Method for determining a component quality of a component to be produced by means of powder bed-based fusion manufacturing

Abstract

The present invention relates to a method for determining the quality of a component (10) to be manufactured by melt processing based on a powder bed, the component being formed of a plurality of sequentially successive layers (12).

Description

[0001] The present invention relates to a method for determining the component quality of a component manufactured by powder bed-based melting, as described in claim 1.

[0002] To ensure the quality of components manufactured using Laser Powder Bed Fusion (LPBF), it is necessary to identify relevant anomalies. This is typically achieved using emission data collected through in-situ melt pool monitoring (MPM) or optical tomography (OT). The collected data is usually processed to infer anomalies in the LPBF process. For example, measured emission intensity data can indicate conditions such as localized overheating.

[0003] In a similar manner, the simulated process temperature distribution can be evaluated to identify anticipated process problems before the build process begins, and to adapt the parameters of the tool path or laser path as necessary.

[0004] The molten pool at least describes the region in the powder bed in which powder is melted into layers, for example by means of a laser.

[0005] A common method for identifying anomalies is to set a threshold for the monitored signal and mark values ​​that deviate from that threshold as anomalies.

[0006] A problem with assessing anomalies, process anomalies, or localized overheating based on simulation predictions is that not all critical areas identified at a particular layer will disrupt the manufacturing process of the produced component, especially causing process interruptions. Often, a single layer can identify too many so-called process problems, triggering an excessive number of false alarms. If, in the case of process simulation, work data is corrected based on identified alarms, such misidentification can lead to unnecessary corrections, resulting in, for example, unnecessary process pauses, and consequently, extended component build times.

[0007] Another advanced approach to processing in-situ monitoring data is the use of machine learning techniques. Essentially, machine learning algorithms for identifying anomalies in LPBF processes offer the advantage of particularly automated feature recognition. Therefore, machine learning can help identify patterns and anomalies in complex LPBF processes, for which a large amount of monitoring data must be collected and prepared to train the machine learning algorithm. Collecting sufficient data with the correct target features is crucial for the quality and accuracy of the machine learning model's predictions, as the algorithm can only discover patterns encompassed by the features of its input data.

[0008] Today, training data is collected based on standard geometries, such as cubes, bars, or cylinders, where process parameters, such as laser power, laser speed, and fill spacing, can vary. Based on these parameter variations and the collected data, it is possible to infer the porosity or crack distribution of a selected process used to manufacture the component.

[0009] When evaluating thresholds, especially for monitored or simulated data, the threshold is applied to a layer of data to identify critical regions. Subsequently, regions deemed too small can be filtered out based on size thresholds, ensuring that only critical regions exceeding the threshold size are marked as relevant. This can at least suppress some false alarms.

[0010] For example, Goh, GD, Sing, SL & Yeong, WY's "A review on machine learning in 3D printing: applications, potential, and challenges," published in *Artificial Intelligence Review*, Volume 54, pp. 63–94 (2021), outlines machine learning in 3D printing. Furthermore, Felix, S., Ray Majumder, S., Mathews, HK, et al.'s "In situ process quality monitoring and defect detection for direct metal laser melting," published in *Scientific Reports*, Volume 12, p. 8503 (2022), demonstrates in-situ process quality monitoring and defect detection in direct metal laser melting.

[0011] The technical problem to be solved by the present invention is to provide a method by which the quality of components can be determined or, in particular, predicted in a particularly advantageous manner before or during the manufacture of components by powder bed-based melting, and thereby using particularly efficient training data and / or being able to advantageously ignore anomalies unrelated to component quality.

[0012] The technical problem described herein is solved by the technical solution provided in the independent claims. Advantageous designs and improvements of the invention, as well as related advantages, are given in the dependent claims, the specification, and the drawings.

[0013] A first aspect of the invention relates to a method for determining the quality of a component to be manufactured by melt processing based on a powder bed, the component being capable of being manufactured or constituted by a plurality of sequentially successive layers. The method of the invention is based on at least one dataset and / or on a modeling model, the dataset and the modeling model being formed by means of layer data of corresponding layers, or using such layer data as input, the corresponding layers being corresponding layers of at least a portion of sequentially successive layers of the component and / or corresponding layers of a model component.

