AI model for controlling and / or regulating a forging process

Abstract

A method for creating and applying an AI model (60) for controlling an apparatus (50) for controlling and / or regulating and / or monitoring a forging process, preferably a free forging process, of a workpiece (2), wherein the method comprises: generating a database by performing at least one computer simulation, preferably a FEM simulation, of at least one virtual forging process based on at least one set of virtual process parameters; creating the AI model (60) by training a neural network with the aid of the generated database; and applying the AI model (60) by inputting process parameters of a forging process of the workpiece (2) into the neural network, wherein the neural network of the AI model (60) generates an output field representing a characteristic of the workpiece (2), preferably comprising a temperature distribution and / or a deformation distribution of the workpiece (2).

Description

Technical Field

[0001] This invention relates to an AI model for creating and applying control devices to control and / or regulate forging processes, particularly free forging processes, of workpieces. The invention also relates to a system comprising forging equipment for forging workpieces, preferably a free forging press, operating equipment for conveying and orienting the workpieces, and control devices for controlling and / or regulating the forging process. Background Technology

[0002] Forging, as a chipless forming method, is primarily used for steel, but also for some non-ferrous metals. The work required for shaping is introduced into the workpiece in the form of kinetic energy. If the forming motion is not restricted by fixed supports or other mechanical means during this process, it is called free forging.

[0003] In free forging, for example, a rough estimate of the geometric evolution of the component is made based on analytical equations to formulate a pass schedule. Through semi-empirical models, it is possible to predict the deformation distribution along the core fibers of the workpiece by applying analytical equations and fitting parameters specific to the material and temperature. For this, see the following references: Dominik Recker (2014), “Entwicklung von schnellen Prozessmodellen und Optimierungsmöglichkeiten für das Freiformschmieden”, PhD dissertation at Aachen IBF; Dirk Rosenstock (2018), “Schnelle Prozessmodellierung, Online-Visualisierung und Optimierungbeim Freiformschmieden”, PhD dissertation at Aachen IBF.

[0004] To date, 3D prediction of deformation distribution can only be achieved based on finite element modeling (FEM). Finite element modeling can also be used to predict grain size distribution, pore closure behavior, and damage development in materials. For further information, please refer to the following references: Wolfgarten, M., Rosenstock, D., Rudolph, F. and Hirt, G. (2019), “New approach for the optimization of pass-schedules in open-die forging”, International Journal of Material Forming, Vol. 12, pp. 973-983; Rudolph, F., Wolfgarten, M., Keray, V. and Hirt, G. (2021), “Optimization of Open-Die Forging Using Fast Models for Strain, Temperature, and Grain Size in the Context of an Assistance System”, Forming the Future: Proceedings of the 13th International Conference on the Technology of Plasticity, pp. 1145-1159, Springer International.

[0005] While a rough estimate of geometric evolution using simple analytical equations can be made very quickly, for example, within seconds, the predictive quality provided by such a rough estimate for calculating material behavior in the forging process is insufficient. The same is true for predicting the deformation distribution along the core fibers of a workpiece using semi-empirical models with analytical equations; this information is particularly inadequate for estimating overall component quality. Beyond the core fibers, grain size distribution and damage development also decisively influence component performance.

[0006] On the other hand, predictions using FEM are very time-consuming. A single simulation can take anywhere from several hours to several days. Therefore, predictions using FEM are not suitable for rapid production planning or for optimizing the free forging process in terms of the quality of the components produced.

[0007] In practice, for free forging, the prediction of component quality and the required number of passes has so far relied primarily on experience. In this context, experiential knowledge is not necessarily acquired in a structured manner within the company, but rather held by the individual employees. This can lead to individualized fluctuations and corresponding risks of disruption.

[0008] Currently, model- and computer-aided deformation distribution prediction (which serves as the basis for component quality prediction in free forging) is either insufficiently accurate, only applicable to the core fibers, or extremely time-consuming, as described above. Therefore, it has been impossible to design, control, or optimize the free forging process as a complex modification process in terms of material properties that decisively influence component quality and performance. Furthermore, it has been impossible to perform 3D process monitoring during free forging. Summary of the Invention

[0009] The purpose of this invention is to improve the control and / or adjustment and / or monitoring of the forging process of workpieces, especially the free forging process.

