and on the other hand, these subsets themselves can be different and can vary over time during an illumination process. Predetermined and advantageous illumination specifications can be identified in the training state for various metal powders and subsequently used in the operating state to achieve improved classification of the metal powder.

[0028] In order to achieve uniform illumination of the powder stream of the metal powder, one embodiment of the invention can provide for the illuminating means of the illumination unit to be arranged symmetrically, preferably point-symmetrically, around the image recording unit. In this case, several illuminating means of the illumination unit can form an illumination ring, wherein several illumination rings with different ring diameters are arranged at different distances from the metal powder. Accordingly, the image recording unit is preferably arranged in a central region of the illumination rings and secured thereto, wherein in this example the illumination rings are arranged concentrically around the image recording unit. According to this aspect of the invention, the metal powder can be uniformly illuminated in one region by the illumination unit, and image data can be recorded by the image recording unit at the same time.

[0029] Alternatively, the light sources of the illumination unit can form one or more illumination rectangles or triangles, although a combination of illumination rings, rectangles, and / or triangles is also conceivable. Optionally, the illumination unit and the image acquisition unit can consist of a microscope into which the (LED) light sources are integrated.

[0030] For example, it can be provided that this has a computing unit, wherein a computing unit is designed to execute a trained classification algorithm for classification, which carries out the classification of the metal powder based on the recorded image files, wherein the results of the classification comprise information about a metal powder type. For example, the metal powder type comprises a chemical composition of the metal powder and / or information about a degradation of the metal powder. Executing the trained classification algorithm could have the advantage of providing an improved classification of a metal powder, which contains information about the chemical composition or chemical structure of the metal powder as well as information about the quality or degradation of the metal powder.Thus, the classification algorithm could provide a simple, cost-effective, fast and automated way to obtain important information about a metal powder (composition, quality) in order to decide, for example, whether and for which purpose the metal powder can be used in the future.

[0031] For example, a supporting classification of the metal powder can be provided, which classifies the metal powder on the basis of a recorded particle size distribution of the metal powder and detected oxide layers of the metal powder, wherein the particle size distribution is determined by an analysis of the image data which is recorded with the image recording unit during illumination of the metal powder by the illumination unit with white light, and wherein the oxide layers are determined by an analysis of the image data which is recorded with the image recording unit during illumination of the metal powder by the illumination unit with colored light, in particular violet light.

[0032] When the metal powder is illuminated with white light, the white light is scattered in different directions by the metal powder particles depending on the illumination angle of the white light, so that only a portion of the white light is scattered back toward the image acquisition unit at a favorable scattering angle and can be recorded by the image acquisition unit. The particle size of the metal powder and the illumination angle influence the amount of light scattered by the particles to the image acquisition device. Therefore, the image data exhibits white areas of varying sizes, which directly correlate with the particle size, with spaces between the white areas being dark. Conclusions about the particle size distribution of the metal powder can be drawn from the analysis of the image data recorded with white light and image segmentation.

[0033] Accordingly, the analysis of the image data captured with the image acquisition unit of the metal powder illuminated with colored light can be used to draw conclusions about oxide layers on the metal powder particles. Colored areas in the image data around the particles indicate degraded particles of reused metal powder. Tests have shown that reddish to violet and bluish light, in particular, indicate titanium oxide layers. Using a combination of illumination angles (which can be horizontal or vertical), white light, and colored light, the particle size distribution and the degree of degradation of the metal powder can be determined. A classification algorithm trained on this basis can reliably classify the illuminated metal powder.

[0034] For example, assisted classification involves matching the particle size distribution and detected oxide layers captured from the image data to a metal powder known from training. This matching could have the advantage of supporting the assignment of the examined metal powder to a known metal powder by comparing the particle size distribution and the detected oxide layers. Thus, assisted classification provides a simple, cost-effective, fast, and automated way to verify the classification results.

[0035] For example, if it is not possible to assign the particle size distribution and detected oxide layers recorded from the image data to a metal powder known from training, the metal powder is classified as an unknown powder. This could have the advantage of avoiding errors in the characterization of the metal powder and, in a subsequent step, also allowing unknown metal powder types to be added to the training. In order to obtain improved results in determining the particle size distribution and the oxide layers, according to an advantageous embodiment of the device according to the invention, the metal powder can be illuminated with the white light at a steeper illumination angle, preferably a steeper vertical illumination angle, than the illumination with the colored light.One aspect of the invention includes that the illumination angles can be in a lower range of 20° to 40° up to an upper range of 60° to 80° to the metal powder. Colored light can correspond to monochromatic light with a limited light wavelength range, which is not necessarily limited to the light wavelength ranges visible to humans, but can also correspond to an infrared or ultraviolet light wavelength range - thus, all types of LEDs and laser LEDs are also included in the range of colored light emitting lamps.

