Computer-implemented method for estimating the pull-out direction of formed components, method for generating geometry models, formed component and method for its manufacture, and vehicle

Abstract

The invention relates to a computer-implemented method for estimating the draw-out direction (1) of formed components (2). The method according to the invention is characterized by the following process steps: - Providing a geometric model (3) of the formed component (2); - Discretizing the geometric model (3) to derive a component mesh (5) composed of a plurality of elements (4); - Determining the normal vector (6) for each element (4) of the component mesh (5); - Determining an averaged normal vector (7) of at least a subset of the elements (4) of the component mesh (5); and - Equating the orientation of the averaged normal vector (7) with the estimated draw-out direction (1).

Description

The invention relates to a computer-implemented method for estimating the direction of extension of forming components, two methods for generating geometric models, a method for producing forming components, a forming component produced using such a method, and a vehicle having such a forming component.The forming according to DIN 8580 is one of the six main groups of production methods. In the product development process of forming components, information about the pull-out direction, i.e. the direction in which the respective forming component is formed, is required at a plurality of points. The extension direction represents, for example, relevant information for crash simulations or simulations for evaluating the feasibility of corresponding forming components. The forming process results in material changes in the form of plastic expansions and thickness changes, which are only referred to below as material changes. These deviations from the initial state of the material must be taken into account appropriately, in particular in the following simulations, since otherwise the simulation results can deviate from reality.Up to now, information about the pull-out direction is manually predefined in corresponding simulations, which firstly increases the user complexity and secondly can lead to errors.DE 10 2016 001 837 A1 describes a method for computer-assisted prediction of the shape of a sheet metal component to be formed in a forming process. The method provides for carrying out a finite element simulation of the corresponding forming process, wherein at least one local characteristic value is determined from a plurality of result variables by means of a fuzzy logic calculation, said local characteristic value indicating whether or not at least one surface defect is present at a defined location of the component to be formed.Furthermore, US 2008 / 0243442 A1 discloses the determination of process sequences for describing the forming process of a forming component. The document describes the derivation of process parameters for producing a forming component by the analysis of a geometry model representing the forming component with the aid of a CAD system. To produce the forming component, a metal sheet is subjected to a plurality of individual forming or cutting steps. After each step, the geometry of the component changes. The geometry of the geometry model representing the finished forming component is analyzed and geometry parameters are derived. The geometric parameters can be assigned to the geometric changes imposed in the individual forming or cutting steps. From the final component geometry, geometry models are thus derived in reverse step by step, which are established after a respective machining of the metal sheet. It is thus possible to determine in an automated manner how many and which machining steps are necessary for producing the forming component from a correspondingly unprocessed metal sheet.The present invention is based on the object of specifying a computer-implemented method for estimating the pull-out direction of forming components, which allows a rapid and reliable determination of the pull-out direction at least approximately. A further object of the present invention is to specify shaped components and the means required for realizing these shaped components, which are distinguished by a rapid and cost-effective development and a load capacity corresponding to the respective requirements.According to the invention, these objects are achieved by a computer-implemented method for estimating the direction of extension of forming components having the features of claim 1, a method for generating a geometry model having the features of claim 6, a method for generating a geometry model having the features of claim 7, a method for producing a forming component having the features of claim 8 and a forming component produced accordingly. Advantageous embodiments and developments and a vehicle comprising such a forming component are evident from the claims dependent thereon.A computer-implemented method for estimating the direction of extension of forming components has, according to the invention, the following method steps:providing a geometric model of the forming component;defining the geometry model for deriving a component network composed of a plurality of elements;determining the normal vector for each element of the component network;determining an averaged normal vector of at least a subset of the elements of the component network; andequating the orientation of the averaged normal vector to the estimated pull-out direction.The computer-implemented method according to the invention allows automated and thus rapid and reliable estimation of the pull-out direction of forming components. In this case, only one geometry model of the finished forming component is required as the single input variable. The geometry model can preferably be a CAD model of the forming component. This can be a 2D or 3D model.In order to determine the normal vectors, it is necessary to discretize the geometry model. Corresponding component networks are already available in practice, if appropriate. Thus, finite element networks usable for finite element simulations can be used. This can be, for example, the finite element network usable for deformation simulation or a finite element network usable for crash simulation. The corresponding geometry model can contain free-form surfaces. These free-form surfaces can be tapped in order to derive point clouds therefrom. Corresponding elements can likewise be derived from these point