Ai model for controlling and / or regulating a ring rolling process

Abstract

The invention relates to a method for generating and applying an AI model (60) for a control device (50) for controlling and / or regulating and / or monitoring a ring rolling process relating to a workpiece (2), the method comprising: creating a database by executing at least one computer simulation, preferably an FEM simulation, of at least one fictitious, partial or complete ring rolling process on the basis of at least one set of fictitious process parameters; generating the AI model (60) by training a neural network using the created database; and applying the AI model (60) by inputting process parameters of the ring rolling process of the workpiece (2) into the neural network, the neural network of the AI model (60) generating an output field that is characteristic of the workpiece (2), said output field preferably comprising a temperature and / or shape-change distribution of the workpiece (2).

Description

[0001] Kl model for controlling and / or regulating a ring rolling process

[0002] Technical field

[0003] The invention relates to a method for creating and applying a computer model for a control device for controlling and / or regulating and / or monitoring a ring rolling process. The invention further relates to a device for rolling a ring-shaped workpiece with a control device for controlling and / or regulating and / or monitoring the ring rolling process.

[0004] Background of the invention

[0005] Among the devices for rolling rings, ring rolling mills are known which have a main roll and a domed roll which forms a rolling gap with the main roll for radially rolling the ring.

[0006] A ring, positioned on a mandrel rolling table such that the mandrel roll passes through the ring, is rolled while the main and dome rolls rotate. The rolling process itself occurs through the resulting roll gap between the main and dome rolls. Axial rolls can also be installed, which act on the ring in the axial direction.

[0007] Currently, estimating the geometry development of a workpiece based on analytical equations to predict the process progression is only possible for rectangular or very simple cross-sections. For complex, such as profiled ring cross-sections, only a rough estimate of the geometry development and the temperature and strain distribution based on experience can be made within a reasonable timeframe. A 3D prediction of the development of the ring geometry and the temperature and strain distribution during ring rolling is possible based on finite element modeling (FEM). In principle, this also allows for the prediction of the grain size distribution, pore closure behavior, and damage development in the material.In this regard, reference should be made to the following literature references: Vahid Jenkouk (2014), “Integration of control algorithms into the finite element simulation of ring rolling processes”, Dissertation IBF Aachen; Gideon Schwich (2019), “Investigation of influencing factors on microstructure and material damage in ring rolling”, Dissertation IBF Aachen.

[0008] While a rough estimate of the geometry and temperature development using analytical equations can be performed very quickly for rectangular cross-sections, within a few seconds, the calculation only provides an example cross-section and does not consider differences across the circumference of rings, nor can it predict the geometry development of profiled rings. Suitable technology software can simplify process design and the determination of an ideal preform, but iterative solutions based on rolling operations on the relevant equipment are often necessary to find an ideal preform. These are correspondingly time-consuming and expensive.

[0009] Prediction using FEM is very time-consuming. A simulation can take hours or even days, making it unsuitable for rapid production planning or for optimizing the ring rolling process with regard to the resulting component quality.

[0010] In practice, predicting component quality and the necessary process strategy for ring rolling, particularly for the production of profiled ring cross-sections, has so far been primarily experience-based. This experiential knowledge is not necessarily structured and readily accessible within the company, but rather resides with the respective employee. This can lead to personnel-dependent fluctuations and a corresponding risk of downtime.

[0011] A lack of knowledge about the anticipated geometry development often complicates process design, particularly with regard to achieving the most ideal preform possible, ensuring that the profiled cross-section is completely filled (and not already emptied) by the end of the process. Without knowledge of the geometry development, it is generally necessary to conduct trials on the existing ring rolling machine, in addition to using analytical methods and software, to iteratively determine a suitable preform.

[0012] Currently, model- and computer-aided prediction of the deformation distribution as a basis for predicting component quality in ring rolling is only possible to a limited extent or with considerable time expenditure. Therefore, it is not yet possible to design, control, or optimize the ring rolling process as a complex forming process with regard to material properties that significantly determine component quality and performance.

[0013] Description of the invention

[0014] One object of the present invention is to improve the control and / or regulation and / or monitoring of a ring rolling process of a workpiece.

[0015] The problem is solved by a method with the features of claim 1 and a device with the features of dependent device claim 9. Advantageous embodiments follow from the dependent claims, the following description of the invention, and the description of preferred embodiments.