[0010] This objective is achieved by a method having the features of claim 1 and a system having the features of claim 8. Advantageous improvements are derived from the dependent claims, the following description of the invention, and the description of preferred embodiments.

[0011] This invention relates to the forging of workpieces, particularly free forging. The forging is carried out in a forging apparatus, preferably configured as a free forging press. Metallic materials, particularly steel, nickel, titanium, zirconium, and aluminum, are considered as materials for the workpiece. Preferably, forging is performed at a temperature above 800°C. The desired geometry achieved through forging includes, for example, circles, rectangles, squares, and quadrilaterals.

[0012] Specifically, the method according to the invention is used to create and apply an AI model for a control device to control and / or regulate and / or monitor the forging process of a workpiece, preferably a free forging process.

[0013] According to this method, a database is first generated by performing at least one computer simulation, preferably a FEM simulation, of at least one virtual forging process based on at least one set of virtual process parameters. "Virtual" here means that the process parameters used for (multiple) simulations do not necessarily have to be exactly the same as the process parameters of the (real) forging process to be controlled, adjusted, or monitored.

[0014] Based on the database generated in this way, AI models are created by training neural networks.

[0015] Neural networks can be classified as convolutional recurrent neural networks, which include at least one convolutional layer and a pooling layer for downsampling, at least one recurrent layer for describing the time history, and, if necessary, at least one transcription layer and an output layer. The exact number of layers and the number of neurons in each layer can vary and be determined depending on the application.

[0016] By inputting the process parameters of the forging process into a neural network, an AI model thus created is applied to the (real) forging process, wherein the neural network of the AI ​​model generates an output field representing the characteristics of the workpiece. Preferably, this output field includes the temperature distribution and / or deformation distribution of the workpiece.

[0017] The input of process parameters into the neural network can also be automated, for example, through measuring instruments in the process, such as cameras, displacement measurement systems for mold positions, timers, scales, etc.

[0018] This paper describes an AI model using at least one neural network that allows for highly accurate prediction of characteristic material properties in forging processes, such as geometric evolution, temperature distribution, and deformation distribution, within a short timeframe. Based on this, pass planning calculations can be performed.

[0019] Building on this, fast computation times of a few seconds (<10 seconds) enable optimization of component quality and performance in terms of selected target variables, such as uniform and fine grain size, maximization of pore closure, or minimization of introduced damage.

[0020] Therefore, preferably, at least one material property of the workpiece is obtained from the output field of the neural network of the AI ​​model, wherein the at least one material property includes local grain size distribution and / or pore closure behavior and / or damage development in the workpiece. The neural network-based prediction system can combine predictions of grain size distribution in the material, predictions of pore closure behavior at casting shrinkage cavities and similar defect sites in the material, predictions of damage development, and predictions of component defects that may occur during the forging process. These properties can be predicted in the component, and the manufacturing process can be designed specifically according to these criteria. Therefore, possible pass planning calculations can take into account established criteria regarding component quality and performance, thereby optimizing the process.

[0021] Preferably, the output field of the neural network of the AI ​​model and / or at least one material property of the acquired workpiece is used to control and / or regulate and / or monitor the forging process. This feedback to the equipment controller enables the process execution to be as close as possible to the optimized process design. Process monitoring can provide the equipment operator with further information on the evolution of the geometry and can issue warning messages in cases where the process deviates from the planned pass calculations and therefore there is a risk that the component will not meet the required quality and performance.

[0022] Preferably, a database for neural networks is generated through finite element modeling by changing different virtual process parameters and generating a multidimensional grid of virtual process parameters, such that each point in the grid corresponds to a combination of virtual process parameters.

[0023] Preferably, the virtual process parameters include workpiece size and / or mold size and / or temperature and / or pause time and / or pass compression.

[0024] Different parameter ranges can be used for rasterization using FEM, and these parameter ranges involve geometry type, material, size, process parameters, etc., which are exemplarily given below in the description of the implementation.

