EP4118493A1

EP4118493A1 - Method and device for optimised production of sheet metal parts

Info

Publication number: EP4118493A1
Application number: EP21711844.7A
Authority: EP
Inventors: Carina Mieth; Jens Ottnad; Alexandru RINCIOG; Frederick STRUCKMEIER
Original assignee: Trumpf Werkzeugmaschinen SE and Co KG
Current assignee: Trumpf Werkzeugmaschinen SE and Co KG
Priority date: 2020-03-13
Filing date: 2021-03-10
Publication date: 2023-01-18
Also published as: CN115335780A; WO2021180816A1; DE102020203296A1; US20230004880A1; US11586996B2

Abstract

The invention relates to a method for optimising production of sheet metal parts. The method optimises the allocation of sheet metal parts for processing on various production machines (14) and outputs an optimised production plan. An algorithm (20) is provided for this, which has a decision tree in the form of a Monte Carlo tree search framework (22) and a neural network (24). With each new query, the algorithm (20) is trained by means of self-play and reinforcement learning. Pre-training of the algorithm (20) is achieved by means of supervised learning. The algorithm (20) preferably optimises the production plan primarily with regard to minimally behind-schedule production deadlines for the sheet metal parts and secondarily with regard to minimal waste. Both goals can be evaluated jointly through the awarding of scores. The method can comprise receiving query-triggering events 46 and/or operating production machines (14) in accordance with the production plan. The invention further relates to a device (18) for carrying out the method.

Description

Process and device for the optimized production of sheet metal parts

Background of the invention

The invention relates to a method for optimizing the production of Blechtei sources. The invention also relates to a device for performing such a method.

Sheet metal parts appear in a wide variety of products in a wide variety of geometries. To manufacture products with sheet metal parts, the sheet metal parts are cut out of a large sheet metal, separated, deburred, bent, joined, coated and / or assembled.

The sheet metal parts are manufactured in so-called orders. An order includes i) the production of a cut, separated, bent and / or assembled sheet metal part or ii) the production of several cut, separated, bent and / or assembled sheet metal parts within a given production period. The individual sheet metal parts should be cut out of the sheet metal in such a way that as little residual material (waste) as possible remains as waste from the sheet metal. Since the sheet metal parts of different orders can have different geometries, it can be advantageous to optimize waste to provide sheet metal parts of different orders to save space together on a metal sheet.

The resulting mixing of orders in time increases the complexity of production planning. In addition, the sheet metal parts can be produced on several identical or similar production machines. For example, several identical or similar bending machines can be provided for bending the separated sheet metal parts. The production machines should be operated with the highest possible utilization.

Production planning, i.e. planning when which sheet metal part is processed on which production machine, becomes very complex due to the variables described, especially in the case of events such as production machine failures, rush orders and / or production machine capacities that are freed up.

Optimal production planning is known as the solution to a job shop scheduling problem (JSSP). Solutions and approaches to this can be found in the following publications:

[1] F. Pfitzer, J. Provost, C. Mieth, and W. Liertz, "Event-driven production scheduling in job shop environments", in 2018 IEEE 14th International Conference on Automation Science and Engineering (CASE), IEEE, 2018, pp. 939-944; [2] M. Putz and A. Schlegel, "Simulation-based investigation of priority and picking rules for controlling the material flow in the sheet metal industry"; [3] LL Li, CB Li, L. Li, Y. Tang, and QS Yang, "An integrated approach for remanufacturing job shop scheduling with routing alternatives.", Mathematical biosciences and engineering: MBE, vol. 16, no. 4, pp. 2063-2085, 2019;

[4] M. Gondran, M.-J. Huguet, P. Lacomme, and N. Tchernev, "Comparison between two approaches to solve the job-shop scheduling problem with routing", 2019;

[5] J. J. van Hoorn, "The current state of bounds on benchmark instances of the job-shop scheduling problem," Journal of Scheduling, vol. 21, no. 1, pp. 127-128, 2018;

[6] S.-C. Lin, E. D. Goodman, and W. F. Punch III, "A genetic algorithm approach to dynamic job shop scheduling problem," in ICGA, 1997, pp. 481-488;