[0016] The invention relates to the rolling of a ring-shaped workpiece. With regard to the geometry of the workpiece, examples include disc- or sleeve-shaped rings with rectangular cross-sections, symmetrically or asymmetrically profiled rings, rings profiled on the outside and / or inside, rings profiled on the top and / or bottom, two-stage radially profiled rings, two-stage axially profiled rings, flanges, gears, solid discs, and externally and / or internally toothed rings. Regarding the material of the workpiece, iron-based materials and steels, as well as materials based on aluminum, titanium, nickel, copper, zinc, and alternative formable materials, can be considered.

[0017] The method according to the invention specifically serves to create and apply a computer model for a control device for controlling and / or regulating and / or monitoring a ring rolling process, in particular for the production and optimization of a preform for profiled ring cross-sections.

[0018] According to the procedure, a database is first generated by executing at least one computer simulation of at least one fictitious, partial, or complete ring rolling process based on at least one set of fictitious process parameters. "Fictitious" here means that the process parameters used for the simulation(s) do not necessarily have to be identical to the process parameters of the (real) ring rolling process to be controlled, regulated, or monitored. The computer simulation is preferably based on a finite element method (FEM) simulation. Analytical prediction methods, recordings of experiments, and known simulation methods are also possible. Based on the database thus generated, a computer intelligence (CL) model is created by training a neural network. Alternatively or additionally to the FEM approach, the result can also be improved through supervised learning by incorporating existing knowledge into the CL's target variables.

[0019] The neural network can be classified as a convolutional recurrent neural network, comprising at least one convolution layer and one pooling layer for downsampling, at least one recurrent layer to describe the temporal evolution, and optionally at least one transcription layer and one output layer. The exact number of layers and the number of neurons in each layer can be varied and determined depending on the application.

[0020] The resulting AI model is applied within the (real) ring rolling process by inputting process parameters of the ring rolling process into the neural network, whereby the neural network of the AI ​​model generates an output field that is characteristic of the workpiece. Preferably, the output field includes a temperature and / or deformation distribution of the workpiece.

[0021] The AI ​​model presented herein, using at least one neural network, allows for a very accurate prediction of characteristic material properties, such as geometry development as well as temperature and deformation distribution, for ring rolling processes within a short time.

[0022] The rapid computation time of, for example, a few seconds (<10), enables the optimization of component quality and performance with regard to selected target parameters, such as homogeneous and small grain size, maximization of pore closure, or minimization of introduced damage. Preferably, at least one material property of the workpiece is determined from the output field of the neural network of the AI ​​model, wherein the at least one material property preferably comprises a local grain size distribution and / or pore closure behavior and / or damage development in the workpiece. The prediction system based on the neural network allows for combination with the prediction of the grain size distribution in the material, the pore closure behavior of casting cavities and similar defects in the material, and damage development, as well as the formation of possible component defects during the ring rolling process.The aforementioned properties can be predicted in the component, and the manufacturing process can be specifically designed with these criteria in mind. Any process calculation can thus take defined criteria regarding component quality and performance into account, thereby optimizing the process accordingly.

[0023] Preferably, the output field of the neural network of the AI ​​model and / or the determined at least one material property of the workpiece are used to control, regulate, and / or monitor the ring rolling process. Such feedback to the plant control system enables the process to be carried out as closely as possible to the optimized process design. Process monitoring can provide the plant operator with further information about the geometry development and allows warning messages to be issued if the process deviates from the prediction or ideal process control, thus posing a risk that the component will not meet the required quality and performance standards.

[0024] Preferably, the output field of the neural network of the AI ​​model and / or the determined at least one material property of the workpiece is used to determine a preform of the workpiece. Particularly for profiled cross-sections of the component to be manufactured, this allows for a significantly more precise determination of an ideal preform than has previously been possible. Preferably, the database for the neural network is generated by finite element modeling, whereby various fictitious process parameters are varied and a multidimensional grid of these fictitious process parameters is created, such that each point in the grid corresponds to a combination of the fictitious process parameters.

[0025] Preferably, the fictitious process parameters include workpiece dimensions and / or tool dimensions and / or temperatures and / or pause times and / or ring rolling speeds and / or rolling curves.

[0026] Various parameter ranges are possible for scanning using FEM, relating to the geometry type, the material, the dimensions, process parameters, etc., which are given as examples in the description of embodiments below.