clouds. In this way, normal vectors can be generated for the surface elements spanning between the points of the point cloud. Thus, a finite element network of a FEM simulation does not necessarily have to be used.In the simplest case, the averaged normal vector for the normal vectors of all elements of the component network is determined by applying statistical methods. Thus, the orientation of the respective normal vectors can be determined and an average value can be formed over all normal vectors, such as the arithmetic mean, the median or some other mean. The orientation of the averaged normal vector determined in this way corresponds to the actual pull-out direction in a rough approximation. This rough approximation may already be sufficient for use in later simulation steps, which will be discussed in more detail below.The invention is based on the concept that, in particular in the case of flat components, most elements are oriented orthogonally to the pull-out direction. Thus, the most common direction of the normal vectors of the elements of the component corresponds to a good estimate of the actual pull-out direction. This makes it possible to ensure mikability for flat components in most cases. Flat components are distinguished in that flanks produced by deep-drawing are smaller in terms of surface area than the rest of the respective component.An advantageous development of the method according to the invention provides that the elements of the component network are sorted into groups depending on the orientation of their respective normal vector and the averaged normal vector is formed for that group with the highest number of elements. This allows an even more accurate pull-out direction to be estimated. The groups can also be referred to as "clusters". The orientation of the respective normal vectors of the elements of the component will not assume a uniform distribution, but typically one, two or more groups of elements will exist whose respective normal vectors have a certain similarity to one another, i.e. differ only slightly from one another.By considering only that group with the highest element number for determining the averaged normal vector, the negative influence of the remaining elements on the direction determination of the averaged normal vector can be neutralized. Since the normal vectors of the other elements do not correspond to the pull-out direction, the additional consideration of the elements of these groups leads to a deviation of the orientation of the averaged normal vector from the actual pull-out direction. This effect can thus be counteracted accordingly. A wide variety of mathematical methods can be used to divide the elements of the component network into the respective groups.Preferably, a machine learning model on the basis of unsupervised learning examines the normal vectors of the elements of the component network for characteristic patterns, and assigns the elements to the respective groups depending on the patterns found, wherein the machine learning model is based on the k-means algorithm, for example. With the aid of artificial intelligence, in particular in the form of methods of machine learning, particularly preferably on the basis of unsupervised learning, characteristic patterns in the distribution of the orientation of the normal vectors can be found particularly quickly and reliably. Since the normal vectors of the elements contain the orientation of the respective elements as information, the corresponding characteristic patterns will thus be a similarity of the respective orientation of the normal vectors. The machine learning model determines the degree of tolerance to how much the orientation of the individual normal vectors may deviate from one another in order to be assigned to the corresponding groups. The number of groups may be fixedly predefined or may be determined by the machine learning model itself. Unsupervised learning refers to a method of machine learning in which the algorithm learns to exploratively recognize patterns and relationships in data independently and without monitoring. In general, all common unsupervised learning algorithms can be used, wherein in particular k-means is particularly distinguished on account of its efficiency and simplicity.In this case, the hyperparameters on which the machine learning model is based are preferably fine-adjusted, in particular by means of a network search. Fine tuning of the hyperparameters of machine learning models is generally well known, see for example: Four Data Teacher, Online courses and lessons about data science, machine learning and artificial intelligence, hyperparameter tuning. Grid search and random search, by Giantloca Malato, 19 May 2021, https: / / www.yourdatataker.com / 2021 / 05 / 19 / hyperparameter-tuning-grid-search-andrandom-search / , retrieved 24.07.2023 at 16.40 o'clock.In particular, the number of groups is optimized in this case. Whether an optimum number of groups has been found can be evaluated, for example, with the aid of a visualization of the groups formed. For this procedure, reference is made, for example, to: Schweyer, F., Palm, G. (1995). Methods for Cluster Analysis and Visualization of High-Dimensional Data Sets. In: Sagerer, G., Posch, S., Kummert, F. (EDS) Pattern Recognition 1995. Information is up to date. Springer, Berlin, Heidelberg, https: / / doi.org / 10.1007 / 978-3-642-79880-8_64.The term "network search" is also better known under the term "grid search".Fine adjustment of the hyperparameters allows an even more accurate finding of suitable groups, which ultimately allows an even more accurate determination of the estimated pull-out direction.According to a further advantageous embodiment of the method according to the invention, only groups are taken into account whose elements are undercut-free. This also allows an estimated pull-out direction for non-flat components to be determined. A group or a cluster center is undercut-free if a projection of the component grid onto a plane orthogonal to the latter does not have any overlap of the elements of the projected grid. This makes it possible to estimate pull-out directions for an even larger group of forming components.According to the invention, a method for generating a geometry model provides that the component geometry on which the geometry model is based is optimized by means of a simulation for evaluating the feasibility of a formed component, wherein the simulation is parameterized with an