Apparatus, method and computer program product for controlling a winder, and winder
By using neural network control devices in the winder, receiving multiple state signals and optimizing control through AI strategies and reinforcement learning, the problem of difficulty in achieving smooth material flow when the processing speed of the winder changes rapidly, improving energy efficiency and reducing stress.
Patent Information
- Application Number
- CN202411814324.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2023-12-14
- Filing Date
- 2024-12-11
- Publication Date
- 2025-06-17
AI Technical Summary
In existing winding machines, it is difficult to achieve smooth material flow and reduce stress on mechanical components and materials when handling winding materials, especially when processing speeds change rapidly.
The computer-implemented device adopts at least one neural network, receives multiple state signals to control the storage device and the buffering system, and achieves smoother control through AI strategies and reinforcement learning optimization control algorithms.
It improves the energy efficiency of the winder, reduces stress on mechanical parts and materials, and enhances control capabilities under rapidly changing conditions.
Smart Images
Figure CN120163186A_ABST
Abstract
Description
Field of the Invention
[0001] The present invention relates to a computer-implemented apparatus for controlling a winding machine for winding a wound material onto a target device, the winding machine including a storage device for storing the wound material and a buffer system for buffering the wound material between the storage device and the target device. Furthermore, the present invention relates to a system including a winding machine for winding a wound material onto a target device and a computer-implemented apparatus for controlling the winding machine. Furthermore, the present invention relates to a computer-implemented method and a computer program product for controlling a winding machine for winding a wound material onto a target device. Background Art
[0002] For example, in a winding / unwinding application, the material stored on a roll, i.e., the wound material, is wound / unwound onto a target device and fed into / consumed from a discontinuous process. Examples of wound materials include plastic films, metal foils, packaging labels, paper, battery separator films, etc.
[0003] To avoid having to accelerate and decelerate the entire material roll for each process step, regulating devices with smaller rollers are used, on which the wound material runs. Some rolls of such regulating devices are freely movable, which allows the continuous rolling / unrolling of the wound material to be coupled with the discontinuous process. If the regulating device has motors attached to some of the rollers, it is called an active regulating device. If the rollers are attached to, for example, springs to provide tension, it is called a passive regulating device. The amount of wound material that can be fed into the regulating device is limited, and thus for a film roll, some form of control is necessary to account for sudden changes in the processing speed or a previous processing speed, as well as for quickly stopping and restarting the winding machine. Without any additional control, the tension of the material may be too low, causing problems in the process, or increased to the extent of deforming or damaging the material. In both cases, the winding machine must be stopped and the material must be corrected by an operator. In many cases, this control is further complicated because the diameter of the material on the roll is unknown or measured with a large uncertainty, for example, by an ultrasonic sensor.
[0004] There are several conventional solutions using general control theory and motion control techniques to solve this problem. An example of wide application is SINAMICS DCC (available at https: / / support.industry.siemens.com / cs / document / 38043750 / ). This document contains solutions for basic unwinding and winding applications, but also includes a section on unwinding specifically via regulating devices. As with many other applications from other suppliers, a series of PID controllers are used to control the amount of material unwound. Adjusting these PID controllers requires in-depth knowledge of control theory and machine parameters. For many customers, this knowledge is concentrated in a single machine operator who adjusts these parameters by feel rather than by rules and trial and error. The uncertainty of the diameter complicates the search for optimized parameters. These conventional solutions are also limited to directly controlling the unwinding relative to the regulating position. Other process values, such as the machine speed of the previous machine or the material inflow, can usually be disregarded. Summary of the Invention
[0005] Accordingly, an object of the present invention is to enhance the control of a winding machine.
[0006] According to a first aspect, there is proposed a computer-implemented device for controlling a winding machine for winding a winding material onto a target device, the winding machine including a storage device for storing the winding material and a buffer system for buffering the winding material between the storage device and the target device. The computer-implemented device includes: a receiving unit for receiving N state signals, each of the N state signals including a specific indication of the current process state of the material flow of the winding material, where N≥1, a computing unit using at least one neural network, the at least one neural network being configured to provide a plurality of output signals for controlling the storage device and / or the buffer system using the received N state signals as inputs, and a control unit for controlling the storage device and / or the buffer system using the provided output signals.