[0027] Preferably, FEM simulations are performed based on the defined grid, the results of which preferably include nodal displacements and / or temperatures and / or deformations, such as strains and shears, and / or deformation increments and / or deformation rates. This can be supplemented by results from analytical calculation models, records, and the like.

[0028] The data basis for training the neural network therefore consists in particular of the multidimensional grid of process parameters including the associated FEM input parameters, the respective definition of the FEM model as well as the element- and node-based results of the simulation(s).

[0029] Preferably, the neural network of the AI ​​model, trained on the dataset, generates the output field for combinations of process parameters not considered during the creation of the dataset. In other words, by training on the dataset, the neural network is able to generate an output field in the form of a temperature and strain distribution for previously unconsidered combinations of process parameters or grid points within a short time, particularly within a few (<10) seconds. The prediction accuracy is very high within the grid boundaries; outside the grid, the predictions are less reliable.

[0030] The above-mentioned problem is further solved by a device for rolling a ring-shaped workpiece with a rolling unit for ring-shaped rolling of the workpiece and a control unit for controlling and / or regulating and / or monitoring a ring rolling process that can be carried out by the rolling unit, wherein the control unit comprises a computer model with a neural network that is configured to generate an output field on the basis of process parameters of the ring rolling process that is characteristic of the workpiece, preferably comprising a temperature and / or shape change distribution of the workpiece.

[0031] The features, technical effects, advantages and embodiments described in relation to the method apply analogously to the device.

[0032] For the reasons stated above, the control device is preferably configured to determine at least one material property of the workpiece from the output field of the neural network of the AI ​​model, wherein the at least one material property preferably comprises a local grain size distribution and / or a pore closure behavior and / or damage development in the workpiece. Preferably, for the reasons stated above, the control device is configured to use the output field of the neural network of the AI ​​model and / or the determined at least one material property of the workpiece for controlling and / or regulating and / or monitoring the ring rolling process.

[0033] Further advantages and features of the present invention will become apparent from the following description of preferred embodiments. The features described therein can be implemented individually or in combination with one or more of the features set out above, provided that the features do not contradict each other. The following description of preferred embodiments is given with reference to the accompanying drawings.

[0034] Brief description of the characters

[0035] Preferred further embodiments of the invention are explained in more detail by the following description of the figures. These show:

[0036] Figure 1 shows a device for rolling a ring-shaped workpiece; and

[0037] Figure 2 schematically illustrates a process for creating and applying a computer model, based on a neural network, for a control device of the apparatus.

[0038] Detailed description of preferred embodiments

[0039] Preferred embodiments are described below with reference to the figures.

[0040] Figure 1 shows a device 1 with a rolling unit 10 for rolling an annular workpiece 2. The rolling unit 10 is implemented in a table-like design, comprising one or more rolling tables 11 on which the workpiece 2 is placed for processing. The rolling unit 10 has a main roll 12, which is rotatable about a main roll axis, and at least one rotatable mandrel roll 14 oriented parallel to it. The main roll 12 and mandrel roll 14 form a rolling gap in which the workpiece 2 is rolled in a radial direction.

[0041] In addition to the main roll 12 and the mandrel roll 14, axial rolls 16 can be installed for axial deformation of the workpiece 2. The rolling unit 10 also includes centering rolls 18, which are designed to hold the workpiece 2 in position during the rolling process.

[0042] The device 1 further comprises a control unit 50, which communicates with the various devices, assemblies, and the like. The control unit 50 is connected via signal transmission to the components to be controlled, regulated, and / or read out, and thus in particular to the rolling device 10 and the manipulation device 20.

[0043] Communication between the control unit 50 and the components to be controlled, regulated, and / or read can be wired or wireless, digital or analog. The control unit 50 can receive and / or send signals (control signals, data, etc.), whereby both unidirectional and bidirectional signal transmission fall under the term "communication" in this context. The control unit 50 does not necessarily have to be implemented as a central computer or electronic control system; it can also include decentralized and / or multi-stage systems, control networks, cloud systems, and the like. The control unit can also be an integral part of a higher-level plant control system or communicate with one. The control unit 50 can also communicate with lower-level plant control systems, i.e., control systems assigned to the respective equipment.The control unit 50 includes or has access to a KL model 60 (where "KL" stands for artificial intelligence) based on one or more neural networks. The KL model 60 allows the prediction of the geometry development of the workpiece 2 as well as the temperature and deformation distribution in the workpiece 2 during ring rolling for defined geometry areas and process strategies within a short time, in particular within a few seconds.