[0036] White light corresponds to a non-monochromatic light wavelength range, which thus encompasses large wavelength ranges visible to humans. Accordingly, white light can encompass a wide range of light wavelengths, for example, with a wavelength of 300 nm to 800 nm. The types of metal powder to be classified determine which wavelength ranges in the image data captured by the image acquisition unit allow conclusions to be drawn about the illuminated metal powder.

[0037] According to this aspect, it can optionally be provided that the illumination resulting from the different light wavelengths comprises at least two or more of the following light spectra: white light, violet light, infrared light, UV light. Thus, it is optionally possible for the light sources to comprise LEDs and, optionally, a laser. In practice, laser LEDs with a light wavelength range around 750 nm have proven successful.

[0038] For example, the classification algorithm can comprise an artificial neural network, in particular a convolutional neural network. According to the invention, (Convolutional) artificial neural networks can be advantageously used to classify the multispectrally illuminated metal powder. Before being used in the operating state, the artificial neural network must be trained and adapted to the metal powders to be classified in the training state. This is typically done using test image data of known metal powders. In practice, a VGG16 model for the artificial neural network has proven sufficiently efficient. Optionally, a VGG19 model can also be used.

[0039] For example, it can be provided that the lighting unit is designed to illuminate the metal powder in a lighting area, wherein the image recording unit is designed to record the image data in the lighting area, wherein the lighting unit has a plurality of illuminating means arranged offset and spaced from one another, wherein in particular due to the offset and the spacing of the illuminating means with respect to the lighting area, a smallest illumination angle, preferably a smallest vertical illumination angle, is at least 30° and a largest illumination angle, preferably a largest vertical illumination angle, is at most 70%. The different lighting areas with the different illumination angles are an advantageous aspect of the invention, since this can improve the shadow cast by the particles of the metal powder and enable the image data to be better analyzed.

[0040] Advantageously, it can be provided that a plurality of identical illuminants form a luminous group for simultaneously illuminating the metal powder along a respective circumferential line. Optionally, the device according to the invention can accordingly comprise a plurality of illuminating groups, each illuminating group being designed to provide illumination with a different wavelength of light, the illuminants of at least some of the different illuminating groups being arranged along different circumferential lines with different circular radii, with circumferential lines preferably having a different distance from a center of the illumination region. According to this aspect, the metal powder is not only uniformly illuminated within an illumination region, but the different illuminating groups also do not significantly influence one another.In order to improve the integration of the device according to the invention, it can be provided that the device further comprises a movement unit, wherein the movement unit is designed to move the illumination unit and the image recording unit relative to the metal powder for illuminating and recording the image data of different areas of the metal powder and / or for achieving the different vertical illumination angles and / or for achieving the different horizontal angles, wherein the classification of the metal powder is carried out on the basis of the image data of the different areas as an overall classification. The movement unit can have an electric motor, in particular a stepper motor, with which the illumination unit and the image recording unit can be displaced precisely, evenly and quickly, wherein the movement unit is operatively connected to the control unit in terms of signals and is controlled by the latter.

[0041] For example, the device may further comprise a mixing unit, wherein the mixing unit is configured to mechanically change the mixing of the metal powder. The mixing unit can ensure, both in the training state and in the operating state, that the metal powder is thoroughly mixed between the acquisition of two image data sets. Mixing the metal powder could have the advantage of providing particularly precise characterization of the metal powder, since it enables illumination and acquisition of image data of the metal powder in multiple orientation states of the metal powder particles.

[0042] For example, the mixing unit can comprise a doctor blade and a plate, in particular a rotary plate for receiving the metal powder, wherein the mixing unit is designed to effect mixing at least partially by relative movement of the plate and the doctor blade to one another, in particular by rotation of the rotary plate. The use of a doctor blade and a plate, in particular a rotary plate, could have the advantage of providing a device that enables an automated mixing process of the metal powder. For example, it can be provided that this is designed to control an energy source for sintering or melting the metal powder when the processing device has classified the metal powder as being of sufficient quality.During sintering, the metal powder is heated but not melted, so that the powdery basic shape of the metal powder particles is essentially retained, but shrinkage usually occurs during a sintering process and gaps between the metal powder particles disappear - a laser is often the energy source for sintering, as this can be controlled efficiently, effectively and precisely.