[0007] The present computer-implemented device using at least one neural network can generate smoother control that does not depend on simply starting or stopping the storage device and / or the buffer system. This improves the energy efficiency of the winding machine and reduces the magnitude of stress on the mechanical components of the winding machine. In addition, it reduces the magnitude of stress on the winding material, since starting and stopping can have the highest values of acceleration, which imposes the highest magnitude of stress on the winding material, which is avoided here.
[0008] The neural network used by the computing unit implements an algorithm for an AI (Artificial Intelligence) strategy for a computer-implemented device, which algorithm can be set to also include various machine states or signals from the front / back winders. This enables the neural network and thus the computing unit to have more control in situations of rapidly changing conditions with rapidly changing processing speeds.
[0009] When operating the winder, at least one neural network of the computing unit is a trained neural network. For ease of training, it is beneficial to transform the input signal (i.e., the received status signal) to encode additional information or limit the negative impact of the signal, such as discontinuities. For example, for the estimated amount of wound material in a buffer system, it may be helpful to pass only the maximum or minimum value of the last machine cycle instead of the oscillating present value or to apply a suitable filter to the signal. Additionally, for example, a binary signal can be encoded as a sawtooth function.
[0010] Here the term "winding" refers to winding applications and unwinding applications. The neural network can also be referred to as an artificial neural network.
[0011] According to an embodiment, the receiving unit is configured to receive a plurality, i.e., N, status signals, which include:
[0012] - The current reel speed of the reel of the storage device that stores the wound material,
[0013] - The current speed of the motor of the storage device,
[0014] - The current speed of the motor of the buffer system,
[0015] - The current amount of the wound material stored on the reel of the storage device,
[0016] - The current amount of the wound material buffered in the buffer system,
[0017] - A plurality of light bridge signals of a light bridge arranged at a feed belt for providing an incoming target device,
[0018] - A plurality of position signals of a light bridge arranged at a feed belt for providing an incoming target device,
[0019] - The current speed of the feed belt,
[0020] - A plurality of light bridge signals of a light bridge arranged at a discharge belt for providing an output target device, and / or
[0021] - The current speed of the discharge belt,
[0022] - A processing signal indicating the current state of the processing process.
[0023] In an embodiment, additional status signals can be used, which are configured to describe the current state of a system including a winding machine, a buffer system, and a storage device. In an application, a subgroup of the status signals can also be used.
[0024] According to another embodiment, the output signal provided by the neural network includes a first setpoint for a first motor controller of a motor of the storage device and / or a second setpoint for a second motor controller of a motor of the buffer system.
[0025] In an application, a system including a winding machine can have multiple different motors with different functions. In an embodiment, for each of the motors of the winding machine, if the neural network of the computing unit is trained accordingly, the neural network of the computing unit can provide a suitable output signal.
[0026] According to another embodiment, the neural network is configured to use reinforcement learning and receive as additional inputs a reward for a smooth movement of the wound material, a penalty for a movement of the wound material with an acceleration above an upper threshold value, and / or a penalty for a violation of a boundary. In particular, the smooth movement is defined by an upper threshold value of the acceleration of the wound material in the winding machine.
[0027] Using reinforcement learning, at least one neural network can be trained now. When controlling the winding machine, reinforcement learning can use rewards and penalties to optimize the function of a computer-implemented device. In particular, if the movement of the wound material performed by the winding machine has an acceleration below the upper threshold value, the movement is a smooth movement. In another case, if the movement of the wound material in the winding machine 200 has an acceleration equal to or greater than the acceleration upper threshold value, the movement is not a smooth movement, and the neural network may receive a penalty for violating the upper threshold value. In addition, for reinforcement learning, the neural network can receive a penalty for an action that violates any predefined boundary.