estimated pull-out direction determined in the previously described method. To tauten the development process of forming components, skills simulations are carried out with the aid of which it can be determined whether or not the forming component to be produced can be produced with the selected process parameters. For example, the target shape produced could deviate from the desired shape or the component could fail in the forming process, for example cracks, bulges or holes could form. If this is the case, the component geometry is changed until a component geometry satisfying the requirements has been found.By performing simulations instead of hardware tests, the effort, the development time, and costs can be saved. By using the pull-out direction estimated according to a method described above, the actual pull-out direction can be used for the simulation in a good approximation. The risk of obtaining incorrect simulation results due to an incorrectly selected pull-out direction thus falls. In addition, the pull-out direction is determined automatically, as a result of which the manual effort for transferring the manually determined pull-out direction into the corresponding simulation model is reduced.According to the invention, a further method for generating a geometry model provides that the component geometry on which the geometry model is based is optimized by means of an accident simulation, wherein the simulation is likewise parameterized with a pull-out direction estimated according to a method described above. Material changes arise in the forming component itself as a result of the forming process. If the pull-out direction is indicated incorrectly, the material changes in the component will deviate from reality. It is therefore relevant to define the pull-out direction as accurately as possible, i.e. in the greatest possible agreement with reality. Depending on the simulation environment, the pull-out direction can be entered directly in the corresponding accident simulation or indirectly by parameterizing a forming simulation preceding the accident simulation. Thus, taking into account the pull-out direction estimated by means of a method described above, particularly realistic material changes are predicted by the forming simulation in the forming component. The forming component of this type is then used for accident simulation. Due to a better approximated initial plastic expansion and thickness change of the elements, the corresponding component will then behave in the accident simulation in greater agreement with an actually produced component. Thus, more realistic results can be obtained.The respective simulation for evaluating the feasibility of the forming component and / or the respective accident simulation, optionally with a previous forming simulation, makes it possible if the component on which the simulation is based does not meet the requirements of revising the geometry of the forming component. In this way, the component geometry can be optimized until the requirements can be fulfilled. Using the pull-out direction estimated by means of a method described above facilitates the procedure, on account of the reduced manual outlay and the higher accuracy.According to the invention, a method for producing a forming component provides that the process parameters of a forming machine for producing the forming component are adapted based on the component geometry of a geometry model produced according to a method described above. The optimized component geometry found according to a method described above for generating the geometry model can therefore preferably be used to set a corresponding forming machine for producing the respective forming component. Thus, the forming components satisfying the respective requirements can be actually manufactured. The corresponding components can thus be manufactured with a shorter development duration, on the one hand since the estimated pull-out direction is determined in an automated manner and is incorporated into corresponding simulation models, and on the other hand since fewer iterations are required for optimizing the respective component mixtures, since the respective simulations allow more accurate results to be determined taking into account the estimated pull-out direction. The respective forming component can thus be produced more favorably.A forming component according to the invention is produced accordingly by such a production method.According to the invention, a vehicle has at least one forming component produced in this way. In this case, a plurality of such forming components can also be combined to form a subassembly. The vehicle can be any road vehicle such as a car, truck, transporter, bus or the like, or else a rail vehicle, watercraft or aircraft. The corresponding forming components can be so-called flat components, which are produced, for example, by means of deep drawing. These may be, for example, panels, partitions, floor panels or the like. However, these may also be so-called non-flat components. The advantages described in connection with the development and production of the corresponding forming components according to the invention can be correspondingly transferred to the vehicle according to the invention. Thus, the development time for the vehicle and thus the manufacturing cost can be reduced.Further advantageous embodiments of the computer-implemented method for estimating the pull-out direction of forming components, the method for generating geometric models, the method for producing forming components, a corresponding forming component and a vehicle, comprising at least one such forming component, also result from the exemplary embodiments, which are described in more detail below with reference to the figures.The following are shown: FIG. 1 shows a schematic illustration of the sequence of a method according to the invention for estimating the pull-out direction of forming components; FIG. 2 is a schematic view of a component network of a flat forming component with averaged normal vectors drawn on it; FIG. 3 is a schematic view of a component network of an undercut-free forming component with averaged normal vectors drawn on it; and FIG. 4 shows a flow chart for developing and manufacturing a vehicle according to the invention.The pull-out direction 1 of forming components 2 represents an important parameter for simulations in order to evaluate the feasibility of the respective forming components 2 and for