[7] T. Yamada and R. Nakano, "Scheduling by genetic local search with multistep crossover", in International Conference on Parallel Problem Solving from Nature, Springer, 1996, pp. 960-969;

[8] B. M. Ombuki and M. Ventresca, "Local search genetic algorithms for the job shop scheduling problem," Applied Intelligence, vol. 21, no. 1, pp. 99-109, 2004;

[9] E. S. Nicoara, F. G. Filip, and N. Paraschiv, "Simulation-based optimization using genetic algorithms for multi-objective flexible jssp", Studies in Informatics and Control, vol. 20, no. 4, pp. 333-344, 2011;

[10] L. Asadzadeh, "A local search genetic algorithm for the job shop scheduling problem with intelligent agents," Computers & Industrial Engineering, vol. 85, pp. 376-383, 2015;

[11] B. Waschneck, A. Reichstaller, L. Belzner, T. Altenmüller, T. Bauernhansl, A. Knapp, and Kyek, "Optimization of global production scheduling with deep reinforcement learning", Procedia CIRP, vol. 72, pp. 1264-1269, 2018;

[12] M. Botvinick, S. Ritter, J. X. Wang, Z. Kurth-Nelson, C. Blundell, and D. Hasabis, "Reinforcement learning, fast and slow", Trends in cognitive Sciences, 2019.

Furthermore, it has become known from WO 2017/157809 A1 to provide a production planning with an optimization unit and a distribution unit separate therefrom. Despite extensive efforts, due to the complexity of the task, a satisfactory production planning has not yet been achieved.

Object of the invention

It is therefore the object of the invention to provide a method and a device for the optimized production of sheet metal parts.

Description of the invention

This object is achieved according to the invention by a method according to claim 1 and a device according to claim 13. The subclaims reproduce preferred developments.

The solution according to the invention thus comprises a method for optimizing the production of sheet metal parts. The method comprises at least the following process steps (before, after and / or between the subsequent process steps, a further process step or several further process steps can be provided): a) cutting out and separating the sheet metal parts (in particular by means of punching or laser cutting); b) bending the sheet metal parts.

The method has at least the following method steps (before, after and / or between the following method steps, a further method step or several further method steps can be provided):

A) Training of a neural network carried out on a Monte Carlo tree search framework by means of supervised learning and self-play with reinforcement learning; B) detecting boundary conditions of the sheet metal parts, the boundary conditions including at least geometric data of the sheet metal parts;

C) Creation of an optimized production plan by the neural network;

D) Output of the production plan. According to the invention it is therefore provided to provide an optimization with a neural network (NN). Neural networks are known to the person skilled in the art, for example, from: [13] Günter Daniel Rey, Karl F. Wender, "Neural Networks", 2nd edition, 2010,

Huber.

The neural network has decision nodes connected via edges. In the present case, these are part of a Monte Carlo tree search (MCTS) framework, ie an algorithm with a decision tree. A promising path is selected in the decision tree (selection), the path is expanded, a simulation is carried out on the basis of the extended path (simulation) and, on the basis of the simulation result, feedback, in particular in the form of a strengthening or weakening, is sent to the Decision tree given (back propagation). Details on the implementation of an MCTS framework can be found in the following publication:

[14] G. Chaslot, S. Bakkes, I. Szita, and P. Spronck, "Monte Carlo tree search: A new framework for game ai", in AIIDE, 2008. In the present case, the MCTS is carried out by the neural network, with the neural network Network is pre-trained through supervised learning. Decision-making and further training take place using self-play and reinforcement learning. Reinforcement learning (RL) is understood to be a feedback-based learning process that includes, in particular, the strengthening or weakening of the decision tree of the MCTS framework. Reinforcement learning generally stands for a number of methods of machine learning in which an agent himself constantly learns a strategy to maximize rewards received. The agent is not shown which action is best in which situation but at certain times he receives a reward, which can also be negative. Using these rewards, he approximates a utility function that describes the value of a certain state or action. Details on the implementation can be found in the following publications:

[15] W. Zhang and T. G. Dietterich, "A reinforcement learning approach to job-shop scheduling", in IJCAI, Citeseer, vol. 95, 1995, pp. 1114-1120;

[16] R. S. Sutton, A. G. Barto, et al., Introduction to reinforcement learning, 4. MIT press Cambridge, 1998, vol. 2;

[17] S. Mahadevan and G. Theocharous, "Optimizing production manufacturing using reinforcement learning.", In FI_AIRS Conference, 1998, pp. 372-377;

[18] S. J. Bradtke and M. 0. Duff, "Reinforcement learning methods for continuation-time markov decision problems", in Advances in neural information processing systems, 1995, pp. 393-400;

[19] S. Riedmiller and M. Riedmiller, "A neural reinforcement learning approach to learn local dispatching policies in production scheduling", in IJCAI, vol. 2, 1999, pp. 764-771;

[20] C. D. Paternina-Arboleda and T. K. Das, "A multi-agent reinforcement learning approach to obtaining dynamic control policies for stochastic lot scheduling problem," Simulation Modeling Practice and Theory, vol. 13, no. 5, pp. 389-406, 2005;

[21] T. Gabel and M. Riedmiller, "Scaling adaptive agent-based reactive job-shop scheduling to large-scale problems", in 2007 IEEE Symposium on Computational Intelligence in Scheduling, IEEE, 2007, pp. 259-266;

[22] YCF Reyna, YM Jim ^' enez, JMB Cabrera, and BMM Hernandez, "A reinforcement learning approach for scheduling problems," Investigacion Operacional, vol. 36, no. 3, pp. 225-231, 2015;

[23] S. Qu, J. Wang, S. Govil, and JO Leckie, "Optimized adaptive scheduling of a manufacturing process system with multi-skill workforce and multiple machine types: An ontology-based, multi-agent reinforcement learning ap proach ", Procedia CIRP, vol. 57, pp. 55-60, 2016;

[24] V. Mnih, K. Kavukcuoglu, D. Silver, A. Graves, I. Antonoglou, D. Wierstra, and M. Ried-miller, "Playing atari with deep reinforcement learning", arXiv preprint arXiv: 1312.5602, 2013;

[25] A. Kuhnle, L. Schäfer, N. Stricker, and G. Lanza, "Design, Implementation and evaluation of reinforcement learning for an adaptive order dispatching in job shop manufacturing systems ", Procedia CIRP, vol. 81, pp. 234-239, 2019;

[26] N. Stricker, A. Kuhnle, R. Sturm, and S. Friess, "Reinforcement learning for adaptive order dispatching in the semiconductor industry," CIRP Annals, vol. 67, no. 1, pp. 511-514, 2018;

[27] J. Schulman, S. Levine, P. Abbeel, M. Jordan, and P. Moritz, "Trust region policy optimization", in International Conference on machine learning, 2015, pp. 1889-1897.

Under supervised learning a training with given solutions is understood. This supervised learning is generally a branch of machine learning. Learning here means the ability of an artificial intelligence to reproduce laws. The results are known through the laws of nature or expert knowledge and are used to learn the system. A learning algorithm tries to find a hypothesis that makes predictions that are as accurate as possible. A hypothesis is to be understood as a mapping that assigns the presumed output value to each input value. The method is therefore based on a predetermined output to be learned, the results of which are known. The results of the learning process can be compared, ie "monitored", with the known, correct results. Details on the implementation can be found in the following publications:

[28] M. Gombolay, R. Jensen, J. Stigile, S.-H. Son, and J. Shah, "Apprenticeship scheduling: Learning to schedule from human experts", AAAI Press / International Joint Conferences on Artificial Intelligence, 2016;

[29] H. Ingimundardottir and T. P. Runarsson, "Supervised learning linear priority dispatch rules for job-shop scheduling", in International Conference on learning and intelligent optimization, Springer, 2011, pp. 263-277.

The algorithm is preferably executed in the form of a single player game.