[0028] According to another embodiment, the computer-implemented device is configured to control the winding machine during a discontinuous process according to a discontinuous winding process, in which the winding machine winds a plurality of target devices using the wound material. The discontinuous process of the computer-implemented device can correspond to the discontinuous winding process of the winding machine.
[0029] According to another embodiment, the proximal policy optimization (PPO) algorithm is used to train the neural network. In the PPO algorithm, a first neural network and a second neural network can be trained, where the first neural network is configured to provide actions fed into the winding machine, and the second neural network is trained to estimate the quality of these actions, in other words, how good these actions are.
[0030] Specifically, the PPO algorithm runs on a simulation of the winding machine. Thus, the PPO algorithm does not run directly on the winding machine, but on a simulation of the winding machine. This is beneficial because reinforcement learning algorithms start with random actions that could cause damage or problems to the winding machine. By using a simulation, the simulation can simply be reset after a series of destructive actions, and a negative reward can be returned to the algorithm, enabling the algorithm to learn from its mistakes.
[0031] According to another embodiment, the PPO algorithm runs on a simulation of the winding machine.
[0032] According to another embodiment, at least one neural network of the computing unit has a multi-layer perceptron (MLP) structure.
[0033] According to another embodiment, the at least one neural network is a long short-term memory (LSTM) neural network.
[0034] According to another embodiment, the at least one neural network is a recurrent neural network (RNN).
[0035] In an embodiment, at least one neural network can be trained with a new policy. This can be done very quickly, while adapting traditional methods usually requires a very deep understanding of control theory and the application at hand. This makes it difficult to adapt existing solutions to new regulating devices, such as five rolls instead of three rolls, different positions of the rolls, or connecting some of the rolls in the regulating device to each other. With an AI-based method using at least one neural network, these parameters can be easily adjusted, and the policy can be retrained within just a few days. This can be achieved through cloud computing that provides an easy-to-use interface for machine engineers.
[0036] The algorithm implemented by at least one neural network can also be used for design optimization by running different trainings using changes in machine configuration (such as the number of rolls in the buffer, the size of the material roll, the maximum acceleration and deceleration of the reel, the number and quality of sensors).
[0037] Corresponding units, such as computing units, can be implemented in hardware and / or software. If the unit is implemented in hardware, it can be implemented as a device, such as a computer or as a processor or as part of a system, such as a computer system. If the unit is implemented in software, it can be implemented as a computer program product, a function, a routine, program code, or an executable object.
[0038] Here and hereinafter, a neural network or an artificial neural network can be understood as software code stored on a computer-readable storage medium and representing one or more interconnected artificial neurons or capable of simulating their functions. The software code can also include several software code components, which can, for example, have different functions. In particular, an artificial neural network can implement a non-linear model or a non-linear algorithm that maps an input (here a state signal) to an output (here an output signal). The input can be given by an input feature vector or an input sequence, and the output can include, for example, the output classes of a classification task, one or more determined values, such as a setpoint for a motor controller, or a prediction sequence.
[0039] Any embodiment of the first aspect can be combined with any embodiment of the first aspect to obtain another embodiment of the first aspect.
[0040] According to a second aspect, a system is proposed, which includes a winding machine for winding a winding material onto a target device. The winding machine includes a storage device for storing the winding material and a buffer system for buffering the winding material between the storage device and the target device. The system also includes a computer-implemented device for controlling the winding machine according to the first aspect or any embodiment of the first aspect.
[0041] In an embodiment, the buffer system is implemented as an adjusting device including a plurality of adjusting rollers.
[0042] In an embodiment, at least one of the plurality of adjusting rollers is an active adjusting roller driven by an adjusting motor.
[0043] According to a third aspect, a computer-implemented method for controlling a winding machine for winding a winding material onto a target device is proposed. The winding machine includes a storage device for storing the winding material and a buffer system for buffering the winding material between the storage device and the target device. The computer-implemented method includes:
[0044] Receiving N state signals, each of the N state signals including a specific indication of the current process state of the material flow of the winding material, where N ≥ 1,
[0045] Feeding the received N state signals into at least one neural network for providing a plurality of output signals to control the storage device and / or the buffer system, and
[0046] Using the provided output signals to control the storage device and / or the buffer system.
