Computer-implemented method for generating 3D points in space

Abstract

A computer-implemented method (100) for generating 3D points in space for component positioning in a complete tool (10) with a plurality of components (12) for forming the complete tool (10), wherein the complete tool (10) is a printing tool, in particular a printing die, or a drawing tool, in particular a drawing die (20), the method (100) comprising the steps of: - reading (110) a digital pattern (14) of a body part (16) to be produced by the complete tool (10) into a trained machine learning model; - creating (120) a digital contour profile (18) by a CAD system, in particular by mapping it onto the complete tool (10); - predicting position points and orientations of components (12) for forming the complete tool (10) by the trained machine learning model, based on the contour profile.

Description

[0001] The invention relates to a computer-implemented method for generating 3D points in space for component positioning in a tool comprising a plurality of components. The invention relates to a computing device. The invention relates to a computer program product. The invention relates to a storage device.

[0002] US 2021 / 0397142A1 describes methods, systems, and devices, including computer program products encoded on a medium, for the computer-aided design and manufacture of physical structures using toolpaths generated by reinforcement learning, for use with subtractive manufacturing systems and techniques, which include: obtaining a three-dimensional model of a manufacturable object in a computer-aided design or manufacturing program;Generating toolpaths that can be used by a computer-controlled manufacturing system to produce at least a part of the manufacturable object by providing at least a part of the three-dimensional model to a machine learning algorithm that uses reinforcement learning, wherein the machine learning algorithm includes one or more evaluation functions that contain rewards that correlate with desired toolpath properties, including toolpath smoothness, toolpath length, and collision avoidance with the three-dimensional model; and providing the toolpaths to the computer-controlled manufacturing system to produce at least that part of the manufacturable object.

[0003] US Patent 2023 / 0084639A1 describes a system and method for extrapolating and automating 3D engineering drawings using machine learning (ML). In this description, a 3D computer model of a part to be manufactured is received, and the model is automatically decomposed into labeled surfaces that can be assigned, mapped, and represented by 2D drawings. One or more subprocesses receive data defining the attributes of the 2D drawings and perform calculations to predetermine the positions of drill holes on a machined part. The system then identifies any unintended gaps, overlaps, or other irregularities on the part. The system generates a list of all irregularities and sends a punch list to a human operator for correction.The system uses Amazon Web Services (AWS) to extract data and smooth the 3D model, select the optimal size of the machine material, and optimize its orientation in relation to the head of the manufacturing machine.

[0004] SHAMSUZZOHA, Ahm, et al describe “Application of machine learning algorithm in the sheet metal industry: an exploratory case study”. in: International Journal of Computer Integrated Manufacturing, 2022, Volume 35, No. 2, pp. 145-164.

[0005] Complete tooling systems are known from the prior art. They are used, for example, to manufacture body sheet metal parts. This requires arranging a large number of components that make up the complete tooling system. The positioning and alignment of these components is very complex and therefore perceived as cumbersome.

[0006] The purpose of the invention is to improve the manufacturing of a complete tool.

[0007] The problem is solved in particular by a computer-implemented method for generating 3D points in space, especially for component positioning in a complete tool comprising a plurality of components for forming the complete tool, with the features of claim 1. The problem is solved in particular by a computing device with the features of claim 7. The problem is solved in particular by a computer program product with the features of claim 8. The problem is solved in particular by a storage device with the features of claim 9. Further features and details of the invention will become apparent from the dependent claims, the description, and the drawings.Features and details described in connection with the computer-implemented method for component positioning in a tool comprising a plurality of components naturally also apply to the method for training a machine learning model, the computer program product, the computing device, and the storage device according to the invention. Conversely, the disclosure of the individual aspects of the invention is always, and can always be, mutually interdependent.

[0008] According to one aspect, the problem is solved in particular by a method with the features of claim 1.

[0009] This involves developing a computer-implemented method for generating 3D points in space. The method is designed for component positioning within a complete tool comprising a multitude of components. The complete tool is a printing tool, in particular a printing die, or a drawing tool, in particular a drawing die. The method includes the step of reading a digital pattern of a body part to be produced by the complete tool into a trained machine learning model. The method includes the step of creating a digital contour profile using a CAD system, in particular by mapping it onto the complete tool. The method includes the step of predicting the position points and orientations of components for forming the complete tool using the machine learning model, based on the contour profile.