[0014] Powder bed fusion is an additive manufacturing process in which materials or components are bonded layer by layer to produce or generate a workpiece or component from 3D model data. In powder bed fusion, selective regions of a powder bed are joined or melted by thermal energy, particularly with the aid of a laser beam. The materials used for the component are, in particular, metals and / or plastics. The powder bed itself or the powder typically forms a good thermal insulator. The dataset contains, in particular, information about critical or anomalies in the multiple bonded layers. The modeling model is particularly based on machine learning methods and, for example, can predict anomalies based on a 3D model of the component to be manufactured. Layer data can, in particular, come from measurements such as melt pool monitoring (MPM) or optical tomography (OT). Additionally or alternatively, layer data can be obtained or generated through simulation of the melting process. The model component may have a different shape than the component, but it can also be a component itself and used to acquire layer data based on measurements.

[0015] In other words, in the method for determining component quality according to the present invention, a dataset is used that summarizes information on multiple sequential layers to be manufactured for producing a component by means of melting, in order to determine criticalities or anomalies that occur during the manufacturing process. The layer data on which the dataset is based can be simulated or measured. Alternatively, in the method for predicting or determining component quality according to the present invention, a modeling model is used based on the layer data, which is capable of evaluating component quality based on the layer data, wherein the modeling model employs, in particular, machine learning methods.

[0016] Therefore, the method has the advantage of utilizing layer data from multiple sequentially successive layers, particularly compared to existing solutions, in order to assess the relevance of key regions or identify key vectors, especially illumination vectors. Another advantage is that overheating, which causes surface warping and so-called molten beads, can be detected with exceptional reliability.

[0017] Another advantage is that machine learning models can be particularly compact and require very little experimental training to construct, while still achieving remarkably accurate results.

[0018] In an advantageous embodiment of the invention, during the manufacture of the model component or component, for the layer data of the corresponding layer, measurement data is recorded for the corresponding layer by a sensor device, and / or the layer data is calculated by simulation, wherein the model component may be the component itself. In other words, a sensor device comprising at least one sensor is provided, which records data, particularly during the manufacture of the component or model component, wherein the sensor data generated by the sensor device may, for example, include emission intensity data or long-term exposure data from a camera. Alternatively, the layer data is simulated, particularly by a simulated distribution. The layer data is then derived from the measurement data or simulation. The resulting advantage is, for example, that the layer data can be formed in a particularly accurate manner because this layer data is based on the measurement data or simulation data.

[0019] In another advantageous embodiment of the invention, for the model component, a component model or 3D model is used, the geometry or shape of which predefines different characteristics for multiple irradiation vectors used or available during the manufacture of the model component, thereby allowing for variations in characteristics among the multiple irradiation vectors. Here, the characteristics may in particular include position, vector length, minimum mass integral, vector orientation, or other features. In other words, the model component is predetermined based on a three-dimensional model, i.e., a component model, wherein the geometry or shape of the component model is selected such that, particularly for different layers, the irradiation vectors are different, defining the movement of the laser within that layer in the powder bed. The layers of the component may, for example, be defined by a triangular shape on their outer periphery, thereby allowing for variations in the vector length of the irradiation vectors within the layer to be formed. For example, variations in different characteristics are advantageous.

[0020] The characteristics of a vector can be its position, particularly its position on the build plate, thus including the xy coordinates of the center of the exposure vector. A second characteristic can be the vector length. Alternatively, the characteristic can be the vector time. Another characteristic is the minimum or average mass integral, which determines the maximum or average vector temperature, and can be determined through simulation or semi-analytical methods. The mass integral determines how much of a hemisphere or ellipsoid with a specific, typically 1 mm radius, is contained within the part. To determine the minimum mass integral, test points can be sampled along the irradiation vector at regular intervals, typically 0.1 mm. For these samples, the mass integral can be evaluated and its minimum calculated. The mass integral here depends on the environment around the irradiation point, particularly on whether there is a large amount of molten material or powder surrounding the irradiation point, as powder is a good thermal insulator compared to molten material and can therefore lead to localized overheating. Another characteristic can be the vector orientation. Additional optional features, such as process parameters like laser power, laser speed, fill spacing, etc., can also be considered. By generating model components that cover as many illumination vectors as possible, or large variations in the characteristics of these illumination vectors, the advantage is that a wide distribution of model input parameters can be achieved, thereby obtaining favorable training data for the modeling model.