[0047] In particular, the computer-implemented method is a method using a computer, a computer network, or another programmable device, where one or more features are implemented in whole or in part by a computer program.
[0048] The technical effects and advantages described for the computer-implemented apparatus according to the first aspect are equally applicable to the computer-implemented method according to the third aspect.
[0049] According to a fourth aspect, a computer program product is proposed, which includes program code for performing the computer-implemented method according to the third aspect when running on at least one computer.
[0050] The computer program product, such as a computer program device, can be implemented as a memory card, a USB stick, a CD-ROM, a DVD, or a file that can be downloaded from a server in a network. Such a file can be provided, for example, by transmitting a file including the computer program product from a wireless communication network.
[0051] Other possible embodiments or alternatives of the present invention also cover combinations of features (not explicitly mentioned herein) described above or below with respect to the embodiments. Those skilled in the art can also add individual or isolated aspects and features to the most basic form of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS
[0052] Other embodiments, features, and advantages of the present invention will become apparent from the following description in conjunction with the accompanying drawings and the dependent claims, in which:
[0053] Figure 1 A schematic block diagram showing an embodiment of a computer-implemented apparatus for controlling a winding machine is shown;
[0054] Figure 2 A schematic block diagram showing an embodiment of a system including a winding machine and a computer-implemented apparatus for controlling the winding machine is shown; and
[0055] Figure 3 A schematic flowchart showing an embodiment of a computer-implemented method for controlling a winding machine is shown. DETAILED DESCRIPTION
[0056] In the drawings, unless otherwise specified, the same reference numerals denote the same or functionally equivalent elements.
[0057] In Figure 1 a schematic block diagram showing an embodiment of a computer-implemented apparatus 100 for controlling a winding machine 200 is shown.
[0058] Referring to Figure 2 discussed Figure 1 the computer-implemented apparatus 100, Figure 2 a schematic block diagram showing an embodiment of a system 10 including a winding machine 200 and Figure 1 the computer-implemented apparatus 100 is shown. It can be noted that Figure 1shows the details of a computer-implemented device 100, where Figure 2 the same computer-implemented device 100 is shown in an abstract manner to ensure Figure 2 readability.
[0059] Figure 2 System 10 includes a winding machine 200 for winding a winding material M onto a target device T5. As Figure 2 shown, system 10 includes a feed belt 230 for supplying input target devices T1 - T4 and a discharge belt 240 for supplying output target devices T6 - T9. As Figure 2 shown, there are nine target devices T1 - T9, where target device T5 is being processed by the winding machine 200, target devices T1 - T4 are input target devices supplied by the feed belt 230, and target devices T6 - T9 have been processed by the winding machine 200 and are thus on the discharge belt 240.
[0060] The winding machine 200 has a storage device 210 for storing the winding material M and a buffer system 220 for buffering the winding material M between the storage device 210 and the object device T5 being processed. The storage device 210 includes a roll for storing the winding material M. For example, the winding material M is a plastic film, a metal sheet, paper, or a battery separator. The winding machine 200 wraps the winding material M onto the target devices T1 - T9 provided by the feed belt 230 in a discontinuous process. The buffer system 220 is configured to buffer the winding material M between the storage device 210 and the target device T5 being processed.
[0061] For example, the buffer system 220 is implemented as an adjustment device including a plurality of adjustment rollers. In Figure 2 the example, the adjustment device 220 includes seven adjustment rollers. For readability, only one adjustment roller in Figure 2 is provided with the reference numeral 221. In addition, at least one of the plurality of adjustment rollers of the adjustment device 220 can be implemented as an active adjustment roller driven by an adjustment motor. For example, the adjustment roller with the reference numeral 221 is an active adjustment roller driven by an adjustment motor and can be controlled by a motor controller.