[0025] Preferably, FEM simulation is performed based on a defined grid, and the results of the FEM simulation preferably include nodal displacements and / or temperatures and / or deformations (such as elongation and shear) and / or deformation increments and / or deformation rates. In addition to the local variables in the material mentioned herein, the results may additionally or alternatively include global variables such as process forces, energy consumption, required process time, etc. In this respect, the system enables the optimization and regulation of the pass schedule optimized according to these criteria. Furthermore, the neural network can be trained using results from different equipment (e.g., two hydraulic presses with pressures of 40MN and 50MN). Thus, when generating the pass schedule, preferred modified equipment or adherence to specific maximum forces is also achieved.

[0026] Therefore, the database used to train the neural network consists in particular of a multidimensional grid of process parameters (including relevant FEM input parameters), the corresponding definition of the FEM model, and the cell- and node-based results of (multiple) simulations.

[0027] Preferably, the neural network of the AI ​​model, trained on a database, generates an output field for combinations of process parameters not considered when the database was created. In other words, by training on the database, the neural network can generate an output field in the form of temperature and deformation distributions for combinations of process parameters (or grid points) that have not been considered so far, within a short time, especially within a few seconds (<10 seconds). The prediction accuracy is very high within the grid area, while the prediction reliability is low outside the grid area.

[0028] The above objective is also achieved by a system having forging equipment for forging workpieces, preferably a free forging press, operating equipment for conveying and orienting the workpieces, and control device for controlling and / or regulating and / or monitoring the forging process that can be performed by the forging equipment and the operating equipment, wherein the control device includes an AI model with a neural network adapted to generate an output field based on process parameters of the forging process, the output field representing the characteristics of the workpiece, preferably including the temperature distribution and deformation distribution of the workpiece.

[0029] The features, technical effects, advantages, and embodiments described with reference to the method are similarly applicable to the system.

[0030] Therefore, for the reasons mentioned above, the control device is preferably adapted to obtain 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 includes local grain size distribution and / or pore closure behavior and / or damage development in the workpiece.

[0031] Preferably, for the reasons stated above, the control device is adapted to use the output field of the neural network of the AI ​​model and / or at least one material property of the workpiece acquired to control and / or regulate and / or monitor the forging process.

[0032] Other advantages and features of the present invention will become apparent from the following description of preferred embodiments. The described features may be implemented individually or in combination with one or more of the features set forth above, provided that these features do not contradict each other. Hereinafter, preferred embodiments are described with reference to the accompanying drawings. Attached Figure Description

[0033] Other preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Wherein:

[0034] Figure 1 A system with forging equipment and operating equipment is schematically shown; and

[0035] Figure 2 The process for creating and applying an AI model based on a neural network and used in the control device of a forging equipment is illustrated schematically. Detailed Implementation

[0036] Preferred embodiments are described below with reference to the accompanying drawings.

[0037] Figure 1 The system 1 is schematically shown, having a forging device 10 for forging workpiece 2 and an operating device 20.

[0038] The forging equipment 10 is preferably designed for free forging, and in particular, is designed as a free forging press. In this case, with the aid of a piston-cylinder system and forging dies, the workpiece 2 (in...) Figure 1 The piston force is applied to the workpiece (located outside the forging equipment 10) to deform it in a desired manner.

[0039] As the material used for workpiece 2, metallic materials are considered, especially steel, nickel, titanium, zirconium, and aluminum. Preferably, forging is carried out at a temperature above 800°C. The desired geometry achieved through forging includes, for example, circles, rectangles, squares, and quadrilaterals.

[0040] The operating device 20 is adapted to move the workpiece 2, particularly to move it to a desired position and adjust it to a desired orientation. For this purpose, the operating device 20 includes a conveying section 22 (e.g., implemented as a vehicle, particularly an autonomous vehicle) and a robotic arm section 24 adapted to hold the workpiece 2, rotate it, swing it, etc. The robotic arm section 24 preferably includes clamps for gripping the workpiece 2.