[0043] As an alternative to laser beam sintering, an additive manufacturing process involving powder bed melting using an electron beam can also be used. The electron beam serves as the energy source. The electron beam preheats the metal powder over a larger area and then melts it in predetermined small areas, allowing the component to be manufactured layer by layer. Alternatively, an arc can be used as the energy source.

[0044] For example, it can be provided that the device is designed to control a feed device with which the metal powder is provided.

[0045] According to one aspect of this embodiment, it can be provided that the feed device is designed as a nozzle device with which the metal powder is blown into the build space layer by layer and simultaneously the energy source sinters or melts the metal powder, wherein the nozzle device has a dosing unit comprising the lighting unit and image recording unit, with which the amount of metal powder for blowing into the build space is dosed, and wherein the processing device qualitatively classifies the metal powder before blowing it into the build space. The dosing unit can be a rotating disk, relative to which the lighting unit and the image recording unit are fixed, wherein a rotational speed of the disk influences the amount of metal powder for blowing into the build space. E.g.It can be provided that the feed device is designed as a coating device for coating the metal powder, with which a powder bed of the metal powder is produced layer by layer in a construction space, wherein both the component to be manufactured is produced layer by layer and the lighting unit and the image recording unit are introduced into the construction space, and wherein the powder bed is sintered or melted by means of the energy source after the qualitatively positive classification of the metal powder by the processing device or the powder bed is removed again after a qualitatively negative classification.

[0046] Optionally, the device can be configured to initiate removal of the metal powder and provision of a new metal powder by the feed device if the processing device classifies the metal powder as insufficiently qualitative. The negative and insufficiently qualified powder bed can thus be blown off or removed, for example.

[0047] Further disclosed is a method, the method comprising:

[0048] - illuminating the metal powder with a lighting unit, wherein the lighting unit has at least one illuminating means, preferably a plurality of illuminating means, which are arranged in particular around the metal powder, with which the metal powder is illuminated from different illumination angles, preferably horizontal angles, and with different light wavelengths;

[0049] - Recording image data of the metal powder illuminated by the illumination unit with an image recording unit from the different illumination angles, preferably horizontal angles, and the different light wavelengths and

[0050] - Processing the image data with a processing device, wherein the processing device is operatively connected to the image recording unit in terms of signals, so that the processing device classifies the metal powder based on the image data transmitted by the image recording unit.

[0051] For example, a powder feed stream of the metal powder is optically connected to the illumination unit and the image acquisition unit, so that the metal powder can be illuminated at least in part by the illumination unit, and the image acquisition unit can capture image data of the metal powder. Advantageously, according to this aspect of the invention, the powder feed stream of the metal powder can be qualitatively classified by the processing device before each metal 3D printing process, powder metallurgical manufacturing process, or the like.

[0052] The method for illuminating the metal powder from different illumination angles, in particular from different horizontal and / or vertical illumination angles, ensures, for example, optimal illumination of the metal powder and optimal shadowing of the particles, so that the particle size distribution of the metal powder can be accurately analyzed and determined. Illuminating the metal powder with different light wavelengths, in particular from different horizontal and / or vertical illumination angles, enables the characterization of specific properties of the illuminated metal powder. In particular, oxide layers of the metal powder, which are an indication of degradation, can be identified using different light wavelengths.

[0053] According to the method according to the invention, the image data of the metal powder illuminated by the illumination unit can be recorded by the image acquisition unit with a resolution in the pm range, for example, from approximately 10 pm to over 200 pm, with the image data then being transmitted to the processing device. Advantageously, the image acquisition unit comprises a microscope and a communication interface with which the image acquisition unit and the processing device can be brought into a signal-based connection.In order to be able to advantageously classify the metal powder by the processing device, an optional embodiment of the method according to the invention can provide for a trained classification algorithm to be executed for classification purposes, which performs the classification of the metal powder based on the recorded image files. The results of the classification include information about a metal powder type. For example, the metal powder type includes a chemical composition of the metal powder and / or information about a degradation of the metal powder.

[0054] For example, a supporting classification of the metal powder can be provided, which classifies the metal powder on the basis of a recorded particle size distribution of the metal powder and detected oxide layers of the metal powder, wherein the particle size distribution is determined by an analysis of the image data which is recorded with the image recording unit during illumination of the metal powder by the illumination unit with white light, and wherein the oxide layers are determined by an analysis of the image data which is recorded with the image recording unit during illumination of the metal powder by the illumination unit with colored light, in particular violet light.