[0044] Finite element method (FEM) simulations can be used as the data basis for the underlying neural network of the Kl model 60. In these simulations, process parameters (workpiece and tool dimensions, temperatures, pause times, pass rates, etc.) are varied. For rectangular or simple cross-sections, the FEM simulations can be supported by supervised learning, for example, using standard software. In supervised learning, the results from standard software and / or empirical data / known knowledge are quantified, for example, in the form of an objective function. The Kl's task is to strive for a specific target value and / or a specific minimum and / or maximum of this objective function. A suitable number of support points (combinations with different process parameters) are used to generate a database and train the neural network.At least within this selected range, the control unit 50, with access to the KL model 60, is then able to predict the temperature and deformation distribution of the workpiece 2 (including grain size, damage development, and pore closure behavior). The range is selected according to the possible and targeted spectrum of process parameters.

[0045] Based on the prediction of the temperature and shape change distribution of workpiece 2, additional material properties that are crucial for component quality and performance, such as grain size distribution, pore closure behavior and damage development, can be predicted.

[0046] This enables process design within a few seconds from a starting geometry, for example an ideal preform, to a final geometry using the Kl model 60 based on at least one neural network.

[0047] In addition, by predicting the temperature and deformation distribution based on the Kl-Model 60, the grain size distribution, pore closure behavior, and damage development can be predicted using conventional models. The process design can thus be optimized to achieve or maintain defined criteria for component quality and performance.

[0048] The process design as described above is used as a process specification for ring rolling and can, for example, be created as a specification for the control unit 50 or its plant control software and read into it or transmitted directly to it, so that the component to be manufactured follows the process design as precisely as possible. The process specification (feed rates, axis positioning, additional control options, such as position control Z-axis control of the centner rolls 18 (force on the rolls or diameter of the workpiece 2, etc.)) can also be entered into the system and thus used to control the device 1 during the process.

[0049] Furthermore, the temperature and deformation distribution in workpiece 2 can be calculated almost in real time during the process (for example, with a delay of less than 2 seconds). Based on this, the current grain size distribution, pore closure, and damage development can also be determined and, if necessary, displayed. Such process monitoring can also be used to generate a warning, for example, for the operator, if the process is likely to result in workpiece 2 not reaching the defined component quality range.

[0050] Figure 2 schematically shows an exemplary process for creating and applying the AI ​​model 60 for the control unit 50. The data basis required for training the neural network of the AI ​​model 60 is generated here by FEM simulations.

[0051] In a first step S1, a multidimensional grid is defined for the FEM simulation(s), whereby different parameter ranges for the rasterization, i.e., the grid points to be simulated of the multidimensional grid, are possible, possibly supplemented by the inclusion of the objective function of supervised learning / analytical functions, etc.

[0052] With regard to geometry type, for example, disc- or sleeve-shaped rings with rectangular cross-sections, symmetrically or asymmetrically profiled rings, rings profiled on the outside and / or inside, rings profiled on the top and / or bottom, two-stage radially profiled rings, two-stage axially profiled rings, flanges, gears, solid discs, and externally and / or internally toothed rings can be included.

[0053] With regard to the material of workpiece 2, iron-based materials and steels, as well as materials based on aluminum, titanium, nickel, copper, zinc, and alternative formable materials, can be considered, with workpiece temperatures ranging from -273 °C to 2,800 °C, preferably between 250 °C and 1,300 °C. Alternatively, the forming temperature of the workpiece 2 to be machined can be taken into account. Regarding the mass and dimensions of the component, masses from 0.1 kg to 5,000 t operating weight are included, with outside diameters from 10 mm to 60,000 mm and heights from 10 mm to 25,000 mm.

[0054] The forming unit or rolling device 10 can be designed as a radial ring rolling machine, radial-axial ring rolling machine, table ring rolling mill, axial die ring rolling machine and can each be hydraulically, servo-hydraulically, servo-mechanically or mechanically driven.

[0055] Process parameters include relative stitch reductions of 0.1 to 50% per ring revolution and target ring growth speeds of 0.1 mm / s to 100 mm / s.

[0056] Furthermore, various boundary conditions can be used for the rasterization to create an FEM simulation model, including ambient temperatures from -100 to +900 °C and tool temperatures from -100 to +1400 °C.

[0057] The data basis required for training the neural network of the Kl model 60 is then generated in a second step S2 by FEM simulation(s).