[0041] In addition, system 1 includes a control device 50 that communicates with various devices, components, etc. The control device 50 is connected in signal technology to the components of system 1 that are to be controlled, adjusted, and / or read, and therefore particularly to the forging equipment 10 and the operating equipment 20.

[0042] Communication between the control device 50 and the component to be controlled, adjusted, and / or read can be achieved via wired or wireless means, and digital or analog means. The control device 50 can accordingly receive and / or transmit signals (control signals, data, etc.), wherein, in this document, signal transmission in one direction and signal transmission in two directions are both covered by the term "communication." Here, the control device 50 is not necessarily implemented through a central computing device or electronic control device, but includes distributed systems and / or multi-level systems, control networks, cloud systems, etc. Furthermore, the control device can be an integrated component of a higher-level equipment controller or communicate with such a component. The control device 50 can also communicate with lower-level equipment controllers, i.e., controllers associated with the corresponding devices.

[0043] The control device 50 includes or has access to an AI model 60 (“AI” stands for Artificial Intelligence) based on one or more neural networks. For a given forging process, the AI ​​model 60 allows prediction, within a short time, especially within a few seconds, of the geometric evolution of the workpiece 2 during forging, as well as the temperature and deformation distribution within the workpiece 2.

[0044] As the database for the underlying neural network of AI model 60, FEM simulation (“FEM” here stands for Finite Element Method) can be used, in which process parameters (workpiece and die dimensions, temperature, pause time, pass compression, etc.) are varied. For this purpose, a database is generated using an appropriate number of reference points (in combination with different process parameters), and the neural network is trained using this database. Thus, by accessing AI model 60, control device 50 can predict the temperature and deformation distribution of workpiece 2, at least within a selected range. This range is selected based on the possible, desired spectrum of process parameters.

[0045] Based on the prediction of the temperature and deformation distribution of workpiece 2, it is also possible to predict material properties that play a decisive role in the quality and performance of the component, such as grain size distribution, pore closure behavior and damage development.

[0046] This enables process design and turn planning calculations from initial geometry to final geometry to be performed within seconds using AI model 60 and based on at least one neural network.

[0047] Furthermore, by predicting temperature and deformation distributions and based on AI model 60, grain size distribution, pore closure behavior, and damage development can be predicted, for example, using conventional models. Therefore, process design can be optimized to meet or comply with established standards for component quality and performance.

[0048] The process design described above is used to control and / or regulate and / or monitor the free forging process so that the component manufactured from workpiece 2 follows the process design as precisely as possible.

[0049] Similarly, during the process, the current temperature and deformation distribution in workpiece 2 can be calculated almost in real time (e.g., with a delay of less than 2 seconds), and based on this, the current grain size distribution, porosity, and damage development can also be determined and displayed if necessary. Furthermore, this process monitoring can be used, for example, to issue a warning to the operator if the process is expected to cause the workpiece to fall outside the predetermined component quality range.

[0050] Figure 2 An exemplary process for creating and applying an AI model 60 for control device 50 is illustrated schematically. Here, the database required for training the neural network of AI model 60 is generated via FEM simulation.

[0051] In the first step S1, a multidimensional grid is defined for (multiple) FEM simulations, wherein different parameter ranges can be used for the grid points to be simulated in the rasterized, or more precisely, multidimensional grid.

[0052] Therefore, in terms of geometric type, it can cover, for example, straight members with circular or polygonal cross-sections, multi-level axes, stepped axes, curved members, twisted members, and combinations of various geometric variations.

[0053] Regarding the material of workpiece 2, iron-based materials and steel, as well as materials based on aluminum, titanium, nickel, copper, zinc, and alternative modified materials can be considered, with workpiece temperatures ranging from -273°C to 2800°C.

[0054] For the mass and dimensions of the components, the range is from 0.1 kg to 5000 t input weight (Einsatzgewicht), from 10 mm to 120 m in length, and from 10 mm to 25 m in side length (or diameter in the case of a circular cross-section).