[0062] Returning to the computer-implemented device 100 configured to control the winding machine 200 as detailed in Figure 2 and Figure 1 above. As Figure 1 highlighted, the computer-implemented device 100 includes a receiving unit 110, a computing unit 120, and a control unit 130. The receiving unit 110 is configured to receive N status signals S1 to S11. In Figure 1 and Figure 2In an embodiment, the receiving unit 110 receives eleven status signals S1 - S11 without loss of generality. Generally, N is equal to or greater than 1 (N ≥ 1). Each of the N status signals S1 to S11 includes an indication or information about the current processing status of the material flow of the winding material M in the winder 200.
[0063] For example, the multiple, i.e., N, status signals S1 - S11 include:
[0064] - The current reel speed S1 of the reel of the storage device 210 that stores the winding material M,
[0065] - The current speed S2 of the motor of the storage device 210,
[0066] - The current speed S3 of the motor of the buffer system 220,
[0067] - The current amount S4 of the winding material M stored on the reel of the storage device 210,
[0068] - The current amount S5 of the winding material M buffered in the buffer system 220,
[0069] - Multiple light bridge signals S6 of the light bridge arranged at the feed belt 230 for providing access to the target devices T1 - T4,
[0070] - Multiple position signals S7 of the light bridge arranged at the feed belt 230 for providing access to the target devices T1 - T4,
[0071] - The current speed S8 of the feed belt 230,
[0072] - Multiple light bridge signals S9 of the light bridge arranged at the discharge belt 240 for providing access to the output target devices T6 - T9, and / or
[0073] - The current speed S10 of the discharge belt, and / or
[0074] - A processing signal S11 indicating the current status of the processing process.
[0075] The calculation unit 120 uses at least one neural network 121. Without loss of generality, Figure 1 the calculation unit 120 of uses one neural network 121. For example, the neural network 121 of the calculation unit 120 has a multi - layer perceptron (MLP) structure. In an embodiment, the neural network 121 can be implemented as a long short - term memory (LSTM) neural network or a recurrent neural network (RNN).
[0076] The neural network 121 is configured to use the received N state signals S1 - S11 as inputs to provide multiple output signals O1, O2 for controlling the storage device 210 and / or the buffer system 220.
[0077] As Figure 1 shown, the neural network 121 includes an input layer 122, multiple hidden layers 123, and an output layer 124. The input layer 122 of the neural network 121 receives the provided state signals S1 - S11 from the receiving unit 110, processes these input state signals S1 - S11 by means of the multiple hidden layers 123, and provides the output signals O1, O2 to the control unit 130 via its output layer 124.
[0078] The control unit 130 is configured to use the provided output signals O1, O2 to control the storage device 210 and / or the buffer system 220.
[0079] In an embodiment, the output signals O1, O2 provided by the neural network 121 are adapted to control both the storage device 210 and the buffer system 220. In particular, for the case where the buffer system 220 discussed above is implemented as an adjustment device including at least one actively adjustable roller, the neural network 121 can provide the output signals O1, O2 for controlling both the storage device 210 and the adjustment device.
[0080] For example, the output signals O1, O2 provided by the neural network 121 include a first setpoint O1 of a first motor controller for at least one motor of the storage device 210 and a second setpoint O2 of a second motor controller for at least one motor of the buffer system 220. In Figure 2 the example, the output signal O1 is adapted to control the first motor controller for at least one motor of the storage device 210, and the output signal O2 is adapted to control the second motor controller for at least one adjustment motor of the adjustment roller 221.
[0081] Furthermore, in an embodiment, the neural network 121 is configured to use reinforcement learning. In this case, the neural network 121 is adapted to receive a reward for the smooth movement of the wound material M as an additional input, which is defined by an upper limit threshold of the acceleration of the wound material M in the winding machine 200. In other words, if the movement of the wound material M performed by the winding machine 200 has an acceleration lower than the upper limit threshold, then the movement is a smooth movement. In another case, if the movement of the wound material M in the winding machine 200 has an acceleration equal to or greater than the acceleration upper limit threshold, then the movement is not a smooth movement, and the neural network 121 may receive a penalty for violating the upper limit threshold. In addition, for reinforcement learning, the neural network 121 can receive a penalty for actions that violate any predefined boundaries.