[0058] In this regard, at least one FEM simulation model is established that corresponds to the aforementioned parameters of the grid points of the multidimensional grid to be simulated. A simulation can encompass a single pass or multiple passes through a rolling gap (radial or axial), one or more ring rotations, up to and including an entire ring rolling process including any intermediate reheating.

[0059] The workpiece 2 under consideration is meshed in 2D with preferably at least 12 or in 3D with preferably at least 64 nodes. A maximum of preferably 10,000,000 or 1,000,000 nodes, respectively, comprising the FEM mesh, are used in each simulation.

[0060] The FEM simulation is started, and after the calculation is complete, the simulation results are saved. The saved data includes, in particular, nodal displacements, temperatures, deformations (strains and shear), deformation increments, and deformation rates. If a machine is simulated with automatic consideration of machine limits (power, torque, forces, etc.), the movements of the tools are also part of the results. The simulation results generated in this way are then exported, preferably automatically, as a database.

[0061] The sum of all input data (possibly also from supervised learning) paired with the respective stored simulation results forms the data basis for the neural network.

[0062] It is helpful if the individual FEM simulations are defined in such a way that the number of elements, and thus the number of support points in the model, is as similar as possible for all simulations within a given database. However, this is not strictly necessary.

[0063] The data basis for training the neural network thus consists of the multidimensional grid of process parameters including the associated FEM input parameters, the respective definition of the FEM model, and the element- and node-based results of the simulations.

[0064] In a third step, S3, the neural network of the Kl model 60 is trained. This is done in a suitable programming environment using the generated dataset. The neural network can be classified, for example, as a Convolutional Recurrent Neural Network, consisting of at least one Convolutional Layer and one Pooling Layer for downsampling, at least one Recurrent Layer to describe the temporal evolution, and optionally at least one Transcription Layer and one Output Layer. The exact number of layers and the number of neurons in each layer are determined by the size of the dataset and, in particular, the definition of the finite element model, i.e., the number of elements and nodes.

[0065] In step S4, when applying the trained neural network, an input point within the multidimensional grid can now be used, which does not necessarily have to correspond to a grid point. It is important to note that the data set should be chosen so that no points outside the grid are required for future predictions. While results will still be generated for such points, the deviation from reality increases with the distance from the grid.

[0066] For such a selected point, the neural network then generates a prediction for the selected simulation results, i.e., at least node displacements, temperatures, shape changes (strains and shearing), shape change increments and shape change rates, and possibly the tool movements.

[0067] For a machine or rolling mill 10 with automatic consideration of machine limits, the reaching of the machine limits can be predicted and the process strategy can be assessed.

[0068] By training on the database, the neural network is thus able to generate an output field in the form of a temperature and shape change distribution for previously unconsidered combinations of process parameters (or grid points) within a short time, for example, within a few (<10) seconds. The accuracy of the prediction is very high within the grid boundaries; outside the grid, the predictions are less reliable.

[0069] In a fifth step S5, the results from production can be fed back into the KL, thus into step S3.

[0070] The output field can be used to predict material properties, crucial for component quality and performance, using suitable models based on temperature and strain distribution. This includes correct geometric dimensions, mechanical properties, particularly yield strength, tensile strength, fracture toughness, elongation at break, and workpiece lifetime (such as wear / abrasion), etc. It is understood that the product properties, as mentioned above, result from physical / chemical states / properties, such as grain size. These relate, among other things, to the local grain size distribution, pore closure behavior, and / or damage development within the component.

[0071] The generated results can be coupled with conventional models for predicting grain sizes, damage values ​​and / or pore closure, for example, in order to predict and, if necessary, optimize the performance of components.

[0072] The optimized process design is used by the control unit 50 to control, regulate, and / or monitor the ring rolling process. If deviations from the optimized process occur, a recalculation and, if necessary and possible, a course correction of the subsequent process can be performed. If the process deviates so significantly from the specification that the required component quality can no longer be achieved, or if reheating becomes necessary, the control unit 50 can issue a corresponding warning message, to which an operator can then react. Alternatively or additionally, automated plant control or process monitoring can be used to ensure autonomous operation.