[0055] Illuminating the metal powder with white light allows the classification algorithm to better capture the particle size distribution of the metal powder. The white light is partially scattered by the particles toward the image acquisition unit, while the gaps between the metal powder particles do not cause any scattering toward the image acquisition unit. Using image segmentation, the dark and light areas captured in the image data can be identified, and the particle sizes of the metal powder can be determined.

[0056] Illuminating the metal powder with colored light, however, highlights oxide layers of the metal powder particles depending on the light wavelength range of the colored light and the type of metal powder, so that the image data can be transformed into a color space, for example an HSV color space, and a color value can be detected for each pixel of the image data, which can be used to identify degradation of the metal powder and its particles.

[0057] Based on the particle size distribution and the detected oxide layers of the metal powder, the classification algorithm, after training with known image data of the various metal powders, can qualitatively classify the metal powder in question and assign it to a metal powder type known to the classification algorithm.

[0058] According to an advantageous embodiment of the method according to the invention, the classification algorithm can comprise an artificial neural network, in particular a convolutional neural network. (Convolutional) artificial neural networks are excellently suited to training with known image data, so that the trained artificial neural network can correctly recognize and classify known metal powders. In practice, a VGG16 model for the artificial neural network has proven sufficiently efficient. Optionally, a VGG19 model can also be used.

[0059] Advantageously, the method according to the invention can be provided for the neural network to be trained using image data, comprising:

[0060] - inputting a set of image data of a known metal powder illuminated from different illumination angles, preferably from different horizontal angles and / or vertical illumination angles, and with different light wavelengths into the neural network;

[0061] - Training the neural network using a classification output of the neural network and a classification matching the image data set of the known metal powder.

[0062] The training is designed, for example, to compare the classification output of the neural network with the known classification output and to attribute any error between the classification output and the known classification output. The error typically has a quadratic effect on the weights of the neural network and adjusts the weights of the neural network such that the error is minimized after a certain amount of image data has been obtained. For this purpose, characteristic properties of various metal powders are highlighted by the recorded particle size distribution and the detected oxide layers, then fed to the neural network as test image data, and subsequently evaluated based on the known metal powders and the lighting types used.

[0063] In a further aspect, the invention relates to a neural network obtainable by the method described above.

[0064] Furthermore, the device and the method are not limited to use in the field of metal 3D printing, but are also applicable, for example, to powder metallurgical manufacturing processes such as metal powder injection molding or thermal spraying with metal powders.

[0065] The invention is illustrated below by way of example using schematic representations. They show:

[0066] Figure 1 shows a device for classifying a metal powder,

[0067] Figure 2 shows a holding device of a lighting unit of the device,

[0068] Figures 3 and 4 show the lighting unit and an image recording unit of the device from a lower perspective view,

[0069] Figures 5 and 6 show an illumination of a particle of a metal powder from two illumination angles,

[0070] Figures 7 and 8 show two images of the metal powder particles taken from two illumination angles,

[0071] Figures 9 and 10 show images of two illuminated different metal powders from different illumination angles, Figure 11 shows an application of the algorithm for determining the particle size distribution,

[0072] Figure 12 detected pixels in an image of titanium fumes taken under violet light and

[0073] Figure 13 is a flowchart of an artificial neural network for classifying a metal powder.

[0074] Figure 1 shows a device 1 for classifying a metal powder 2, wherein the device 1 has an optional electric motor-driven turntable 3 which is designed to provide a build space 4 for the component to be manufactured. A control unit 5 comprised by the device 1 can control a feed device by means of a signal connection 6, such that the feed device introduces a powder feed stream of the metal powder 2 into the build space 4, wherein, for example, a doctor blade 7 in the build space 4 distributes the metal powder 2 into a powder bed via the rotating turntable 3. The device 1 also has an energy source in the form of, for example, a laser (not shown), with which the device 1 can heat and sinter the powder bed in certain areas according to the specifications of the control unit 5, such that with each application of a new powder layer of the metal powder 2, the component can be manufactured layer by layer.

[0075] Figure 1 further shows a lighting unit 8 and an image recording unit 9, wherein the metal powder 2 in the build space 4 can be illuminated from different vertical illumination angles 11, horizontal angles, and with different light wavelengths using the lighting unit 8 and a plurality of light sources 10 comprised by the lighting unit 8. The metal powder can be illuminated from the same horizontal angle with different light wavelengths, as well as from the same vertical illumination angle 11 with different light wavelengths. The lighting unit 8 comprises an image recording unit 9 in a central region 12, which can record image data of the metal powder 2 illuminated by the lighting unit 8 from the different illumination angles 11 and the different light wavelengths.According to this specific embodiment of the device 1 according to the invention, the image acquisition unit 9 is a microscope that is operatively connected to the control unit 5 via signal technology. The microscope has a specific magnification characteristic and can record fewer pixels per 100 pm with increasing magnification. The image data recorded from the illuminated metal powder 2 can be transmitted from the image acquisition unit 9 via the signal connection 6 to the control unit 5, so that a processing unit 13 included in the control unit 5 can analyze the image data and classify the metal powder 2.