[0021] In an advantageous embodiment of the method, measurement data acquired during the manufacture of the model component, along with the component model itself (which may also contain information about the irradiation vector or possible manufacturing processes), is used as training and / or validation data for a machine learning method that provides or forms the modeling model. In other words, data from a model component with a specific geometry is used as training data to cover a wide range of variations in the irradiation vector. For this purpose, the model component is manufactured based on a powder bed-based melting process, and the resulting measurements or measurement data are used as layer data for at least a portion of the layers and layer data. This data and information from the component model are aggregated and can be used as training data. The machine learning method or modeling model can, for example, be a self-learning algorithm and / or a neural network. The resulting advantage is that, due to the shape of the component model and the model component manufactured using a pre-defined component model, particularly advantageous measurement data can be generated, thus the training data can be particularly advantageously used to determine component quality.

[0022] In another advantageous embodiment of the invention, the intensity distribution or emission intensity distribution is used as measurement data, determined by melt pool monitoring (MPM), particularly in-situ melt pool monitoring, and / or long-exposure imaging for optical tomography. Long-exposure imaging specifically corresponds to exposure imaging during the formation of each layer. In other words, the emission of the melt pool during the process is monitored, particularly automatically, wherein at least one sensor detects the emission of the melt pool layer by layer, particularly in a position-resolved manner. If the melt pool is too cold or too hot, this can be detected in the measurement data. In optical tomography, optical imaging is acquired, for example, by long exposure of each layer using a camera, making it possible to detect the light distribution of all illumination vectors constituting that layer. Multi-layer images can be generated, for example, using tomography algorithms. The resulting advantage is that the measurement data can be used particularly advantageously to perform the method.

[0023] In another advantageous embodiment of the invention, layer data describes criticality. In other words, for example, measurement results or measurement data, such as intensity distribution, are reinterpreted in the layer data, for example, through classification. By classifying criticality, the risk of anomalies at different locations in the corresponding layers of the component can be assessed. For example, corresponding measurement points in the corresponding layers can be classified as, for example, subcritical, semi-critical, or critical in relation to the formation and quality of the component. The resulting advantage is that layer data can be used particularly advantageously to determine component quality.

[0024] In another advantageous embodiment of the invention, for layer data, multiple layers, particularly no more than 10 layers, preferably 3 to 5 layers, are arithmetically and / or logically combined, thereby deriving at least one anomaly for a component. This anomaly can describe the component quality. In other words, information from multiple layers is arithmetically combined, i.e., combined by statistical methods such as averaging, or logically combined, for example, based on criticality. For example, if in three consecutive layers, there is a particularly high intensity at the xy coordinate only in one layer, while there is normal intensity at the same xy coordinate in the other two layers, then the absence of a related anomaly can be inferred by, for example, the arithmetic mean at that point. This similarly applies to logical combining; for example, if, in the case of three layers having the same xy coordinate, only one criticality is displayed and two sub-criticalities are displayed, then the absence of an anomaly can also be inferred. The resulting advantage is that the method for determining component quality can be performed particularly stably.

[0025] In another advantageous embodiment of the invention, key regions are formed from layer data based on point data and / or key vectors are formed based on point data and / or key vectors are formed based on vector criticality, thereby deriving component quality. In other words, key regions are formed from point data distributed on corresponding illumination vectors in multiple layers, wherein a single-layer criticality bitmap can be generated. Alternatively or additionally, key regions can be determined in multiple layers by not performing contour search, but by calculating the overlap with the key regions for the illumination vectors of a layer, and marking the vectors or illumination vectors as critical if the overlap threshold is exceeded. Additionally or alternatively, a single-layer vector criticality index is generated, wherein pixel values ​​of a single-layer bitmap can be created along the vectors. The resulting advantage is that various questions about the formed layers can be answered, such as whether a region, vector, or illumination vector is critical. Therefore, this method can be particularly advantageously used to determine component quality.

[0026] In another advantageous embodiment of the invention, the model components are manufactured multiple times for training data, particularly in different orientations to fill the build space. In other words, multiple variations of the component model are arranged in a usable powder bed such that they are printed individually and independently, while simultaneously filling the build space or powder bed as completely as possible. This allows for the acquisition of measurement data at multiple locations and multiple vectors or illumination vectors via appropriate optical tomography or in-situ melt pool monitoring, enabling the model or modeling template to be trained particularly advantageously. The resulting advantage is that component quality can be evaluated particularly advantageously using machine learning.

[0027] In another advantageous embodiment of the invention, the component model uses a scaling factor. In other words, the component model is provided such that different layers, although having substantially the same shape, are scaled at different ratios. This allows, for example, the creation of shapes of particular interest at the edges, so that measurement data of these shapes can be obtained in a particularly advantageous manner during the manufacture of the model component, thereby enabling the prediction of component quality in a particularly advantageous way.