accident simulations. The manual effort for ascertaining and inputting the actual pull-out direction as a parameter into the respective simulation is associated with a high effort and can lead to errors. With the aid of a computer-implemented method according to the invention for estimating the pull-out direction 1 of the forming components 2, this procedure can be automated and thus the required time duration can be reduced and the risk of errors occurring can be reduced.FIG. 1 shows the process sequence for this purpose. In a method step 101, a geometry model 3 of the forming component 2 to be produced is provided. For example, this can be a CAD file.In a method step 102, the geometry model 3 is discretized in order to derive a component network 5 composed of a plurality of elements 4 therefrom. This can be, for example, a finite element network.In a method step 103, the respective normal vectors 6 are determined for each element 4 of the component network 5. FIG. 1 shows, by way of example, a section from the component network 5 having four elements 4. not all figure elements are provided with reference symbols in order to maintain the clarity.In a subsequent method step 104, an averaged normal vector 7 is determined for at least a subset of the elements 4 of the component network 5. Proven methods of statistics can be used for this purpose. Particularly preferably, the normal vectors 6 of the elements 4 of the component network 5 are aggregated according to their orientation into groups, preferably using an unsupervised learning model. The corresponding machine learning model is preferably based on the k-means algorithm.The averaged normal vector 7 is then preferably formed for that group of elements 4 with the highest number of elements and used for the next method step.In method step 105, the orientation of the averaged normal vector 7 is then set equal to the estimated pull-out direction 1. The estimated pull-out direction 1 corresponds approximately to the actual pull-out direction, but can deviate from it within tolerable limits.FIG. 2 shows the representation of a geometry model 3 representing the forming component 2 with the component network 5 drawn on it and the corresponding elements 4. The position at which the respective averaged normal vectors 7 are represented in FIGS. 2 and 3 can be chosen as desired, since only the orientation of the respective vectors is relevant, but not their absolute positioning with respect to the forming component 2. the elements 4 assigned to the group having the largest number of elements are highlighted in FIG. 2 approximately by hatching. As FIG. 2 shows, all these elements 4 are oriented approximately horizontally. The averaged normal vector 7 of this group is represented by a solid arrow and the two remaining averaged normal vectors 7 by dashed arrows. The corresponding further elements 4 of the component network 5 are assigned to one of the other two averaged normal vectors 7. FIG. 2 shows a so-called flat forming component 2.FIG. 3 shows a corresponding illustration for a forming component 2 without any undercut.FIG. 4 shows the sequence for developing and manufacturing a vehicle according to the invention and illustrates the advantageous application of the estimated pull-out direction 1 determined by means of the described computer-implemented method according to the invention.In a method step 401, the pull-out direction 1 is derived from the respective geometry model 3, as described above. The estimated pull-out direction 1 is then provided in a method step 402. Method steps 403 to 405 describe the execution of a simulation for evaluating the feasibility of the forming component 2, and method steps 406 to 408 describe the execution of an accident simulation using the corresponding component geometry of the forming component 2.FIG. 4 shows by way of example that the respective simulations can be carried out in parallel with one another. However, it would also be possible to carry out the simulations successively.In method step 403, the simulation is carried out for evaluating the feasibility of the forming component 2, wherein the estimated pull-out direction 1 was used in the respective simulation model. In method step 404, the result is obtained. In method step 405 it is checked whether the component geometry used meets the requirements, i.e. whether the respective forming component 2 can be produced under the respective process conditions. If this is not the case, the method begins again in method step 401 with a modified component geometry. If, on the other hand, the requirements are fulfilled, the method proceeds to method step 409.In method step 406, an accident simulation is carried out based on the respective forming component 2 and the estimated pull-out direction 1. The accident simulation can be preceded by a forming simulation taking into account the pull-out direction 1 in order to estimate the prestresses which are established in the component on the basis of the forming process (not illustrated). In method step 407, the result of the accident simulation is obtained. In method step 408, an evaluation is then carried out as to whether the forming component meets the respective requirements, that is to say whether there is sufficient crash safety. If this is not the case, method step 401 is carried out again, based on a changed component geometry of forming component 2; if, on the other hand, the requirements are fulfilled, the method proceeds to method step 409.In method step 409, the optimized component geometry of the forming component 2 is found. The respective forming component 2 can now be produced.For this purpose, in the next step 410, a corresponding forming machine is set in order to produce the corresponding component geometry.In method step 411, the corresponding forming component 2 is produced.In method step 412, the forming component 2 or a corresponding assembly of a plurality of such forming components 2 is installed in the vehicle or serves as a basic framework for producing the vehicle.The respective vehicle is manufactured as usual and finished in method step 413.References included in the specificationThis list of documents cited by the applicant has been produced in an automated manner and is only included for the better information of the reader. The list is not part of the German patent application or utility model application. The DPMA does not take any adhesion for any faults or omissions.Patent Literature citedDE 10 2016 001 837 A1