[0055] As a modification device, the forging equipment 10 may include: one or more hammers with impact energy from 1 Nm to 125,000 kNm; a screw press with modification energy from 1 J to 250,000 kJ; and a hydraulic press, servo-hydraulic press or servo-mechanical press with a maximum force from 1 kN to 10,000 MN; a crank lever press, etc.

[0056] As process parameters, they cover the relative compression of each pass from 0.1% to 90%, the bite width from 1 mm to 10 m, the rotation angle between each pass or stroke from 0.1° to 180° if necessary, and the bending or twisting angle between each pass from 0.1° to 75°.

[0057] In addition, different boundary conditions can be used for rasterization to create FEM simulation models, including ambient temperatures of -100°C to +900°C and mold temperatures of -100°C to +1400°C.

[0058] Subsequently, in the second step S2, the database required for training the neural network 60 is generated through FEM simulation.

[0059] In this regard, at least one FEM simulation model is established, which corresponds to the aforementioned parameters of the grid points to be simulated in a multidimensional mesh. Each point in the grid corresponds to a combination of process parameters subsequently calculated for at least one pass using a suitable finite element program (e.g., Simufact, Forge NxT, Deform, Q-Form, or similar programs). Here, the simulation may include a single or multiple strokes, one or more passes, or even the entire forging process, including intermediate heat treatments.

[0060] Here, for the workpiece 2 under consideration, the mesh is preferably generated with at least 4 nodes in the 2D case, or preferably with at least 8 nodes in the 3D case. In each simulation, preferably a maximum of 10,000,000 or 1,000,000 nodes are used to form the FEM mesh.

[0061] Initiate the FEM simulation and store the simulation results after the calculation is complete. The stored data includes nodal displacements, temperatures, deformations (elongation and shear), deformation increments, and deformation rates. The simulation results thus generated are preferably automatically exported to a database.

[0062] The sum of all input data is paired with the corresponding stored simulation results to form a database for neural networks.

[0063] Helpful here is to define each FEM simulation so that the number of elements in the model, and consequently the number of reference points, is as similar as possible across all simulations in the database. However, this is not mandatory.

[0064] Therefore, the database used to train the neural network consists of a multidimensional grid of process parameters (including relevant FEM input parameters), the corresponding definition of the FEM model, and simulation results based on cells and nodes.

[0065] In the third step S3, the neural network of AI model 60 is trained. This is done in a suitable programming environment based on the generated database.

[0066] The neural network can be classified, for example, as a convolutional recurrent neural network, comprising at least one convolutional layer and a pooling layer for downsampling, at least one recurrent layer for describing the time history, and, if necessary, at least one transcription layer and an output layer. The exact number of layers and the number of neurons in each layer are determined by the size of the database and, in particular, the definition of the finite element model (i.e., the number of units and nodes).

[0067] Now, when applying the neural network trained in this way (step S4), input points within the multidimensional grid can be used, which do not necessarily correspond to grid points. It should be noted that the database should be selected such that future predictions do not require points outside the grid. Although results are generated for such points, the deviation from reality increases with distance from the grid.

[0068] Subsequently, for each input point, the neural network generates predictions for the selected simulation results, namely at least nodal displacement, temperature, deformation (elongation and shear), deformation increment, and deformation rate.

[0069] Therefore, by training on a database, the neural network can generate an output field in the form of temperature and deformation distributions for combinations of previously unconsidered process parameters (or grid points) within a short time, such as a few seconds (<10 seconds). The prediction accuracy is very high within the grid area, but the reliability of predictions outside the grid is low.

[0070] The output field can be used, with the aid of appropriate models and based on temperature and deformation distributions, to predict material properties that play a decisive role in component quality and performance. These material properties particularly relate to local grain size distribution, pore closure behavior, and / or damage development within the component.

[0071] For example, the resulting variables can be combined with conventional models used to predict grain size, damage values, and / or pore closure to predict component performance and optimize it manually or automatically.

[0072] The optimized process design is used to control and / or adjust and / or monitor the free forging process (press movement and robot movement) via control device 50. In case of deviations from the optimized process, recalculation can be performed, and subsequent processes can be corrected as needed and possible.