[0010] The computer-implemented method can be designed to generate 3D points in space. These points can be used, for example, for component positioning and for design tasks.

[0011] The computer-implemented method can be designed for component positioning within a complete tool containing a large number of components. Specifically, it is a method in which digital models of the body part can be read into a trained machine learning model. A digital contour profile can be created that can predict the position and orientation of all components within the complete tool, particularly through a CAD system, especially a CAD software product. The contour creation process can be automated.

[0012] In one embodiment, the digital contour is not generated by the trained machine learning model, but rather the digital contour is (merely) read into the trained machine learning model. The digital contour can be generated automatically in the CAD system and then read into the trained machine learning model.

[0013] The CAD system generates, in particular, a digital contour profile that can be read into the machine learning model. The machine learning model then provides, as a result, a prediction for the position and orientation of components, especially all components.

[0014] One possible approach to implementing this process would be to use artificial neural networks (ANNs) to train the model and make predictions. ANNs are a form of machine learning that enables the recognition of complex relationships between a model's inputs and outputs and the making of predictions. Another approach would be to use evolutionary algorithms (EAs), such as genetic algorithms or particle swarm optimization (PSO), to find the optimal positioning pattern for the overall tool.

[0015] Predicting the position points and orientations of components within the overall tool can also be achieved using other machine learning algorithms such as Random Forest or Support Vector Machines (SVM). These algorithms can increase the complexity of the model by using a variety of decision trees or classifiers to make predictions.

[0016] Another possible approach is to use a computer-aided optimization (CSO) method to find the best possible position and orientation of the components within the overall tool. CSO methods can help to consider a multitude of variables and find the optimal compromise between different objective functions.

[0017] It is also possible to use a component positioning method that combines several methods. For example, a KNN-based method can be combined with an EA to find an optimal solution for positioning the components within the overall tool.

[0018] Component positioning refers in particular to the sorting and arrangement of the individual components of a tool to form a complete tool.

[0019] The complete tool is, in particular, the final tool that is assembled from the individual components.

[0020] Reading a digital pattern of a body part is, in particular, the process in which a digital model of the body part is loaded into a computing device, such as a computer system, and read by it.

[0021] The body part is, in particular, a part of a car body produced by the interaction of the components of the overall tooling, such as a body sheet metal part, which contributes to the shaping of the body of the final product.

[0022] A digital contour profile is, in particular, a digital representation of the contours of a body part or other part in the overall tooling, which is created by a computer-implemented method.

[0023] Position points are, in particular, the exact points where a component should be located in the overall tool, especially according to a prediction by the computer-implemented method using the trained machine learning model.

[0024] The orientations of the components refer specifically to the direction of a component within the overall tool. A reference can be made to an overall axis construction of the entire tool to determine the orientation of the individual components.

[0025] Predicting position points and orientations refers in particular to the method in which a machine learning model is used to predict the position points and orientations of the individual components in the overall tool.

[0026] From one perspective, the overall tool can comprise a printing tool, in particular a printing die, or a drawing tool, in particular a drawing die. A drawing die is specifically a tool for machining metal parts, in which the die is pressed into the material by a mechanical process to create a negative form of the die in the material. In the context of component positioning, the components, for example in the form of sliding plates, could be used to position a sheet metal holder via their position and orientation. The sheet metal holder can move up and down on the die within the tool via the sliding plates. Reading in a digital pattern of a body part and creating a digital contour, in particular using a CAD system, can be used to create a digital representation of the overall tool with the position points and orientations of the components.This digital model can then be used to optimize the drawing die process, particularly deep drawing, and to precisely propose the positioning and orientation of the individual components within the overall tool. A sheet metal holder guide can function as part of the overall tool, and its mounting can be created quickly and efficiently by determining the positioning and orientation of the components. It can also be improved and optimized, as the best positioning and mounting positions can be incorporated into the machine learning model through appropriate training. The sliding plates are mounted on the stationary drawing die and primarily serve to position and guide the moving sheet metal holder.