[0004] US 2008 / 0243442 A1

[0005] Cited Non-Patent LiteratureDIN 8580

[0002]

Claims

Computer-implemented method for estimating the pull-out direction (1) of forming components (2), characterized bythe following method steps: - providing a geometry model (3) of the forming component (2); - discretizing the geometry model (3) for deriving a component network (5) composed of a plurality of elements (4); - determining the normal vector (6) for each element (4) of the component network (5); - determining an averaged normal vector (7) of at least a subset of the elements (4) of the component network (5); and - equating the orientation of the averaged normal vector (7) with the estimated pull-out direction (1).Method according to Claim 1, characterized in that the elements (4) of the component network (5) are sorted into groups depending on the orientation of their respective normal vector (6), and the averaged normal vector (7) is formed for that group having the highest number of elements.Method according to Claim 2, characterized in that a machine learning model on the basis of unsupervised learning examines the normal vectors (6) of the elements (4) of the component network (5) for characteristic patterns, and assigns the elements (4) to the respective groups as a function of the patterns found, wherein the machine learning model is based in particular on the k-means algorithm.Method according to claim 3, characterised in that the hyperparameters on which the machine learning model is based are fine-adjusted, in particular by means of a network search.Method according to one of Claims 2 to 4, characterized in that only those groups are taken into account whose elements (4) are undercut-free.Method for generating a geometry model (3), characterized in that the component geometry on which the geometry model (3) is based is optimized by means of a simulation for evaluating the feasibility of a forming component (2), wherein the simulation is parameterized with an estimated pull-out direction (1) determined by a method according to one of Claims 1 to 5.Method for generating a geometry model (3), characterized in that the component geometry on which the geometry model (3) is based is optimized by means of an accident simulation, wherein the simulation is parameterized with an estimated pull-out direction (1) determined by a method according to one of Claims 1 to 5.Method for producing a forming component (2), characterized in that the process parameters of a forming machine for producing the forming component (2) are adapted on the basis of the component geometry of a geometry model (3) produced by a method according to Claim 6 and / or 7.Forming component (2), characterised bya production with a method according to claim 8.Vehicle, characterised byat least one forming component (2) according to claim 9.

Images