[0076] For this purpose, the control unit 5 is also operatively connected to the illumination unit 8 via the signal connection 6 and can transmit to the illumination unit 8 an illumination specification according to which the metal powder 2 is to be illuminated so that it can be adequately classified by illuminating the metal powder 2 from different illumination angles 11 and with different light wavelengths. The illumination specification can, for example, contain information that several image data sets of the metal powder are produced per minute, each at different light wavelengths, vertical illumination angles, and horizontal angles. It is also conceivable, for example, that the metal powder is mixed or moved using the turntable and / or the squeegee in the interval between the acquisition of two image data sets.In one example, multiple image data are taken, each at different light wavelengths, vertical illumination angles, and horizontal angles, then the metal powder is mixed or agitated using the turntable and / or squeegee, and the process begins again.

[0077] Figure 2 shows a holding device 14 of the lighting unit 8 according to the invention from Figure 1, wherein the holding device 14 is circular and designed to hold, position and fix a plurality of illuminating means 10 in the form of LEDs arranged concentrically around an image recording opening 12. An alternative arrangement of the illuminating means 10 is possible. For example, in Figure 2, the holding device 14 has positioning openings 16 for at least two different illuminating groups of the illuminating means 10 along different circular circumferential lines 15 with different circular radii, wherein the circular circumferential lines 15 are spaced apart from one another along an illumination direction 17, so that the illuminating means 5 of the different illuminating groups can illuminate the metal powder 2 in sections according to the illumination specification of the control unit 5. The illuminating means 10 have, for example,above the upper circumferential line 15 an illumination angle 11 of approximately 70° to a plane 18 of the powder bed, wherein the illuminants 10 below the upper circumferential line 15 or above a lower circumferential line 15 have an illumination angle 11 of approximately 30°.

[0078] Figure 2 also shows the image recording opening 12, which is arranged centrally in the holding device 14 by way of example, with the image recording unit 9 in Figure 1 being inserted into it and secured to the holding device 14. Accordingly, when the metal powder 2 is partially illuminated with the illuminants 10, image data can be simultaneously recorded with the image recording unit 9 through the image recording opening 12.

[0079] Figures 3 and 4 show the illumination unit 8 and the image recording unit 9 of the device 1 from Figure 1, each shown enlarged from a lower perspective view. Shown are the different types of illuminants 10 arranged concentrically around the image recording unit 9 and secured to the holding device 14 of the illumination unit 8. The holding device 14 is inserted into the illumination unit 8 and secured thereto, so that the illumination unit 8 can illuminate the metal powder 2 in regions along the illumination direction 17, while the image recording unit 9 can record image data from the metal powder 2 illuminated from the different illumination angles 11 and the different light wavelengths.According to the invention, the different illumination angles 11 are obtained by the position of the illuminants 10 in the respective positioning openings 16, wherein the different light wavelengths are determined by the affiliation of the respective illuminant 10 to a lighting group and can be specified by the illumination specification of the control unit 5. Furthermore, Figures 3 and 4 show a total of four optional laser LEDs as illuminants 10, which are fixed outside the holding device 14 directly to the illumination unit 8. The laser LEDs 10 have, for example, a light wavelength of approximately 750 nm and a steeper illumination angle 11 than the other illuminants 10, so that the shadow cast by the particles 19 of the metal powder 2 can be improved.

[0080] Figures 5 and 6 each show a basic illumination of a particle 19 of a metal powder 2 from a different vertical illumination angle 11 - specifically, Figure 5 shows an illumination of the metal powder 2 from a vertical illumination angle 11 of 70° to a plane 18 of a powder bed of the metal powder 2, while Figure 6 shows a vertical illumination angle 11 of 30°.

[0081] Both Figures 5 and 6 show a particle 19 of the metal powder 2, which is illuminated (only by way of example) by a light source 10. Light rays emitted by the light source 10 strike the particle 19 at various points and are scattered by it, with a portion of the light rays 20 radiating past an image recording unit 9 and not being displayed in the image data of the image recording unit 9 or being displayed as dark pixels.