[0073] The following is a concrete example to illustrate the comprehensibility of the above process for creating and applying the Kl-Model 60:

[0074] Assume rectangular rings with unit weights of 100 kg to 25 t made of structural steels, a starting geometry in the range of 600 mm / 410 mm (outer diameter / inner diameter) x 90 mm and 4,000 mm / 3,550 mm x 2,100 mm and finished dimensions of 750 mm / 600 mm x 80 mm to 8,000 mm x 7,740 mm x 2,000 mm, which are manufactured at a starting temperature of 1250 °C.

[0075] If a workpiece 2 cools below 900 °C during production, reheating may be necessary. For ring rolling, an automatic radial-axial ring rolling mill with a maximum radial / axial force of 10,000 kN / 5,000 kN is used as the rolling unit 10. The relative pass reductions are in the range of 2 to 15%, the targeted ring growth speeds are in the range of 10 to 25 mm / s, the diameters of the mandrel roll(s) 14 are in the range of 250 mm to 500 mm, and the main roll diameter is fixed at 1200 mm. The ambient temperature is approximately 35 °C, and the tool temperature is approximately 300 °C.

[0076] To generate the database according to step S2, these ranges of dimensions and process parameters, as well as the fixed boundary conditions, are simulated. For this purpose, the boundary points and 18 further linearly interpolated points within the dimensional range are simulated for each dimension. With 20 simulations each for "start geometry" and "final geometry," plus 6 for "mandrel roller diameter," 7 for "referenced pass rate," and 7 for "ring wax rate," this results in 20 x 20 x 6 x 7 x 7 = 11,760 simulations. By applying physically meaningful limitations to the start and final geometries (the smallest final geometry cannot be produced from the largest start geometry) and the mandrel roller diameter (which cannot be larger than the inner diameter of the start geometry), this number can be reduced to 8,500 simulations.

[0077] A complete process simulation, for example, takes 8 days, so the initial effort is very high. However, since many stitches and ring turns are very similar, the processes can be broken down into their individual ring turns, with each process consisting of 10 to 100 ring turns. By grouping thermally and geometrically similar ring turns, one obtains 200,000 simulations, each requiring approximately 60 minutes of simulation time. With a total simulation time of 200,000 hours (instead of 1,632,000 hours for the original 8), the process becomes significantly more efficient.(500 simulations), whereby in principle any number of these simulations can be carried out and evaluated in parallel, all simulations can be automatically started, calculated and evaluated within 9 days using a high-performance cluster with, for example, 1000 computing cores, including storing the simulation results together with the input parameters in a database in such a way that they can be used for the subsequent training of the neural network.

[0078] The generated database is then used, for example, to train a neural network in the C# programming language. This network generates an output within a few seconds, corresponding to the result of a finite element method (FEM) simulation. The results generated in this way, like the results of the FEM simulations, can be used to evaluate the performance of the manufactured components. Existing models for predicting grain size, damage, and / or pore closure behavior are used for this purpose. The short computation time of just a few seconds for generating a result can be leveraged to optimize the rolling strategy and the preform used with regard to minimizing damage development.

[0079] The rolling strategies optimized in this way for the entire product range are used in further production to control and / or regulate the rolling unit 10. This prevents machine overload and minimizes potential production downtime. The device 1 does not need to have fully automatic machine limit adjustment capabilities.

[0080] In addition to optimizing the rolling strategies, the condition of workpiece 2 during the rolling process can be calculated for each ring rotation. If the process deviates from the defined rolling strategy and the calculation shows that the required component quality can no longer be achieved at this temperature, the machine operator can be notified that reheating is necessary. A further adjusted rolling strategy is then calculated for the reheated component, enabling the required component quality to be achieved, and this strategy is implemented by the machine if necessary.

[0081] In a further development of the concept presented herein, software for the automatic generation of the database (creation, calculation and evaluation of the simulations) can be developed in a suitable simulation program.

[0082] Furthermore, a system for the simplified generation / prediction and, if necessary, automatic optimization of shape change distributions by the neural network (possibly with regard to a goal, for example, with regard to a defined grain size range, maximization of pore closure, or minimization of damage) can be developed.

[0083] The Kl model 60 presented herein, using at least one neural network, allows for a very accurate prediction of the geometry development as well as temperature and deformation distribution for ring rolling processes within a few seconds and, if necessary, in 3D. These can also form the basis for predicting component quality and performance.

[0084] For profiled cross-sections (including not only, but also rectangular cross-sections) of the ring-shaped component to be manufactured, an ideal preform can thus be determined much more accurately than has been the case so far.