[0027] From one perspective, position points and orientation can be represented using an axis system with an origin point. The position points and orientations of the components of a complete tool can be mapped by reading a digital sample of a body part into a trained machine learning model and then creating a digital contour profile on an axis system with an origin point. This axis system enables a coordinate-accurate representation of the position points and orientations, where the position points are described by the coordinates of the axes and the orientations are described by the angle between the axes. This approach allows the positions and orientations to be recorded precisely and accurately to ensure optimal component positioning within the complete tool.

[0028] From one perspective, the multitude of components can include at least one of the following: a sliding plate, a gas spring, or a spacer. These are just a few examples of components to which the method can be applied. The method is not limited to drawing tools but can be applied to any type of tool used for manufacturing body components.

[0029] The multitude of components in a complete tool can exhibit at least one of the following: A sliding plate is, in particular, a flat, smooth plate made of metal or other material. Sliding plates can be used, especially exclusively, for guiding individual assemblies within the overall tooling. Sliding plates, in particular, do not have contact with the body panel. Alternatively or additionally, a sliding plate can serve as a positioning and / or guiding element for the sheet metal holder.

[0030] A gas spring is, in particular, a component that can be used to exert pressure on various components.

[0031] A distance can be a component that can serve to create or regulate a specific distance between the different parts of the overall tool.

[0032] Regarding the position points and orientations of the components within the overall tool, these can be represented by an axis system with an origin point. This allows for the precise determination of each component's position and orientation relative to the other components in the overall tool. All determinations and predictions can be performed digitally, thereby creating a digital twin of the entire tool.

[0033] From one perspective, the machine learning model can be a neural network. A neural network is an example of a machine learning model based on the human brain, which can be used to solve complex problems. It consists of a large number of artificial neurons that can recognize patterns and make predictions. Each neuron receives input from other neurons, specifically from all neurons in a preceding layer of a so-called fully connected neural network, and outputs a signal that is passed on to other neurons. The neurons are interconnected, thus forming a network.

[0034] The neural network can be used in a computer-implemented method for component positioning within a complete tool to predict the position points and orientations of components for forming the overall tool. It can also be used to prepare the input of a digital pattern of a body part into the model and to perform the creation of a digital contour profile.

[0035] A neural network is particularly useful for complex problems that require consideration of many factors and variables. It can also be capable of recognizing patterns in data access and making predictions based on these patterns, such as the position and orientation of components.

[0036] The machine learning model can be modularly structured as AI and in particular features a data pipeline with input, data preprocessing with feature engineering (e.g. creating new features or reducing dimensionality) and scaling, as well as a machine learning algorithm.

[0037] Training can be carried out as follows: The respective ML libraries of the algorithms provide the training methods, e.g., neural networks (Deep Neural Networks, or DNN; Convoluted Neural Networks, or CNN) are trained using the backpropagation method, and parameters such as the learning rate, number of epochs, and batch size can be optimized and can be part of the AI ​​creation process.

[0038] Alternative algorithms that can be used include at least one of the following: deep neural networks (DNNs), one-dimensional convolutional neural networks (1D CNNs), "You Only Look Once" networks (YOLO), or K-nearest neighbor (KNNs).

[0039] A deep neural network (DNN) is a machine learning model based on the principle of neural networks. It features multiple layers of neurons that are interconnected, particularly in adjacent layers, and capable of performing data processing. Each layer contains a multitude of neurons, each assigned a weight and using an activation function to process data and perform feedforward operations.

[0040] A deep neural network (DNN) is trained using a training process where it can be fed data and its weights adjusted to minimize errors between layers. The result is a machine learning model capable of processing new data and making predictions based on its internal representation of that data.

[0041] A key advantage of DNNs is their ability to capture and learn complex relationships between data points. Due to their depth of field, DNNs can learn to classify information from the point cloud, including the associated arrangements and orientations of components, and then apply this classification to a new contour.

[0042] Another advantage of DNNs is their ability to process large amounts of data. Due to their depth of field, DNNs can utilize many neurons within a single layer to capture complex relationships between data points. This allows DNNs to be effective at processing large datasets and outperform other machine learning models.