[0082] Other light rays 20, however, are favorably scattered by the particle 19 in the direction of the image recording unit 9 and guided to an image recording sensor 22 with the aid of an optional optics 21, wherein the vertical illumination angle 11 from which the illuminant 10 illuminates the particle 19 essentially determines the position at which the favorably scattered light rays 20 strike the image recording sensor 22.

[0083] In this respect, Figures 7 and 8 show two image recordings 23 of the particles 19 of the metal powder 2 recorded from the two vertical illumination angles 11; Figure 7 shows the image recording 23 of the metal powder 2 with a vertical illumination angle 11 of 70°, and Figure 8 with a vertical illumination angle 11 of 30°. Thus, in both image recordings 23, the same region of the metal powder 2 is shown from different vertical illumination angles 11. If both image recordings 23 are merged, an improved particle size distribution of the metal powder 2 can be determined, wherein the particle size distribution describes the distribution of the differently sized particles 19 in the total amount of metal powder 2.

[0084] Figures 9 and 10 show two different metal powders 2, each with different lighting. Figure 9 shows the illumination of titanium powder 2, wherein the left image 23 was taken at a vertical illumination angle 11 of 70° and the right image 23 was taken at a vertical illumination angle 11 of 30°. Figure 10 shows the illumination of stainless steel powder 2, wherein the left image 23 was taken at a vertical illumination angle 11 of 70° and the right image 23 was taken at a vertical illumination angle 11 of 30°. In this respect, it is further shown that under illumination with steep illumination angles 11 (for example 70°), closed reflection points are formed on particles 19 of the respective metal powder 2, whereby larger particles 19 are often not closed at flatter illumination angles 11 (for example 30°).

[0085] Figure 11 shows an exemplary determination of a particle size distribution using a stainless steel 316L metal powder 2 in three image recordings 23. The left image recording 23 shows an input image 24 of the metal powder 2 illuminated by the illumination unit 8 in one area, wherein the larger particles 19 are shown in clear white and some areas between the clear white particles 19 remain dark because no light rays 20 were reflected from there to the image recording unit 9.

[0086] The areas of the left-hand input image 24 that are not clearly displayed as white or black could be problematic for computer-assisted image recognition, so that in a first image processing step, a threshold image 25 shown in the center is created. The threshold image 25 now only shows pixels of the image 23 that lie above a brightness threshold, so that faintly white and gray areas of the input image 24 are removed. In a subsequent image processing step, an output image 26 shown on the right is created, in which only pixel regions that were clearly identified as a particle 19 are shown, so that possible artifacts and image acquisition errors that could distort the particle size distribution are removed.In this respect, a particle size distribution of the input image 24 and thus of the metal powder 2 can be reproducibly and reliably recognized from the particles 19 of the metal powder 2 clearly recognized in the output image 26.

[0087] Figure 12 shows an exemplary determination of oxide layers (27, 28) based on a reused and degraded titanium metal powder 2 in three image recordings 23. A left-hand input image recording 24 shows a region of the metal powder 2 that was illuminated with colored—in particular, violet—light by an illumination unit 8, so that an image recording device 9 could record image data. Two regions (not representable in color) can be determined in the input image recording 24: a lower reddish region 27 and an upper bluish region 28. Both reddish and bluish reflections or regions 27, 28 indicate degradation of the metal powder 2. An HSV color space is particularly well suited for segmenting specific colors from image data, so that the input image recording 24 is first transformed into HSV coordinates in an evaluation algorithm.The evaluation algorithm then searches the HSV coordinates to determine which pixels fall into defined HSV color ranges, so that the average number of reddish or bluish pixels 27, 28 per input image 24 can be calculated.

[0088] In this respect, the middle image 23 shows the detected regions 27, 28, which are represented by red pixels 27 in the input image 24, and the right image 23 shows the detected regions 27, 28, which are represented by bluish pixels 28 in the input image. Both detected regions 27, 28 indicate a degree of degradation of the titanium metal powder 2, which can advantageously be used to classify the metal powder 2. Figure 13 shows a method 29 for classifying a metal powder 2, wherein, in an initial input step 30, image data of a metal powder 2 are recorded by an image recording unit 9, while the metal powder 2 is illuminated by an illumination unit 8 from various vertical illumination angles 11, horizontal angles, and with various light wavelengths. The image data serves as input values ​​for a convolutional artificial neural network and is, for example, 244x224 pixels in size.

[0089] In a subsequent convolution step 31, the image data generated in the input step 30 are processed and propagated by a convolution layer of the neural network. A filter matrix, serving as a filter kernel, is convolved section by section with the image data, and information reduction is then performed.