[0085] Based on this, predictions can be made regarding the grain size distribution, pore closure behavior, and / or damage development in the component, and the process can be specifically designed with respect to these criteria. The sampling plan calculation can thus take defined criteria regarding component quality and performance into account, and the process can be optimized accordingly.

[0086] Feeding the data back into the plant control system allows the process to be carried out as close as possible to the optimized process design.

[0087] Process monitoring can provide the machine operator with further information about the geometry development and enables the issuance of warning messages if the process deviates from the prediction or ideal process control, thus posing a risk that the component will not meet the required quality and performance standards. If necessary, the autonomous system can decide on corrective measures itself. Where applicable, all individual features shown in the exemplary embodiments can be combined and / or exchanged without departing from the scope of the invention.

[0088] Reference symbol list

[0089] 1 Device for rolling a ring-shaped workpiece

[0090] 2 workpieces 10 rolling device

[0091] 11 Rolling table

[0092] 12 Main roller

[0093] 14 Dome roller

[0094] 16 Axial roller 18 Centner roller

[0095] 50 Control unit

[0096] 60 Kl model

Claims

Patent claims 1. Method for creating and applying a KL model (60) for a control device (50) for controlling and / or regulating and / or monitoring a ring rolling process of a workpiece (2), wherein the method comprises: Generating a database by performing at least one computer simulation, preferably FEM simulation, of at least one fictitious, partial or complete ring rolling process based on at least one set of fictitious process parameters; Creating the AI ​​model (60) by training a neural network using the generated data set; and Applying the KL model (60) by inputting process parameters of the ring rolling process of the workpiece (2) into the neural network, wherein the neural network of the KL model (60) generates an output field that is characteristic of the workpiece (2), preferably comprising a temperature and / or shape change distribution of the workpiece (2).

2. Method according to claim 1, characterized in that at least one material property of the workpiece (2) is determined from the output field of the neural network of the Kl model (60), wherein the at least one material property preferably comprises a local grain size distribution and / or a pore closure behavior and / or a damage development in the workpiece (2).

3. Method according to claim 1 or 2, characterized in that the output field of the neural network of the KL model (60) and / or the determined at least one material property of the workpiece (2) is used for The control and / or regulation and / or monitoring of the ring rolling process is used.

4. Method according to claim 3, characterized in that results from the ring rolling process are fed back into the Kl model (60).

5. Method according to one of the preceding claims, characterized in that the output field of the neural network of the Kl model (60) and / or the determined at least one material property of the workpiece (2) is used to determine a preform of the workpiece (2).

6. Method according to one of the preceding claims, characterized in that the database for the neural network is generated by finite element modeling by varying various fictitious process parameters and generating a multidimensional grid of the fictitious process parameters such that each point in the grid corresponds to a combination of the fictitious process parameters, and / or the database for the neural network is generated or supplemented by means of supervised learning.

7. Method according to claim 6, characterized in that the fictitious Process parameters workpiece dimensions and / or This includes tool dimensions and / or temperatures and / or pause times and / or ring growth speeds and / or rolling curves.

8. Method according to claim 6 or 7, characterized in that on the Based on the defined grid, FEM simulations are performed, the results of which preferably include nodal displacements and / or Temperatures and / or changes in shape and / or Include shape change increments and / or shape change rates.

9. Method according to one of the preceding claims, characterized in that the neural network of the Kl model (60) trained on the data basis generates the output field for combinations of process parameters not taken into account when creating the data basis.

10. Device (1) for rolling a ring-shaped workpiece (2) with a rolling device and a control device (50) for controlling and / or regulating and / or monitoring a ring rolling process that can be carried out by the rolling device (1), wherein the control device (50) comprises a computer model (60) with a neural network configured to generate an output field characteristic of the workpiece (2) based on process parameters of the ring rolling process, preferably comprising a temperature and / or deformation distribution of the workpiece (2).

11. Device (1 ) according to claim 10, characterized in that the control device (50) is configured to determine at least one material property of the workpiece (2) from the output field of the neural network of the KL model (60), wherein the at least one material property preferably comprises a local grain size distribution and / or a pore closure behavior and / or a damage development in the workpiece (2).

12. Device (1) according to claim 10 or 11, characterized in that the control device (50) is configured to use the output field of the neural network of the KL model (60) and / or the determined at least one material property of the workpiece (2) for controlling and / or regulating and / or monitoring the ring rolling process.

Images