[0043] A 1D Convolutional Neural Network (CNN) is a special type of neural network that can be used to process (sequential) data such as contour lines. It is based on the principle of convolution and has multiple layers that can be interconnected. In each layer, the input data can be convolved with a filter to refine the features. The results of the convolution operation in each layer of the neural network are then transformed, specifically by an activation function, and passed on to the next layer.

[0044] A 1D CNN is particularly well-suited for processing sequential data such as contour profiles, as it is able to recognize and extract patterns that may be present in the input data sequence. Due to its ability to capture and learn complex relationships between data points, a 1D CNN can achieve very good results in classification and predict the position and orientation of components accordingly.

[0045] Another advantage of a 1D CNN is its ability to process large amounts of data. Due to its depth of field, 1D CNNs can utilize many neurons within a single layer to capture complex relationships between data points. In this way, 1D CNNs can be effective at processing large datasets and outperform other machine learning models.

[0046] Another advantage of a 1D CNN is its ability to adapt to new data. Through the learning process, the model is specifically adapted to minimize errors between layers by adjusting the weights that connect the CNN's neurons via appropriate functions.

[0047] The YOLO (You Only Look Once) network is a machine learning model specifically designed for real-time object recognition. It is based on a convolutional neural network (CNN) and requires only a single forward propagation through the CNN to detect objects. Compared to other approaches like RetinaNet, YOLO is faster.

[0048] The YOLO network differs from other approaches primarily in that it can treat object recognition as a single regression problem. This means the network can establish a direct correlation between the object's spatial coordinates and its class probability. In this way, the model can be trained very quickly and will learn generalizable representations of objects, enabling it to test even unexpected inputs.

[0049] Another advantage of the YOLO network is its ability to detect smaller, closely spaced objects. Specifically, the network uses a large number of classification nodes and bounding boxes distributed throughout the image to detect a wider range of objects. This allows the model to identify even smaller objects that might be missed by other approaches.

[0050] K-Nearest Neighbors (KNN) is a machine learning model that can be based on the local area principle. It is particularly used to make predictions for new data points by determining the value of a new point based on the values ​​of neighboring data points. KNN is a method that can be used in data classification and clustering.

[0051] The model has three main steps: It gathers data points and determines their position in feature space. This is done primarily by using a distance function to calculate the distance between a new point and all previously known points.

[0052] A specific value for K, indicating the number of neighbors within the radius of the new point, is determined. This radius can range from 1 to 5.

[0053] It calculates the mean or majority of the values ​​of its K-neighbors and uses this value as a prediction for the new point. This allows for point clustering and highlights in the machine learning model how the points are distributed and how orientation can be achieved using these points, enabling the arrangement and orientation of components based on a contour profile.

[0054] From one perspective, the digital contour profile can map an outline as a point cloud onto the overall tool to be manufactured. This mapping corresponds specifically to a mathematical representation in a digital space, allowing us to learn something about the real object. In particular, the point cloud of a contour profile achieves a simplification, which is mathematically and resource-wise easier to handle than, for example, a polygon representation.

[0055] A digital machine learning model can be trained (or already trained) by reading digitized patterns of the body part into the model and reading in the corresponding associated complete tools. The model then creates a digital contour path that maps the outline of the drawing die as a point cloud onto the complete tool (or its precursor), for example, to enable the precise positioning of a drawing die and sheet metal holder.

[0056] This digital contour can then be used to predict the position points and orientations of the various components in the overall tool and to optimally design the overall tool.

[0057] Predicting position points and orientations can be achieved through a computer-implemented method that considers a multitude of components within a complete tool. The model can represent the position points and orientations using an axis system with a central origin. A machine learning model, particularly a neural network, can be used to make predictions from the digital patterns of the components and to optimally plan the positioning of the entire tool.

[0058] In other words, what is described here is a method for positioning components using machine learning. A tool for manufacturing sheet metal body parts can contain a multitude of components that together form the overall tool. Each component can be referenced via an axis system with an origin point. These reference "origin points" of the components can be defined in space using machine learning, depending on the component geometries. The result is, in particular, the output and display of 3D-generated reference points where a component can be inserted.