[0028] For application situations or scenarios that may occur in the method but are not explicitly described herein, it may be stipulated that the method outputs error reports and / or requests for user feedback, and / or sets standard settings and / or predefined initial states.

[0029] In this patent application, nouns and pronouns referring to people generally do not specify a particular gender.

[0030] In the attached diagram:

[0031] Figure 1 A schematic side view of an apparatus for melt manufacturing of components based on a powder bed is shown;

[0032] Figure 2 A schematic perspective view of a component model is shown, which can be manufactured as a model component using this method;

[0033] Figure 3 It shows according to Figure 1 A perspective view of a powder bed of an apparatus, wherein multiple [unclear] are arranged in the powder bed. Figure 2 A copy of the component model shown;

[0034] Figure 4 It shows Figure 2 The top view of the component shown, with corresponding illumination vectors on two layers;

[0035] Figure 5 A top view of the layers of another component model is shown; and

[0036] Figure 6 It shows having Figure 5 A three-dimensional view of the component model of the layer shown.

[0037] Based on the accompanying drawings, a method is described for determining or evaluating the quality of a component 10 to be manufactured by melt processing based on a powder bed, the component being manufactured or formed from a plurality of sequentially successive layers 12.

[0038] To manufacture component 10, an apparatus 16 with a powder bed 14 is used. This apparatus has a light source 18, which in particular can provide a laser beam 20 that moves along an irradiation vector 22 to selectively melt the powder 24 contained in the powder bed 14 at specific locations within the powder bed 14. Thus, after the laser beam 20 moves along the irradiation vector 22, a layer 12 of component or model component 26 is produced or formed. For other layers 12, a build platform with the manufactured portion of component 10 is lowered, and new layers are applied and melted on it with new powder 24. The bottom of the powder bed can be formed, in particular, by a build plate 32.

[0039] Powder 24 may be, in particular, a metallic material and / or a plastic, wherein the material may be mixed with other materials, such as ceramics.

[0040] In the method for determining the component quality of component 10, the component quality is determined based on at least one dataset and / or based on a modeling model, the dataset and / or modeling model being formed by means of layer data of a corresponding layer 12, the corresponding layer being a corresponding layer of at least a portion of the layers 12 of sequentially successive layers of component 10 and / or a corresponding layer 12 of model component 26.

[0041] In other words, the dataset is provided, in particular, through layer data. Additionally or alternatively, the layer data can be used as input to a modeling model, which is provided, in particular, through machine learning methods such as self-learning algorithms and / or neural networks, wherein layer data can additionally be generated as model data or training data for the modeling model. Thus, by training the modeling model using, for example, training data based on layer data of model component 26, the modeling model is able to predict the component quality of component 10. For this purpose, the trained modeling model obtains a three-dimensional model of component 10 as input.

[0042] Advantageously, the layer data for the corresponding layer 12 can be formed from measurement data during the manufacturing of model component 26 and / or component 10, and this measurement data is recorded by sensor device 28 for the corresponding layer 12. Alternatively or additionally, the layer data can be calculated or simulated.

[0043] The measurement data used in the method may advantageously be the intensity distribution of melt pool monitoring (MPM) and / or long-exposure images for optical tomography.

[0044] Based on a dataset, particularly containing measured or simulated layer data, component quality can be evaluated, for example, by assessing the relevance of critical regions contained in sensor and / or simulated data or layer data. This evaluation is performed by combining a configurable number of sequentially occurring layers 12 or their layer data. Typically, fewer than 10 layers 12 can be combined, preferably 3 to 5 single layers or layer 12 measurements or measurement data and / or criticality. In this combination, arithmetic operations can be performed, such as averaging the measurements of each individual layer 12 or single-layer measurements, or logically associating single-layer criticality indices. This allows the layer data of the corresponding layer 12 to describe criticality. Criticality can be represented, for example, by a two-dimensional distribution, particularly in the form of a two-dimensional distribution with a configurable list of bitmaps, such that for each layer 12, the corresponding layer data exists as this bitmap. This allows each pixel to correspond to a point of criticality in a 2D region (typically the lateral dimension of component 10).

[0045] Preferably, multi-layer keyness is formed by averaging the measurements of each layer. Thus, a multi-layer keyness bitmap can be derived or formed, for example, by comparing thresholds of an average distribution. If single-layer keyness is used as input data, then, for example, counting of single-layer keyness can be used. For example, if at least two of the single-layer keynesses considered are set as key, then a value is set to 1 in the spatial multi-layer keyness distribution.