The combination of a Monte Carlo tree search framework based neural network and training of this neural network by means of supervised learning and self- play with reinforcement learning leads to an optimization that significantly surpasses the known optimizations in sheet metal processing.

Preferred embodiments

The output in method step D) can be made to a manufacturing execution system (MES). This allows the production plan to be implemented directly on the production machines.

The method according to the invention can have one or more of the following process steps in addition to those already mentioned: c) deburring the sheet metal parts; d) joining, in particular welding and / or soldering, the sheet metal parts; e) Coating the sheet metal parts, in particular by painting and / or

Powder coating; f) Assemble the sheet metal parts.

Each of these process steps can be carried out by production machines and optimized by the method according to the invention.

In a preferred embodiment of the invention, the method according to the invention is carried out with the AlphaGo algorithm, in a particularly preferred embodiment with the AlphaGo Zero algorithm. In this case, the algorithm comprises the previously described Monte Carlo tree search framework with the neural network trained by means of supervised learning and self-play with reinforcement learning. AlphaGo or AlphaGo Zero has proven to be a very powerful algorithm for optimizing the production of sheet metal parts within the scope of the implementation of the invention.

The AlphaGo Zero algorithm can be viewed on the following websites:

• https://tmoer.github.io/AlphaZero/

• https://towardsdatascience.com/alphazero-implementation-and-tutorial- f4324d65fdfc • https://medium.com/applied-data-science/how-to-build-your-own-alpha- zero-ai-using-python-and-keras-7f664945cl88

AlphaGo or AlphaGo Zero is preferably implemented in Python and / or Tensorflow. Further details on the implementation of AlphaGo or AlphaGo Zero can be found in the following publications:

[30] D. Silver, A. Huang, CJ Maddison, A. Guez, L. Sifre, G. Van Den Driessche, J. Stepwieser, I. Antonoglou, V. Panneershelvam, M. Lanctot, et al., "Mas - tering the game of go with deep neural networks and tree search ", nature, vol. 529, no.7587, p. 484, 2016.

[31] G. Chaslot, S. Bakkes, I. Szita, and P. Spronck, "Monte-Carlo tree search: A new framework for game ai.", In AIIDE, 2008.

[32] D. Silver, J. Stepwieser, K. Simonyan, I. Antonoglou, A. Huang, A. Guez, T. Hubert, L. Baker, M. Lai, A. Bolton, et al., "Mastering the game of go without human knowledge, "Nature, vol. 550, no.7676, p. 354, 2017.

[33] D. Silver, T. Hubert, J. Stepwieser, I. Antonoglou, M. Lai, A. Guez, M. Lanctot, L. Sifre, D. Kumaran, T. Graepel, et al., "Mastering chess and shogi by self-play with a general reinforcement learning algorithm ", arXiv preprint arXiv: 1712.01815, 2017.

The disclosure of all publications and websites cited here is fully incorporated into the present description (incorporated by reference).

The training in method step A) is more preferably carried out with heuristically determined solutions of optimized production plans. This gives the neural network a good starting point for further optimization.

In particular, optimized production plans in the form of earliest due date (EDD) solutions can be used. These solutions have proven to be particularly advantageous, since in practice there are often rush orders that make previous production planning obsolete. A particularly preferred embodiment of the method relates to the case that the optimization includes both the minimization of waste and the optimization of production times. This enables both fast and inexpensive and resource-saving production. The goals of production time optimization are, in particular, the minimum total delay and / or the minimum total production time.

The boundary conditions in process step B) can include the production deadlines for the sheet metal parts. The production time optimization can then take into account compliance with the production deadlines. Compliance with production deadlines can be given a higher priority than other goals.

As an alternative or in addition to this, the boundary conditions in method step B) can include the values, that is to say the monetary values or prices, of the sheet metal parts. This allows production to be optimized depending on the values of the respective sheet metal parts. In general, this allows the value of a sheet metal part, for example the price of its late manufacture, to be qualified within the scope of the optimization according to the invention. It is further preferred that a waste score is allocated to the waste and, when the production deadline is reached, a production deadline score is assigned that is based on the value of the sheet metal parts, with the optimization minimizing both the waste score and the production deadline score. By assigning the scores, the production time minimization can be treated or optimized on the same scale as the waste minimization.