[0073] If the process deviates significantly from the preset parameters, making it impossible to achieve the required component quality, or if reheating is required, the control device 50 can issue a corresponding warning message, which the operator can then respond to.

[0074] The following specific examples illustrate the process of creating and applying AI model 60 described above:

[0075] Assume a rectangular workpiece 2 forged from structural steel, with a single piece weight ranging from 100kg to 25t, an initial geometry ranging from 320mm×320mm×125mm to 1265mm×1265mm×2,000mm, and finished dimensions ranging from 160mm×160mm×500mm to 632.5mm×632.5mm×8,000mm, and the workpiece is manufactured at an initial temperature of 1250℃.

[0076] If workpiece 2 cools to below 900°C during manufacturing, it needs to be reheated. For free forging, a hydraulic press with a maximum pressure of 25MN is used as forging equipment 10. The relative compression per pass is in the range of 3% to 15%, and the bite width is 50mm to 250mm. Workpiece 2 rotates 90° between each pass. The ambient temperature is approximately 35°C, and the die temperature is approximately 300°C.

[0077] To generate the database based on step S2, simulations were performed on the range of these dimensions and process parameters, as well as fixed boundary conditions. For this purpose, boundary points and 18 additional linear points within the parameter spectrum were simulated separately. By performing 20 simulations for each variable (“initial geometry”, “final geometry”, “relative pass compression”, and “bite width”), a total of 20 × 20 × 20 × 20 = 16,000 simulations were obtained. By restricting the initial and final geometries (the minimum final geometry cannot be derived from the maximum initial geometry) and the bite width (which cannot exceed the total length of the component) in a physically meaningful manner, the number of simulations could be reduced to 10,000.

[0078] A complete process simulation typically takes about 3 days, making it initially very time-consuming. However, because many forging passes are very similar, the process can be broken down into its individual passes, each consisting of 4 to 40 passes. If passes with similar thermodynamics and geometry are aggregated, this results in 200,000 simulations, each requiring approximately 30 minutes of simulation time. With a total simulation time of 100,000 hours (instead of the 720,000 hours used for the initial 10,000 simulations), where, in principle, any number of simulations can be executed and evaluated in parallel, all simulations can be automatically started, computed, and evaluated within a week using, for example, a high-performance cluster with 700 computing cores. This includes storing the simulation results along with the input parameters in a database, making them usable for subsequent training of neural networks.

[0079] The generated database is now used, for example, to train a neural network using the C# programming language, which generates outputs corresponding to FEM simulation results within a parameter field within seconds. These generated results can then be used, just like FEM simulation results, to evaluate the performance of the manufactured components. For this purpose, existing models for predicting grain size, damage, and / or porosity closure behavior are used, where the short computation time of generating results within seconds is used to optimize the process flow, particularly the forging strategy comprised of press and robot movements, to optimize for maximizing porosity closure behavior.

[0080] The forging strategy optimized for the entire product portfolio in this manner is used to control the forging equipment 10 and operating equipment 20 in subsequent production. In the event of occasional deviations from the process strategy, an adjusted forging strategy is calculated within seconds, and the remaining process flow is replaced by the newly calculated operation. If the calculation determines that the required component quality can no longer be achieved at this temperature, a prompt is issued to the equipment operator indicating that reheating is necessary. Subsequently, for the reheated component, a forging strategy that achieves the required component quality is recalculated in an adjusted manner and implemented.

[0081] If existing measuring instruments detect deviations from the process strategy, the forging strategy can be automatically adjusted. In addition to die position, measuring instruments may also include timers (for deviations during pauses or non-processing times), cameras for position and geometry recognition, and / or scales for use in cases of excessive material loss.

[0082] Further improvements to the concept described in this paper could lead to the development of software for automatically generating databases (creating, calculating, and evaluating simulations) within suitable simulation programs.

[0083] Furthermore, systems can be developed to simplify generation / prediction through neural networks and, where necessary, automatically optimize deformation distribution (with goal-oriented approaches, such as maximizing pore closure or minimizing damage for defined grain size ranges).