[0059] A computer-implemented method for component positioning in a complete tool with a multitude of components to form the complete tool can be realized through various technologies and approaches.

[0060] One possibility would be to use a machine learning algorithm, as described elsewhere herein, to train the machine learning model, in particular a neural network. Digitized samples of the body part can be used as input for the machine learning model. The machine learning model can then generate a digital representation of the entire tool, including position points and orientations of the various components.

[0061] Another option would be to use CAD software to create digital models of the individual components and then position them within the context of the overall tooling, creating this as a training dataset. The machine learning model can then learn to automatically predict and display the position points and orientations of the components to ensure optimal positioning of the body part, especially sheet metal body parts, within the overall tooling.

[0062] Another possibility would be to use sensors and cameras to determine and display the positions of individual components in relation to the overall tooling and to digitize them accordingly. Sensors can be attached to the various components, or cameras can be pointed at them to measure their positions and orientations and transmit this information to the machine learning model. When a digital sample of a body part is fed into the model, the trained machine learning model can then use this information to suggest an optimal positioning and orientation of components within the overall tooling.

[0063] From an independent perspective, a computer-implemented method for training a machine learning model can be provided. This method specifically includes the step of reading in digital patterns of components with an axis system and origin point. The method specifically includes the step of reading in digital patterns of complete tools with appropriately positioned components. The method specifically includes the step of training the machine learning model to learn the position and orientation of the components in manufactured complete tools.

[0064] From an independent perspective, a computing facility may be trained and configured to perform at least one of the procedures described elsewhere herein.

[0065] The computing device can be described by the features, properties, and advantages of one or both of the methods, the storage device, and the computer program product. This also applies across the category boundaries of method, device, and system. Thus, one or both of the methods, the storage device, and the computer program product can also be described by the features, properties, and advantages of the computing device. For the sake of readability and conciseness, a repetition of all these features, properties, and advantages is omitted.

[0066] From an independent perspective, a computer program product may contain machine-readable instructions to cause a machine, when executed on a computing device, particularly as described elsewhere herein, to perform one or both of the procedures described elsewhere herein.

[0067] The computer program product can be described by the features, properties, and advantages of one or both of the methods, the computing device, and the storage device. This also applies across the category boundaries of method, device, and system. Thus, the computing device, the storage device, and one or both of the methods can also be described by the features, properties, and advantages of the computer program product. For the sake of conciseness and readability, a repetition of all these features, properties, and advantages is omitted here.

[0068] The computer program product can be designed and configured to be executed on a machine. When the computer program product is executed on the machine, one or both of the procedures described elsewhere herein can be performed. The computer program product is, in particular, machine-readable code and / or an electronic signal configured to be read by a machine (such as a computing unit) to transmit instructions to the machine, such as to perform one or both of the procedures described elsewhere herein. A computer program product can, in particular, be configured as machine-readable code, especially as an algorithm capable of performing a corresponding prediction and classification of time series from at least one sensor channel.Alternatively or additionally, the machine-readable code can also execute instructions to perform one of the algorithms described below.

[0069] The computer program product may, in particular, contain instructions that, when executed on a computing device as described elsewhere herein, perform one of the procedures described elsewhere herein. These instructions may be in the form of algorithms, as described below.

[0070] One implementation of a machine learning model can, for example, implement the concept of a neural network. In this case, input data can be read into and processed by the machine learning model through different data processing blocks of the neural network – so-called layers.

[0071] First, the input data can be transformed by an input layer and distributed across multiple processing layers. Each layer, in particular, contains a number of artificial neurons, each of which performs a specific sequence of actions (mathematical mappings) to extract information from the input data. The information can be passed on at each step.

[0072] In the next step, the processed data from one or more connection pools can be prepared and transformed into a simpler format to make the information easier to process for further operations within the neural network. This is achieved in particular by summarizing information from different neurons in a layer into a single value (of which there can be many different values).