In this case, the estimated maximum achievable total score value is preferably stored in the decision node; The probability (= weighting) that the respective decision of the decision node is the best is preferably stored on the edges connecting the decision nodes. The waste score and the production deadline score can be used, for example, in the form of a price. Then the price for waste material can be weighed against the price of a sheet metal part produced too late. As part of the procedure, the following function can be used to optimize:

Where c (W) represents the value for the total material used (including waste, i.e. waste), T, and v, respectively represent the delay and the value of the order part i l is a parameter that penalizes delay. r _abs reflects the sum of the sheet metal parts, each reduced proportionally to the production deadlines, minus the total material costs. The formula can be used to generate a reward for the neural network, in particular scaled to [0, 1], the maximum possible score being r _max (without delay and without waste). Process steps B) to D) can be triggered as required by the presence of an event, the event being read in via an event interface.

The event is preferably in the form of a request for further processing of a sheet metal part, in the form of production machine capacity being freed up, in the form of a production machine failure and / or in the form of a rush order.

The event can be triggered automatically and read in via the event interface. The event is particularly preferably triggered by a production machine, an indoor localization system and / or a manufacturing execution system and read in via the event interface. In the case of an indoor localization system, the planning can be further optimized in an automated manner using events transmitted by the tags of the indoor localization system. To further improve the neural network, a user evaluation of the production plan output in process step D) can be read in in a method step E). The invention further relates to a method for the production of sheet metal parts, in which a previously mentioned method is carried out and then the process steps a) and b) are carried out on the basis of the optimized production plan. In the method for producing sheet metal parts, after process steps a) and b), process steps c), d), e) and / or f) can be carried out on the basis of the optimized production plan.

The object of the invention is also achieved by a device for performing a method described here, the device having a computer for storing and executing the neural network, a boundary conditions interface for reading in the boundary conditions and a production plan interface for outputting the production plan. A user evaluation interface can be provided for reading in the user evaluations. The neural network can be cloud-based in order to facilitate training with, in particular anonymized, user reviews.

The device according to the invention can have the event interface and furthermore have a production machine, an indoor localization system (with several tags that transmit events) and / or a manufacturing execution system, one of the production machine, the indoor localization system and / or The event triggered by the manufacturing execution system can be read in via the event interface. In this case, the device can be optimized in an automated or partially automated manner.

Further advantages of the invention emerge from the description and the drawing voltage. The aforementioned and those detailed below can also be used Features according to the invention can be used individually or collectively in any combination. The embodiments shown and described are not to be understood as an exhaustive list, but rather have an exemplary character for describing the invention.

Detailed description of the invention and drawing

Fig. 1 shows schematically the production sequence in the manufacture of sheet metal parts. Fig. 2 shows schematically the optimization of the production process. Fig. 1 shows schematically the production of various orders. In Fig. 1, the orders Aoi to Aio are shown by way of example. The orders Aoi-Aio include the manufacture of products Poi to Pio, which are made from several, in particular different, sheet metal parts with their respective geometric data. For the sake of clarity, only the sheet metal parts Bi and B _{2 are provided} with a reference symbol in FIG. 1.

As indicated by clock symbols in FIG. 1, the individual sheet metal parts Bi, B _{2 have} different production times. Furthermore, orders A _0i to A _{i0 have} different production _{deadlines F 01} to F ₁₀ . Piggy banks indicate that the sheet metal parts Bi, B _{2 have} different (money) values. The specifications described ben represent boundary conditions 10 of the sheet metal parts Bi, B ₂ .