[0046] In order to assess component quality based on anomalies such as excessively high melting temperatures caused by irradiation vector 22, it is necessary to identify critical irradiation vectors 22 that need to be corrected if necessary. For example, the overlap between the irradiation vector 22 of layer 12 and critical regions can be determined. Here, to identify critical regions, contour search and, optionally, size filtering can be performed.

[0047] In general, for the detection of multi-layered critical areas, the following situations can or should be distinguished:

[0048] In the first case, multiple key regions can be detected from point data.

[0049] Based on data points distributed along the illumination vector 22 (the data points being, for example, based on MPM intensity values ​​or simulated temperature), a single-layer criticality bitmap is created by setting the pixel values ​​of the bitmap to the corresponding measured values. Here, either the raw values ​​of the measured data can be used directly, or criticality indices, such as subcritical, semi-critical, or critical, can be determined. These criticality indices are generated from the raw data through threshold comparisons and can be used as logical data.

[0050] When generating bitmaps, care should be taken to ensure that no undefined pixel remnants remain within the illuminated faces (especially faces in layers of 12). Otherwise, faces with holes may be generated when subsequently associating individual bitmaps. This can be achieved by using a sufficiently low bitmap resolution, i.e., larger pixels, or by a bitmap dilation process, such as through morphological closing operations. Multi-layer critical bitmaps are then formed from the individual bitmaps through arithmetic or logical associations. Subsequent contour searches (which describe, for example, the length or perimeter of the faces) and optional size filtering can then characterize critical regions or anomalies in multi-layer data involving multiple layers.

[0051] In the second case, multi-layer key vectors are detected from point data.

[0052] Here, similar to the first case, multiple key regions are identified; however, contour search is not required. Subsequently, for the illumination vector 22 of a layer or layer 12, the overlap with the key regions is calculated, and the illumination vector 22 is marked as key if the overlap exceeds a threshold. For performance reasons, the overlap calculation can be performed by an electronic computing device using equidistant sampling points of the key on the counting vector or illumination vector 22. Alternatively, if the key regions are determined beforehand via contour search, the length of the vector polygon overlap can also be determined.

[0053] Therefore, performing the method according to the first case can determine the region, while according to the second case, it can, for example, answer the question of how long the corresponding illumination vector 22 has been running at the key point, thereby determining, for example, whether the vector itself is key.

[0054] Alternatively, or as an alternative, multi-level key vectors can be detected from vector criticality based on the third case.

[0055] In this case, a single-layer vector criticality index (e.g., non-critical and critical) can be used as input data.

[0056] The pixel values ​​of a single-layer bitmap are set accordingly along the vector or illumination vector 22.

[0057] Following an optional dilation step, the bitmaps are simulated and / or logically combined into multi-layer key bitmaps, and then, as in the second case, overlap in the illumination vectors of one layer is determined. Similar to case 2, multi-layer keyness of the vectors is derived through overlap threshold comparisons.

[0058] In the described scenario, the corresponding layer data or the respective individual multi-layer bitmaps are used as a dataset, based on which the quality of the component can be determined, since key regions or vectors can be identified, for example, through the dataset.

[0059] To assess the relevance of critical areas or identify key vectors of measured simulation data or simulated measured data (e.g., temperature), multiple sequentially consecutive layers 12 are used. This allows for particularly reliable detection of overheating, especially that leading to surface warping or molten beads in component 10, more reliably than evaluating each layer individually. Therefore, the proposed method is particularly advantageous for determining component quality.

[0060] Additionally or alternatively, a modeling model may be used, which is provided in particular by or with the aid of machine learning methods. Here, for model component 26, component model 30 may be used, whose geometry or shape can ensure that the illumination vector 22 used to manufacture model component 26 varies, thereby giving the multiple illumination vectors 22 of model component 26 different characteristics.

[0061] Model component 26 or component model 30 is used in particular as a reference component, and its shape is designed to distribute the feature values ​​of the machine learning algorithm and / or neural network as evenly as possible. Therefore, a list of features for the machine learning algorithm can be described first. Model component 26 can be manufactured, for example, through a copy of itself, especially to fill the construction space for training data.

[0062] Each layer 12 of the model component 26 is manufactured using an illumination vector or a set of illumination vectors 22. For example, Figure 4 The illumination vectors 22 of the two layers 12 are shown. The characteristics or properties of the illumination vectors 22 are advantageously varied, especially between the layers 12.