[0090] The image data, filtered several times in the convolutional layer and information-reduced, are used in a subsequent classification step 32 by a classification layer of the neural network to classify the metal powder 2. In a subsequent output step 33, as part of the supporting classification, the image data from the input step 30 are assigned to a metal powder 2 known from training based on the particle size distribution and the detected oxide layers 27, 28. The output step 33 verifies or validates the results of the classification step 32. If there is no agreement between the classification step 32 and the output step 33 within certain limits, the powder is classified as an unknown powder; otherwise, the metal powder type in terms of a chemical composition of the metal powder and an indication of the degradation of the metal powder are output.List of reference symbols for the formation of a metal powder.

Claims

Patent claims 1. Device (1) for classifying a metal powder (2) with: - a lighting unit (8) for illuminating the metal powder (2), wherein the lighting unit (8) has a plurality of lighting means (10) arranged around the metal powder (2) to enable illumination of the metal powder (2) from different horizontal angles and with different light wavelengths; - an image recording unit (9) for recording image data of the metal powder (2) illuminated by the illumination unit (8) from the different horizontal angles and the different light wavelengths, and - a processing device (13) for processing the image data, wherein the processing device (13) is operatively connected to the image recording unit (9) by signal technology, so that the processing device (13) can classify the metal powder (2) based on the image data transmitted by the image recording unit (9).

2. Device (1) according to claim 1, wherein the illumination unit (8) enables illumination of the metal powder (2) additionally from different vertical illumination angles (11).

3. Device (1) according to one of the preceding claims, wherein the metal powder (2) can be illuminated from a same horizontal angle with different monochromatic or non-monochromatic light wavelengths.

4. Device (1) according to one of the preceding claims, further comprising a control unit (5), wherein the control unit (5) is designed to synchronously control the illumination unit (8) and the image recording unit (8) for illuminating the metal powder (2) based on an illumination specification and recording the image data, wherein the illumination specification determines the illumination from the different illumination angles and with different light wavelengths.

5. Device (1) according to claim 4, wherein the illumination specification further comprises a temporal sequence of the illumination with illumination steps from the different illumination angles and the different light wavelengths.

6. Device (1) according to one of the preceding claims, wherein the lighting means (10) of the lighting unit (8) are arranged symmetrically, preferably point-symmetrically, around the image recording unit (9).

7. Device (1) according to one of the preceding claims, wherein a computing unit is designed to execute a trained classification algorithm for classification, which carries out the classification of the metal powder (2) based on the recorded image files, wherein the results of the classification comprise an indication of a metal powder type.

8. Device (1) according to claim 7, wherein the metal powder type comprises a chemical composition of the metal powder and / or an indication of a degradation of the metal powder.

9. Device (1) according to claim 7 or 8, further comprising a supporting classification which classifies the metal powder (2) on the basis of a recorded particle size distribution of the metal powder (2) and detected oxide layers (27, 28) of the metal powder (2), wherein the particle size distribution is determined by an analysis of the image data which is recorded with the image recording unit (9) during illumination of the metal powder (2) by the illumination unit (8) with white light, and wherein the oxide layers (27, 28) are determined by an analysis of the image data which is recorded with the image recording unit (9) during illumination of the metal powder (2) by the illumination unit (8) with colored light, in particular violet light.

10. Device (1) according to claim 9, wherein the supporting classification comprises an assignment of the particle size distribution acquired from the image data and detected oxide layers to a metal powder known from the training.

11. Device (1) according to claim 10, wherein in the event that an assignment of the particle size distribution acquired from the image data and detected oxide layers to a metal powder known from the training is not possible, the metal powder (2) is classified as an unknown powder.

12. Device (1) according to claims 9-11, wherein the illumination of the metal powder (2) with the white light takes place at a steeper vertical illumination angle (11) than the illumination with the colored light.

13. Device (1) according to claims 7-12, wherein the classification algorithm comprises an artificial neural network, in particular a convolutional neural network.

14. Device (1) according to one of the preceding claims 2-13, wherein the illumination unit (8) is designed to illuminate the metal powder (2) in an illumination region, wherein the image recording unit (9) is designed to record the image data in the illumination region, wherein the illumination unit (8) has a plurality of illuminating means (10) arranged offset and spaced from one another, wherein in particular due to the offset and the spacing of the illuminating means (10) with respect to the illumination region, a smallest vertical illumination angle (11) is at least 30° and a largest vertical illumination angle (11) is at most 70%.

15. Device (1) according to claim 14, wherein a plurality of identical illuminating means (10) form a illuminating group for simultaneously illuminating the metal powder (2) along a respective circumferential line (15).