[0082] As described above, the winding machine 200 particularly wraps the plurality of target devices T1 - T9 with the winding material M during a discontinuous winding process. In this regard, the computer - implemented device 100 is particularly configured to also control the winding machine 200 during a discontinuous process. The discontinuous process of the computer - implemented device 100 may correspond to the discontinuous winding process of the winding machine 200.
[0083] Before operation, the neural network 121 is trained. The proximal policy optimization (PPO) algorithm can be used for training. In the PPO algorithm, a first neural network and a second neural network can be trained. The first neural network is configured to provide actions fed into the winding machine 200, and the second neural network can be trained to estimate the quality of these actions, in other words, how good these actions are.
[0084] Particularly, the PPO algorithm runs on a simulation of the winding machine 200. In other words, the PPO algorithm does not run directly on the winding machine 200, but on a simulation of the winding machine 200. This is beneficial because reinforcement learning algorithms start with random actions that may cause damage or problems to the winding machine 200. By using the simulation, the simulation can simply be reset after a series of destructive actions, and a negative reward can be returned to the algorithm, enabling the algorithm to learn from its mistakes.
[0085] In addition Figure 3 A schematic flowchart of an embodiment of a computer - implemented method for controlling the winding machine 200 is shown. Figure 2 An example of such a winding machine 200 is depicted. The winding machine 200 is configured to wind the winding material M onto the target device T5. The winding machine 200 includes a storage device 210 for storing the winding material M and a buffer system 220 for buffering the winding material M between the storage device 210 and the target device T5.
[0086] Figure 3 The computer - implemented method includes method steps 301 - 303:
[0087] In step 301, a plurality of state signals S1 - S11 are received. Each of the N state signals S1 - S11 includes a specific indication of the current processing state of the material flow of the winding material M.
[0088] In step 302, the received N state signals S1 - S11 are fed into at least one neural network 121 to provide a plurality of output signals O1, O2 for controlling the storage device 210 and / or the buffer system 220.
[0089] In step 303, the provided output signals O1, O2 are used to control the storage device 210 and / or the buffer system 220. It can be noted that Figure 1 andFigure 2 The computer-implemented device 100 may be adapted to perform the above method steps 301-303.
[0090] Although the present invention has been described in accordance with the preferred embodiments, it will be apparent to those skilled in the art that modifications can be made in all embodiments.
[0091] List of reference numerals:
[0092] 10 System
[0093] 100 Computer-implemented device
[0094] 110 Receiving unit
[0095] 120 Computing unit
[0096] 121 Neural network
[0097] 122 Input layer
[0098] 123 Hidden layer
[0099] 124 Output layer
[0100] 130 Control unit
[0101] 200 Winding machine
[0102] 210 Storage device
[0103] 220 Buffer system
[0104] 221 Adjusting roller
[0105] 230 Feeding belt
[0106] 240 Discharging belt
[0107] 301-303 Method steps
[0108] M Winding material
[0109] O1-O2 Output signal
[0110] S1-S11 Status signal
[0111] T1-T9 Target device
Claims
1. A computer-implemented device (100) for controlling a winding machine (200) for winding a winding material (M) onto a target device (T5), the winding machine (200) comprising a storage device (210) for storing the winding material (M) and a buffer system (220) for buffering the winding material (M) between the storage device (210) and the target device (T5), the computer-implemented device (100) comprising: A receiving unit (110) is used to receive a number N of status signals (S1-S11), each of the N status signals (S1-S11) comprising a specific indication of a current process state of the material flow of the winding material, a computing unit (120) using at least one neural network (121), the at least one neural network (121) being configured to use the received N state signals (S1-S11) as input to provide a plurality of output signals (O1, O2) for controlling the storage device (210) and / or the buffer system (220), and A control unit (130) is used to control the storage device (210) and / or the buffer system (220) using the provided output signals (O1, O2).