[0063] Here, the first characteristic to be changed could be the position described on the build plate 32 used to form the bottom of the powder bed 14. The build plate 32 could, for example, be provided with an xy coordinate system, and the characteristic of the corresponding irradiation vector 22 could represent the position or path on the build plate 32, wherein the "position" characteristic could be described using the center of the irradiation vector 22 or alternatively using the coordinates of the vector's start and end points. For a circular build plate 32, polar coordinates could be used, for example, instead of xy coordinates to represent the position.

[0064] The second characteristic could be the vector length, or alternatively, the vector time or the minimum or maximum repetition time. The vector time here is the time required to cover the illumination vector 22. The minimum repetition time is the time between the vector's starting point and the previous illumination time of the point closest to the vector's starting point. The maximum repetition time could be the time between the vector's ending point and the previous exposure time of the point closest to the vector's ending point.

[0065] The third characteristic could be the minimum mass integral, or alternatively, the maximum mass integral, the average mass integral, or the average vector temperature, which can be determined, in particular, through simulation and by means of semi-analytical methods. The mass integral determines how much of a hemisphere or ellipsoid with a specific radius, typically about 1 mm, is contained within the irradiated component. To determine the minimum mass integral, for example, the irradiation vector 22 can be scanned at test points at regular intervals, typically 0.1 mm. For these samples, the integral is evaluated and its minimum value is calculated.

[0066] Another characteristic could be vector orientation, or alternatively, fringe orientation. Here, vector orientation is the angle between the exposure vector or illumination vector 22 and the horizontal axis. Fringe orientation can represent the direction of a series of illumination vectors.

[0067] In addition, other features exist, such as process parameters like laser power, laser speed, or fill spacing. These features can be selectively used, particularly when it is necessary to evaluate and / or combine different sets of process parameters.

[0068] The process parameter set may include, for example, measurement data recorded by sensor device 28.

[0069] Figure 2 The model component 26 or the three-dimensional component model 30 based on the model component 26 is shown. Due to its advantageous geometry, the three-dimensional component model can have special variations in terms of the characteristics of the just-outlined illumination vector 22 that can be used in its manufacture.

[0070] therefore, Figure 2The component model 30 has a favorable geometry for machine learning training, for which it is divided into seven segments 34 along the height direction. The number of segments 34 is arbitrary and merely exemplary. Thus, the model component 26 to be manufactured can be made with more or fewer segments 34.

[0071] The height of each segment 34 is specifically equal to the thickness of the corresponding layer 12 to be manufactured by the device 16 multiplied by the number of different layer orientations to be studied or used for layer data.

[0072]

[0073] The order of the stripe orientations in each segment 34 can be the same. This order can be chosen to ensure that the orientations are uniformly distributed within the segment 34. The actual choice of the number of different layers and orientations can be, for example, n. orient =36 or n orient =72 or 360, or other sufficiently large integer factors (>18). The order of stripe orientation in each segment can be designed as follows:

[0074]

[0075] Alternatively, other regular distributions of orientation angles may be used. One condition for such distributions is that the two successive layers 12 are sufficiently different in orientation, and in particular, for example, offset or rotated more than 45° relative to each other.

[0076] For the training geometry of model component 26 used for training data, the same fill pattern used for component 10 can be used. Here, a checkerboard pattern, for example, 5 to 10 mm wide stripes or similar sizes, is typically used depending on the component size. Figures 2 to 4 For example, a 10mm wide stripe pattern can be used.

[0077] The lower segment 34 of the geometry of component model 30 can be designed as a cylinder. Advantageously, the cylinder can have a diameter of, for example, 12 to 16 mm. An excessively large cylindrical base would result in an excessively long illumination vector 22 and would lead to the generation of too many stripes of similar size.

[0078] Conversely, if the cylinder is too small, the long illumination vector 22 is not covered during training. The advantageous geometry of the component model 30 is intended to distribute the features or properties as evenly as possible within layer 12 and throughout the entire geometry.

[0079] exist Figure 2 In the example shown, the diameter of the cylinder is, for example, 15 mm.

[0080] The second segment 34, namely the first segment above the cylinder, and subsequent segments 34, are specifically formed by triangular prisms twisted relative to each other. The base of each triangular prism is an equilateral triangle inscribed in the base circle (the circumference of the cylinder). Alternatively, other regular polygons or other shapes may be used instead of triangles, as long as a broad distribution of the length of the illumination vector 22 is achieved. The triangles do not need to be precisely inscribed in the circle. Slightly smaller triangles are also suitable.