[0084] This paper describes an AI model 60 using at least one neural network, which allows for highly accurate predictions of the geometric evolution, temperature distribution, and deformation distribution in free forging processes within seconds. It can also perform pass planning calculations, which similarly provide a basis for predicting component quality and performance.

[0085] Based on this, it is possible to predict, in particular, the grain size distribution, pore closure behavior, and / or damage development in components, and to design processes specifically for these criteria. Therefore, pass planning calculations can take into account established standards regarding component quality and performance, and optimize processes accordingly.

[0086] Feedback to the equipment controller enables the process to be executed as closely as possible to the optimized process design.

[0087] Process monitoring can provide equipment operators with further information about the evolution of geometry and can issue warning messages when the process deviates from the planned pass calculations and therefore there is a risk that components will not meet the required quality and performance.

[0088] Where applicable, all the various features shown in the embodiments can be combined and / or interchanged with each other without departing from the scope of the invention. List of reference numerals in the attached diagram: 1. A system with forging equipment and operating equipment 2. Workpiece 10 Forging equipment 20 Operating equipment 22 Conveying Section 24. Robotic arm section 50 Control devices 60 AI models.

Claims

1. A method for creating and applying an AI model (60), said AI model being used in a control device (50) to control and / or regulate and / or monitor the forging process of a workpiece (2), preferably a free forging process, wherein, The method includes: A database is generated by performing at least one computer simulation, preferably FEM simulation, of at least one virtual forging process based on at least one set of virtual process parameters; The AI ​​model (60) is created by training a neural network using the generated database; and By inputting the process parameters of the forging process of the workpiece (2) into the neural network, the AI ​​model (60) is applied, wherein the neural network of the AI ​​model (60) generates an output field, the output field representing the characteristics of the workpiece (2), preferably including the temperature distribution and / or deformation distribution of the workpiece (2).

2. The method according to claim 1, characterized in that, The output field of the neural network of the AI ​​model (60) is used to obtain at least one material property of the workpiece (2), wherein the at least one material property includes the local grain size distribution and / or pore closure behavior and / or damage development in the workpiece (2).

3. The method according to claim 1 or 2, characterized in that, The output field of the neural network of the AI ​​model (60) and / or at least one material property of the workpiece (2) acquired are used to control and / or regulate and / or monitor the forging process.

4. The method according to any one of the preceding claims, characterized in that, A database for the neural network is generated by finite element modeling by changing different virtual process parameters and generating a multidimensional grid of the virtual process parameters, such that each point in the grid corresponds to a combination of the virtual process parameters.

5. The method according to claim 4, characterized in that, The virtual process parameters include workpiece size and / or mold size and / or temperature and / or pause time and / or compression per pass.

6. The method according to claim 4 or 5, characterized in that, FEM simulation is performed based on the defined grid, and the results of the FEM simulation preferably include nodal displacement and / or temperature and / or deformation and / or deformation increment and / or deformation rate.

7. The method according to any one of the preceding claims, characterized in that, The neural network of the AI ​​model (60) trained on the database generates an output field for combinations of process parameters that were not considered when the database was created.

8. A system (1) comprising forging equipment (10) for forging a workpiece (2), preferably a free forging press, operating equipment (20) for conveying and orienting the workpiece (2), and control device (50) for controlling and / or regulating and / or monitoring a forging process that can be performed by the forging equipment (10) and the operating equipment (20), wherein, The control device (50) includes an AI model (60) with a neural network, which is adapted to generate an output field based on the process parameters of the forging process, the output field representing the characteristics of the workpiece (2), preferably including the temperature distribution and / or deformation distribution of the workpiece (2).

9. The system (1) according to claim 8, characterized in that, The control device (50) is adapted to obtain at least one material property of the workpiece (2) from the output field of the neural network of the AI ​​model (60), wherein the at least one material property includes local grain size distribution and / or pore closure behavior and / or damage development in the workpiece (2).

10. The system (1) according to claim 8 or 9, characterized in that, The control device (50) is adapted to use the output field of the neural network of the AI ​​model (60) and / or at least one material property of the workpiece (2) acquired to control and / or regulate and / or monitor the forging process.

Images