2. The device according to claim 1, in, The receiving unit (110) is configured to receive a plurality of, i.e., N, state signals (S1-S11), wherein the plurality of state signals include: - the current reel speed (S1) of a roll of the storage device (210), the roll storing the winding material (M), - the current speed (S2) of the motor of the storage device (210), - the current speed (S3) of the motor of the buffer system (220), - the current amount (S4) of the winding material (M) stored on the roll of the storage device (210), - the current amount (S5) of the winding material (M) buffered in the buffer system (220), - a plurality of light bridge signals (S6) of a light bridge arranged at a feed belt (230) for providing incoming target devices (T1-T4), - a plurality of position signals (S7) of a light bridge arranged at a feed belt (230) for providing incoming target devices (T1-T4), - the current speed (S8) of the feed belt (230), - a plurality of optical bridge signals (S9) of an optical bridge arranged at an outfeed belt (240) for providing output target devices (T6-T9), and / or - the current speed of the outfeed belt (S10), and / or - a processing signal (S11) indicating the current status of the processing progress.
3. The device according to claim 1 or 2, in, The output signals (O1, O2) provided by the neural network (121) include a first set point (O1) of a first motor controller for a motor of the storage device (210) and a second set point (O2) of a second motor controller for a motor of the buffer system (220).
4. The device according to any one of claims 1 to 3, in, The neural network (121) is configured to use reinforcement learning and receives as additional input a reward for smooth movement of the winding material (M), a penalty for movement of the winding material (M) with an acceleration above an upper threshold and / or a penalty for boundary violation, wherein the smooth movement is limited by an upper threshold for the acceleration of the winding material (M) in the winding machine (200).
5. The device according to any one of claims 1 to 4, in, The computer-implemented device (100) is configured to control a winder (200) in a discontinuous process according to a discontinuous winding process, in which the winder (200) wraps a plurality of target devices (T1-T9) with a winding material (M).
6. The device according to any one of claims 1 to 5, in, The neural network (121) is trained using a proximal policy optimization (PPO) algorithm, wherein in the PPO algorithm, a first neural network and a second neural network are trained, wherein the first neural network is configured to provide actions fed into the winder (200) and the second neural network is trained to estimate the quality of these actions.
7. The device according to claim 6, in, The PPO algorithm was run on a simulation of the winder (200).
8. The device according to any one of claims 1 to 7, in, The at least one neural network (121) of the computing unit (120) has a multi-layer perception (MLP) structure or a long short-term memory (LSTM) neural network or a recurrent neural network (RNN).
9. A system (10) comprising a winding machine (200) for winding a winding material (M) onto a target device (T5), the winding machine (200) comprising a storage device (210) for storing the winding material (M) and a buffer system (220) for buffering the winding material (M) between the storage device (210) and the target device (T5), the system also comprising a computer-implemented device (100) for controlling the winding machine (200) according to any one of claims 1 to 8.
10. The system according to claim 9, in, The buffer system (220) is implemented as an adjustment device including a plurality of adjustment rollers (221).
11. The system according to claim 10, in, At least one of the plurality of adjusting rollers (221) is an active adjusting roller driven by an adjusting motor.
12. A computer-implemented method for controlling a winding machine (200) for winding a winding material (M) onto a target device (T5), the winding machine (200) comprising a storage device (210) for storing the winding material (M) and a buffer system (220) for buffering the winding material (M) between the storage device (210) and the target device (T5), the computer-implemented method comprising: receiving (301) a number N of status signals (S1-S11), each of the N status signals (S1-S11) comprising a specific indication of a current process state of the material flow of the winding material, feeding (302) the received N state signals S1-S11 into at least one neural network (121) for providing a plurality of output signals (O1, O2) for controlling a storage device (210) and / or a buffer system (220), The storage device (210) and / or the buffer system (220) are controlled (303) using the provided output signals (O1, O2).
13. A computer program product comprising program code for performing the computer-implemented method according to claim 12 when run on at least one computer.