[0073] Finally, the data can be further processed by an output layer and transformed into a format suitable for the ultimate goal of the machine learning model. This could be, for example, a prediction or a classification result. In this process, the results from the different layers are combined and processed to generate the final result.

[0074] Data processing in the neural network can also occur continuously in a loop, where a portion of the data is passed to a different layer each time until the final product is generated. In particular, the results from each layer are used to refine the next step and further train the model. During training, the weights of the mathematical functions represented by the artificial neurons can be adjusted to learn the appropriate assignment of input data, such as sensor channel time series, to output data, such as the predicted time series of a specific sensor channel. This allows the trained machine learning model to make a prediction, for example, about the time series of a specific sensor channel, when given a set of input data, such as sensor channel time series.

[0075] Artificial neural networks are, in particular, supervised learning algorithms used for classifying and predicting target variables. The algorithm consists of a network of interconnected artificial neurons that process data. Each neuron can receive input from at least one other neuron in the network and output data to at least one other neuron (sometimes even to itself), which can then be further processed or forwarded by other neurons. The goal of the algorithm is, in particular, to train the network to recognize complex relationships between variables and to make predictions.

[0076] Multilayer Perceptrons (MLPs) represent a special form of neural network architecture. MLPs are specifically supervised learning algorithms for classifying and predicting target variables. These algorithms consist of multiple interconnected layers of neural networks. Each layer receives input from the previous layer and passes output to the next. The goal of the algorithm is to train the network to recognize complex relationships between variables and make predictions. Specifically, each layer receives input from the previous layer and passes output to the next. The MLP is an example of a neural network architecture where each neuron in the next layer represents a function that processes all the weights and inputs from the previous layer.The MLP is trained in particular by switching back the feedforward model to compare the outputs with the target values ​​and to minimize the cost of errors.

[0077] In one embodiment, an implementation can include the creation of the digital contour profile in the form of a point cloud by the CAD system, in particular the CAD computer program product.

[0078] It can involve reading the point coordinates and passing them to the machine learning model via an interface (XML).

[0079] A generated result from the machine learning model can be transferred to the CAD system via an interface (XML).

[0080] It may involve processing the imported information in the CAD program (inserting the component at a point).

[0081] From an independent perspective, a storage device may include a computer program product, as described elsewhere herein. Alternatively or additionally, the computing device, as described elsewhere herein, may include a computer program product, as described elsewhere herein. The computing device is specifically designed and configured to allow the computer program product to read from the storage device in order to perform at least one of the described procedures. Thus, the storage device and / or the computing device can be described by the features, properties, and advantages described and set forth for at least one of the procedures and / or for the computer program product. The reverse is also true. For the sake of conciseness and readability, a repetition of all these features, properties, and advantages is omitted here.

[0082] Further advantages, features, and details of the invention will become apparent from the following description, in which exemplary embodiments of the invention are described in detail with reference to the drawings. The same reference numerals may be used for structures or features that have the same effect or are of the same type. The features mentioned in the claims and in the description may each be essential to the invention individually or in any combination. The following are shown schematically: Fig. 1 a schematic representation of an exemplary embodiment of a method for positioning components in a complete tool; Fig. 2 a schematic representation of an exemplary embodiment of a training procedure; Fig. 3. A representation of a component with a reference axis and origin point; and Fig. 4 A representation of a drawing die and sheet metal holder with body component in a complete tool.

[0083] Fig. Figure 1 shows a schematic representation of an exemplary embodiment of a method 100 for positioning components 12 in a complete tool 10.

[0084] The computer-implemented method 100 for component positioning in a complete tool 10 with a multitude of components 12 for forming the complete tool 10 can include the step of reading 110 a digital pattern 14 of a body part 16 to be produced by the complete tool 10 into a trained machine learning model.

[0085] The process can include the step of creating a digital contour profile 18, in particular using a CAD program, and further, in particular, mapping it onto the overall tool 10. The digital contour profile 18 can be mapped as an outline, as a point cloud 19, onto the overall tool 10 to be manufactured (digitally / mathematically).

[0086] The procedure can include the step of predicting 130 position points and orientations of components 12 to form the overall tool 10.