16. Device (1 ) according to claim 15, comprising several of the lighting groups, each lighting group being designed to carry out the illumination with a different Light wavelength is formed, wherein the illuminating means (10) of at least some of the different illuminating groups are arranged along different circular circumferential lines (15) with different circular radii, wherein preferably circular circumferential lines (15) have a different distance from a center (12) of the illumination area.

17. Device (1) according to one of the preceding claims, wherein illumination resulting from the different light wavelengths comprises at least two or more of the following light spectra: white light, violet light, infrared light, UV light.

18. Device (1) according to one of the preceding claims, wherein the lighting means (10) comprise LEDs and optionally a laser.

19. Device (1) according to one of the preceding claims, further comprising a movement unit, wherein the movement unit is designed to move the illumination unit (8) and the image recording unit (9) relative to the metal powder (2) for illuminating and recording the image data of different areas of the metal powder (2) and / or for achieving the different vertical illumination angles and / or for achieving the different horizontal angles, wherein the classification of the metal powder (2) is carried out on the basis of the image data of the different areas as an overall classification.

20. Device (1) according to one of the preceding claims, further comprising a mixing unit, wherein the mixing unit is designed to mechanically change the mixing of the metal powder.

21. Device according to claim 20, wherein the mixing unit comprises a doctor blade and a plate, in particular a rotary plate for receiving the metal powder, wherein the mixing unit is designed to effect the mixing at least partially by relative movement of the plate and the doctor blade to one another, in particular by rotation of the rotary plate.

21. Device (1) according to one of the preceding claims, wherein it is designed to control an energy source for sintering or melting the metal powder (2) when the processing device (13) has classified the metal powder (2) as being of sufficient quality.

22. Device (1) according to one of claims 4 to 21, wherein the device (1) is designed to control a feed device with which the metal powder (2) is provided.

23. Device (1) according to one of the preceding claims, wherein the device (1) is designed, in the event that the processing device (13) has classified the metal powder (2) as not being of sufficient quality, to cause the metal powder (2) to be removed and a new metal powder (2) to be provided by the feed device.

24. A method (29) for classifying a metal powder (2), the method (29) comprising: - illuminating the metal powder (2) with a lighting unit (8), wherein the lighting unit (8) has a plurality of lighting means (10) arranged around the metal powder (2) to enable illumination of the metal powder (2) from different horizontal angles and with different light wavelengths; - Recording image data of the metal powder (2) illuminated by the illumination unit (8) with an image recording unit (9) from the different horizontal angles and the different light wavelengths and - Processing the image data with a processing device (13), wherein the processing device (13) is operatively connected to the image recording unit (9) in terms of signals, so that the processing device (13) classifies the metal powder (2) based on the image data transmitted by the image recording unit (9).

25. Method (29) according to claim 24, wherein the illumination unit (8) enables illumination of the metal powder (2) additionally from different vertical illumination angles (11).

26. Method (29) according to one of the preceding claims 24-25, wherein the metal powder (2) can be illuminated from a same horizontal angle with different monochromatic or non-monochromatic light wavelengths.

27. Method (29) according to one of the preceding claims 24-26, wherein a trained classification algorithm is executed for classification, which carries out the classification of the metal powder (2) based on the recorded image files, wherein the results of the classification comprise an indication of a metal powder type.

28. The method (29) according to claim 27, wherein the metal powder type comprises a chemical composition of the metal powder and / or an indication of a degradation of the metal powder.

29. The method (29) according to claim 27 or 28, further comprising a supporting classification which classifies the metal powder (2) on the basis of a recorded particle size distribution of the metal powder (2) and detected oxide layers (27, 28) of the metal powder (2), wherein the particle size distribution is determined by an analysis of the image data which is recorded with the image recording unit (9) during illumination of the metal powder (2) by the illumination unit (8) with white light, and wherein the oxide layers (27, 28) are determined by an analysis of the image data which is recorded with the image recording unit (9) during illumination of the metal powder (2) by the illumination unit (8) with colored light, in particular violet light.

30. The method (29) according to claim 27, wherein the classification algorithm comprises an artificial neural network, in particular a convolutional neural network.

31. Method (29) according to claim 30, wherein the neural network is trained using image data, comprising: - inputting a set of image data of a known metal powder (2) illuminated from different illumination angles and with different light wavelengths into the neural network; - Training the neural network using a classification output of the neural network and a classification matching the image data set of the known metal powder (2).

32. Computer program product with program code for carrying out the method (29) according to one of claims 24 to 31 when the program code is executed on a computer.

33. Neural network obtainable by the method (29) according to claim 31.

Images