The sheet metal parts Bi, B ₂ are arranged on a sheet metal 12 in such a way that the waste is minimal. As can be seen from FIG. 1, this can lead to the mixing of sheet metal parts Bi, B _{2 of} different orders A01-A10. The sheet metal parts Bi, B ₂ are processed on production machines 14, of which in Fig. 1 production machines Ci, C ₂ (cut) for cutting and singling, production _{machines bi, b 2} (bend) for bending and production machines ai, a ₂ (assemble ) for mounting the sheet metal parts Bi, B ₂ are shown. In addition, further production machines 14, not shown in FIG. 1, can be provided for processing the sheet metal parts Bi, B ₂ , for example for deburring, joining and / or coating the sheet metal parts Bi, B ₂ . The finished products comprising the sheet metal parts Bi, B ₂ are shown in FIG. 1 at reference number 16. The distribution of the sheet metal parts Bi, B2 to the production machines 14 represents a highly complex problem with the various boundary conditions 10 of the sheet metal parts Bi, B2. This is particularly true because the individual process steps take different lengths of time, production machines 14 fail and / or urgent orders can be received .

The optimization of the production process according to the invention is shown in FIG. 2 Darge. FIG. 2 shows a device 18 for the optimized production or optimized production _{planning of the sheet metal parts Bi, B 2} from FIG. 1. An algorithm 20 is provided for this purpose. The algorithm 20 is preferably in the form of AlphaGo or AlphaGo Zero. The algorithm 20 comprises a Monte Carlo tree search framework 22. The Monte Carlo tree search framework 22 is modified by a neural network 24. First of all, supervised learning is carried out, i.e. training based on heuristically determined problem solutions.

This is followed by self-play with reinforcement learning as a single-player game. This is shown in FIG. 2 in steps 26 (selection), 28 (expansion), 30 (simulation) and 32 (backpropagation). In step 26, a decision path is selected via certain decision nodes, in step 28 the decision tree with the decision nodes is expanded according to the random principle, the result of this is simulated in step 30 and the decision nodes are reweighted on the basis of this simulation result in step 32 (strengthened or weakened). Steps 26 to 32 are repeated several times. The determination of the optimal division of the production steps carried out in this way preferably takes place both with regard to minimizing waste (nesting) and with regard to production time optimization (scheduling). This process can be described as optimization by a nesting agent and a scheduling agent, in which the agents make decisions in a simulation environment and receive a reward depending on the quality of the decision. The simulation is an image of the sheet metal production. The optimized production plan is output via a production plan interface 34, in particular to a manufacturing execution system 36. The manufacturing execution system 36 controls the production machines 14, that is to say the real sheet metal production, with the optimized production plan.

The algorithm 20 is supplied with the boundary conditions 10 via a boundary condition interface 38. User evaluations 40 can be fed to the algorithm 20 via a user evaluation interface 42. As an alternative or in addition to this, an event interface 44 can be provided, via which an event 46 can be read. The event 46 can be triggered by the manufacturing-execution-system 36, one or more production machines 14 and / or an indoor localization system 48. The event 46 can include, for example, a failure of a production machine 14, capacity of a production machine 14 that is freed up, errors in production, new orders and / or order changes. In particular, the event 46 includes the further production planning for a sheet metal part Bi, B ₂ (see FIG. 1) that has just completed a production step in a production machine 14. The algorithm 20 is executed on a computer 50. The computer 50 can be cloud-based in order to facilitate the use of user reviews 40 from different users. The manufacturing execution system 36 can be executed (as indicated) on the same computer or on a different computer.

Taking all the figures of the drawing together, the invention relates in summary to a method for optimizing the production of sheet metal parts Bi, B ₂ . The method optimizes the allocation of sheet metal parts Bi, B ₂ for processing on different production machines 14 and outputs an optimized production plan. For this purpose, an algorithm 20 is provided which has a decision tree in the form of a Monte Carlo tree search framework 22 and a neural network 24. The algorithm 20 is used with each new query trained through self-play and reinforcement learning. A pre-training of the algorithm 20 is achieved through supervised learning. The algorithm 20 preferably optimizes the production plan primarily with regard to minimally delayed production periods Foi to Fio of the sheet metal parts Bi, B2 and secondarily with regard to a minimal waste. By assigning scores, both goals can be assessed together. The method can include the receipt of query-triggering events 46 and / or the operation of production machines 14 in accordance with the production plan. The invention also relates to a device 18 for carrying out the method.