[0081] Therefore, each segment comprises twisted prisms or triangular prisms stacked in descending order of twist angle. These twists ensure that the features of the minimum mass integral are uniformly covered in the training data. For example, for nickel-based alloys, overhang angles up to 40° can be manufactured without overheating. Therefore, Figure 2 The torsion angles of the six sections 34 in the upper middle part are selected as follows: 90°, 80°, 70°, ..., up to 40°.

[0082] For process parameters or material settings that are less prone to overheating, a profile with a smaller minimum overhang angle can be used.

[0083] As mentioned earlier, the number of segments 34 is arbitrary. Other decreasing sequences down to 40° can also be used.

[0084] The order of the torsion angles is decreasing because only the upper segment 34 is affected under overheating conditions, thus it can be easily excluded from the training dataset. Typically, materials prone to overheating require fewer twisted prism segments 34 (the upper segment 34 has a larger torsion angle).

[0085] Alternatively, a continuous function can be used to represent the torsion angle instead of segment 34.

[0086] Figure 3 The distribution of multiple model components 26 on the construction plate 32 is shown, wherein an advantageous distribution is that the second component model of every two model components 30 is additionally rotated about its center, i.e., adjacent geometries have different orientations. In the case of triangles, a 60° rotation may be advantageous.

[0087] Figure 3 The rotation of nine adjacent model components 26 is shown, wherein more than two different rotations or orientations may also be used to achieve a favorable distribution on the building plate 32.

[0088] Figure 4 An example of a striped pattern is shown in two adjacent sections rotated by 60°, where the lines represent illumination vector 22. Rotating by 60° contributes to a better distribution of vector quantity characteristics.

[0089] The model components 26 are distributed as evenly as possible, especially so that the largest possible area on the building plate 32 is covered by the model components 26, thereby maintaining a very small gap between adjacent model components 26.

[0090] In powder bed-based melting processes, the typical small spacing between components can be 5 to 15 mm.

[0091] In addition, the edges and corners of the rectangular building plate 32 should be covered as much as possible. However, for example, if the fastening elements of the building plate 32 are located nearby, some corners may not always be covered.

[0092] For the circular construction board 32, it is recommended to cover the perimeter as close as possible to the board boundary to obtain favorable training data.

[0093] For training data, it is particularly advantageous to manufacture model components multiple times, especially in different orientations and ways of filling the building space.26

[0094] Depending on the filling of the build plate 32, for training data, there may be situations where it is necessary to extrapolate areas in the build space or powder bed 14 where no model components 26 have been formed. Since there must be gaps between the model components 26, there will always be some uncovered xy positions. If the data collected based on measurements cannot be interpreted well enough, the build plate 32 can be filled again with model components 26, which are built into the previously left gaps, so that the training data can be compensated for for each xy position.

[0095] If multiple process parameters, such as laser power, laser speed, and / or fill spacing, need to be studied, different sets of parameters should be used in repeated builds or fabrications. In contrast, lower / upper skin parameters and corresponding exposure strategies can be integrated into the same build or fabrication.

[0096] Figure 5 and Figure 6 Another advantageous geometry of component model 30 is shown, in which, Figure 5 Layer 12 is shown in particular, which is used in multiple layers or layers 12 of the component model 30 by means of a scaling factor.

[0097] Figure 5 and Figure 6 The shape or geometry of the embodiment of model component 26 can be particularly advantageous in order to determine the correctness of predictions by machine learning algorithms, and the modeling model can be based on such correctness, for example.

[0098] The structure of component model 30 and Figures 2 to 4The structure is similar, but the difference lies in that the lower cylinder has a larger diameter, for example, 23 mm, to form a star-shaped structure. Alternatively, the cylinder can be omitted, and the lower part can begin directly with a star-shaped segment 34. Figure 5 One of the layers, 12, is shown.

[0099] The star-shaped segment can be similar to Figures 2 to 4 The triangular prism is twisted. Furthermore, a scaling factor can be applied from bottom to top. The scaling factor, or growth scaling factor, can correspond to the same angle chosen for the twist in each segment 34; that is, mathematically, the factor can be calculated as follows:

[0100]

[0101] Alternatively, the scaling factor could also be f. growth =1. Alternatively, a large scaling factor can be used without twisting, so that the component has only a straight star geometry.

[0102] As the basic geometric shape or form of the lower cylinder, an inscribed octagon can be chosen, such as... Figure 5 and Figure 6 As shown. Other polygonal options can also be selected as alternative locations.