[0087] The complete tool 10 is designed in particular as a drawing die 20 in order to be able to form a body part, such as a body sheet metal part.

[0088] Position points and orientation can be mapped using an axis system 22 with origin point 24. Fig. Figure 3 shows a representation of a component 12, using the example of a sliding plate 26 with bores 27 and reference axes X, Y, Z as axis system 22 and origin point 24. For example, an axis Z, representing the height axis, can pass through the center of the bores 27 and position the origin point 24 flush with the lower edge of the sliding plate 26. This axis system 22 and origin point 24 can also be arranged within an axis system 22 with an origin point of the overall tool to represent a relative orientation, thereby defining the position and orientation of the component (and correspondingly other components) within the overall tool to be manufactured.

[0089] In general, the components 12 can have at least one of the following: a sliding plate 26, a gas spring 28, or a spacer 30, as is shown in relation to the Fig. 4 is shown as an example. Fig. Figure 4 shows a drawing die 20 with a body component 16 in a complete tool 10. The drawing die 20 can be firmly screwed into the complete tool 10, and the sheet metal holder 900 is positioned, in particular, by the sliding plates 26. The sheet metal holder 900 can move up and down on the drawing die 20 in the tool by means of the sliding plates 26.

[0090] The machine learning model can be a neural network, which can be stored as a computer program product 60 on a storage device 70, to be read by a computing device 50 in order to perform one of the procedures 100, 200 as described elsewhere herein.

[0091] Fig.Figure 2 shows a schematic representation of an exemplary embodiment of a training method 200. The computer-implemented method 200 can be configured to train a machine learning model. In particular, it includes the step of reading 210 digital patterns of components 12 with axis system 22 and origin point 24.

[0092] It can also include the step of reading in 220 digital patterns of complete tools 10 with appropriately positioned components 12.

[0093] It can include the step of training 230 of the machine learning model to learn the position and orientation of the components 12 in manufactured complete tools 10.

[0094] The combinations of features, structures, and configurations described here and elsewhere herein can be extracted from the corresponding combinations in order to combine them in a manner that would be apparent to a person skilled in the art from studying the disclosure made herein. Therefore, other combinations can also be placed under protection.

Claims

[1] Computer-implemented method (100) for generating 3D points in space for component positioning in a composite tool (10) with a plurality of components (12) for forming the composite tool (10), wherein the composite tool (10) is a printing tool, in particular a printing die, or a drawing tool, in particular a drawing die (20), the method (100) comprising the steps of: - reading (110) a digital pattern (14) of a body part (16) to be produced by the composite tool (10) into a trained machine learning model; - creating (120) a digital contour profile (18) by a CAD system, in particular in mapping onto the composite tool (10); - predicting (130) position points and orientations of components (12) for forming the composite tool (10) by the machine learning model, based on the contour profile. [2] Method (100) according to claim 1, characterized by, that position points and orientation are mapped via an axis system (22) with origin point (24). [3] Method (100) according to at least one of the preceding claims, characterized by , that the components (12) have at least one of the following: - a sliding plate (26), - a gas spring (28), or - a spacer (30). [4] Method (100) according to at least one of the preceding claims, characterized by that the machine learning model is a neural network. [5] Method (100) according to at least one of the preceding claims, characterized by , that the digital contour gradient (18) maps an outline as a point cloud (19) onto the overall tool (10) to be produced. [6] Method (100) according to at least one of the preceding claims, characterized by, that the machine learning model is trained by a computer-implemented method (200) for training a machine learning model, comprising the steps of: - reading (210) digital patterns of components (12) with axis system (22) and origin point (24); - reading (220) digital patterns of complete tools (10) with correspondingly positioned components (12); - training (230) the machine learning model to learn the position and orientation of the components (12) in manufactured complete tools (10). [7] Computing device (50) designed and set up to carry out a method (100, 200) according to at least one of the preceding claims. [8] Computer program product (60) comprising machine-readable instructions such that when the computer program product (60) is read by a computing device (50) according to claim 7, the machine-readable instructions cause a machine to perform a method (100, 200) according to at least one of claims 1 to 6. [9] Storage device (70) comprising a computer program product (60) according to claim 8.

Images