Reference list Aoi to Aio orders

Poi to Pio products Bi, B ₂ sheet metal parts

Foi to Fio production deadlines

Ci, c ₂ cutting production machines bi, b ₂ bending production machines ai, a ₂ assembly production machines 10 boundary conditions

12 sheet metal

14 production machines

16 products

18 device 20 algorithm

22 Monte Carlo tree search framework

24 neural network

26 step - selection

28 step expansion 30 step simulation

32 step - backpropagation

34 Production plan interface

36 manufacturing execution system

38 Constraint Interface 40 User Reviews

42 User Rating Interface

44 Event Interface

46 event

48 indoor localization system 50 computers

Claims

1. Method for optimizing the production of sheet metal parts (Bi, B ₂ ) with the

Process steps: a) Cutting out and separating the sheet metal parts (Bi, B ₂ ); b) bending the sheet metal parts (Bi, B ₂ ); wherein the method has the following method steps:

A) training of a neural network (24) executed on a Monte Carlo tree search framework (22) by means of supervised learning and self-play with reinforcement learning;

B) detecting boundary conditions (10) of the sheet metal parts (Bi, B ₂ ), the boundary conditions (10) including geometric data of the sheet metal parts (Bi, B ₂ );

C) creating an optimized production plan by the neural network (24);

D) Output of the production plan.

2. The method according to claim 1, wherein the method has one or more of the following process steps: c) deburring the sheet metal parts (Bi, B ₂ ); d) joining the sheet metal parts (Bi, B ₂ ); e) coating the sheet metal parts (Bi, B ₂ ); f) Assemble the sheet metal parts (Bi, B ₂ ).

3. The method according to claim 1 or 2, wherein the method steps A) to D) is carried out with an algorithm (20), the algorithm (20) based on AlphaGo or AlphaGo Zero and wherein the algorithm (20) comprises the neural network .

4. The method according to any one of the preceding claims, in which the training in method step A) is carried out with heuristically determined solutions of optimized production plans.

5. The method according to claim 4, in which optimized production plans in the form of earliest due date solutions are used.

6. The method according to any one of the preceding claims, in which the optimization comprises both the minimization of waste and the optimization of production time.

7. The method according to claim 6, in which the boundary conditions (10) in procedural step B) additionally include the production periods of the sheet metal parts (Bi, B ₂ ).

8. The method according to claim 7, wherein the boundary conditions (10) in procedural step B) additionally include the values of the sheet metal parts (Bi, B ₂ ).

9. The method according to claim 8, in which a waste score is allocated to the waste and a production deadline score is allocated to the achievement of the production deadline, which is based on the value of the sheet metal parts (Bi, B ₂ ), the optimization of both the waste score and the product onsfristscore minimized.

10. The method according to any one of the preceding claims, in which the procedural steps B) to D) are carried out event-triggered, the event (46) being read in via an event interface (44).

11. The method according to claim 10, wherein the event (46) is in the form of a request for further processing of a sheet metal part (Bi, B ₂ ), in the form of freed up production machine capacity, in the form of a production machine failure and / or in the form of a rush order .

12. The method as claimed in claim 10 or 11, in which the event (46) is triggered by a production machine (14), an indoor localization system (48) and / or a manufacturing execution system (36) and is read in via the event interface (44) .

13. The method according to any one of the preceding claims, in which in a method step E) a user evaluation (40) of the production plan output in method step D) is read and the neural network (24) is further trained with the user evaluation (40).

14. Device (18) for performing a method according to one of the preceding claims, wherein the device (18) has a computer (50) for storing and executing the neural network (24), a boundary condition section (38) for reading in the boundary conditions (10) and a production plan interface (34) for outputting the production plan.

15. The device according to claim 14 in conjunction with claim 12, wherein the device (18) has the event interface (44) and the device (18) further comprises a production machine (14), an indoor localization system (48) and / or a manufacturing execution system (36), an event (46) triggered by the production machine (14), the indoor localization system (48) and / or the manufacturing execution system (36) via the event interface (44 ) can be read.