[0103] The polygon or octagon forms a hole at its center, which represents the area in the corresponding layer 12 that is generally of secondary significance for evaluating component quality. This hole is particularly useful for saving material.

[0104] Since the information obtained from the center of the structure through monitoring the molten pool is not important, model component 26 is used to detect anomalies at the edges and in the outer regions where they appear. In another variation of model component 26, only one or more arms in a star shape can be manufactured to similarly save material or material, especially in the form of powder 2.

[0105] In this form of component model 30, the number of segments 34 can also be arbitrarily chosen. Figure 5 and Figure 6 The embodiment has five segments 34, which have a twisted star shape and an increased scaling factor.

[0106] According to Figures 2 to 4 Compared to the embodiment of model component 26, the torsion angle is chosen more aggressively.

[0107] Therefore, the torsion angle at the top of the structure is especially less than 40°. For the material or components, nickel-based alloys can be used, and the following torsion angles can be selected: 80°, 60°, 45°, 35°, and 25°.

[0108] The resulting geometry is used as a good test of the output of machine learning algorithms and / or neural networks, i.e., the output of modeling models, because it contains both the (lower) segment 34, which can be constructed without anomalies, and the (upper) segment 34, which in particular may exhibit overheating anomalies.

[0109] In practice, multiple model components 26 can also be distributed across the construction plate 32 for printing. The rules for covering the construction plate 32 can be consistent with the use of... Figures 2 to 4 The training phase of model component 26 is less stringent than that of component model 30. Therefore, model component 26 can be placed on the construction plate 32 almost arbitrarily.

[0110] Therefore, this paper demonstrates how the method for determining component quality when using a modeling model is presented. The proposed method allows for the advantageous determination of component quality based on at least one dataset containing layer data, i.e., also based on a modeling model containing layer data. Thus, for example, it is advantageous to avoid the incorrect manufacturing of component 10.

[0111] Therefore, the method is applicable to both multi-layered criticality assessment of quality assurance in LPBF processes and to reference models for identifying anomalies through in-situ monitoring in laser powder bed melting processes.

[0112] List of reference numerals

[0113] 10 components

[0114] 12 floors

[0115] 14 Powder Bed

[0116] 16 devices

[0117] 18 Light Sources

[0118] 20 laser beams

[0119] 22 Irradiation Vector

[0120] 24 Powder

[0121] 26 Model Components

[0122] 28 Sensor Devices

[0123] 30-component model

[0124] 32 Construction board

[0125] Section 34

Claims

1. A method for determining the component quality of a component (10) to be manufactured by melt-processing based on a powder bed, the component comprising a plurality of sequentially successive layers (12), the method being implemented based on at least one dataset and / or based on a modeling model formed by means of layer data of a corresponding layer (12), the corresponding layer being a corresponding layer (12) of at least a portion of the sequentially successive layers (12) of the component (10) and / or a corresponding layer (12) of a model component (26).

2. The method according to claim 1, characterized in that, During the manufacturing of the model component (26), measurement data for the corresponding layer (12) is recorded by the sensor device (28) for the corresponding layer (12), and / or the layer data is calculated by simulation.

3. The method according to claim 2, characterized in that, For the model component (26), a component model (30) is used. The shape of the component model is preset with different characteristics for multiple illumination vectors (22) used in manufacturing the model component (26), so that there are characteristic changes in the multiple illumination vectors (22).

4. The method according to claim 3, characterized in that, Measurement data collected during the manufacturing of the model component (26) and the component model (30) are used as training data and / or validation data for providing machine learning methods for modeling the model.

5. The method according to any one of claims 2 to 4, characterized in that, Intensity distribution monitored by the molten pool and / or long-exposure images used for optical tomography are used as measurement data.

6. The method according to any one of the preceding claims, characterized in that, The layer data description is crucial.

7. The method according to any one of the preceding claims, characterized in that, For layer data, multiple, especially up to 10, preferably 3-5 layers (12) are arithmetically and / or logically merged, thereby deriving at least one anomaly for component (10), which describes the component quality.

8. The method according to any one of the preceding claims, characterized in that, The quality of the components is derived from the key regions formed from the layer data based on point data and / or the key vectors formed from the layer data based on point data and / or the key vectors formed from the layer data based on the key vectors.

9. The method according to any one of claims 2 to 8, characterized in that, For training data, model components are fabricated multiple times in different orientations to fill the building space (26).

10. The method according to any one of claims 3 to 9, characterized in that, The component model uses a scaling factor.

Images