Methods for controlling the flow of goods to be manufactured through an industrial production environment

Abstract

A method for controlling the flow of goods to be manufactured (10a, 10b) through an industrial production environment is described, wherein a large number of identical goods (10a, 10b) are grouped into a production order (Fa, Fb) and pass through several production stations (9a to 9d), and wherein the production environment includes a localization system (1) for capturing position data of the goods (10a, 10b) in production, the method comprising: - holding an initial number of production orders (Fa, Fb) in an order buffer (9), - locating the goods (10a, 10b) in production within the production environment, - analyzing the current utilization of the production stations (9a to 9d) of the production environment based on the position data and analyzing the production orders (Fa, Fb) in the order buffer (9) using a machine learning model.- Prioritizing the production orders (Fa, Fb) located in the order buffer (9) by the machine learning model based on the analysis, and - processing the production orders (Fa, Fb) to produce the goods !0a, 10b) according to the prioritization.

Description

The invention relates to a method for controlling a passage of goods to be manufactured through an industrial production environment by means of a machine learning model which is combined with a localization system for localization of the goods being manufactured.In order to be able to produce industrial goods economically, it is often decisive that the production sequences are designed optimally. In particular in the production of integrated circuits, for example in the assembly of printed circuit boards, a plurality of production steps are necessary at different production stations. In this case, it can be provided that different configurations of the placement machines or different setup trays must be prepared for different production orders. Furthermore, individual manufacturing stations can have free capacities, while other manufacturing stations are completely used and thereby limit the total flow rate of goods to be produced. Lying times of the goods to be produced which are caused in this way also increase the cycle time of the goods of the relevant production orders. By grouping similar manufacturing orders that require identical or similar components for producing the goods, the throughput can be increased. This is typically done manually on the basis of up-to-date production data and thus generally not optimally.It is an object of the invention to provide a method for controlling the passage of goods to be produced through an industrial production environment, with the aid of which production sequences in the production and processing of commercial goods can be optimized, short resting times of the goods and a large passage can be achieved.The object is achieved by a method according to claim 1. Advantageous embodiments of the invention are specified in the dependent claims. To achieve the object, there is further provided a system for controlling the passage of goods to be manufactured through an industrial production environment according to claim 13.In the context of this patent application, an "RTLS anchor" denotes a stationary sensor unit for determining the position of RTLS tags. An RTLS anchor is configured to receive radio signals from RTLS tags and to determine a distance of the respective RTLS tag to the RTLS anchor.Furthermore, an "RTLS tag" within the scope of this patent application denotes a sensor system element which is associated with one or more components or goods located in a production environment and is configured and configured to transmit a radio signal which can be received by RTLS anchors, such that a position of the RTLS tag and thus also of the associated components or goods can be calculated.Advantageously, a method for controlling the passage of goods to be produced through an industrial production environment is specified, wherein in each case a multiplicity of identical goods is combined in a production order and a plurality of production stations are passed through, and wherein the production environment comprises a localization system for capturing position data of the goods located in the production. The method comprising:providing a first number of production orders in an order buffer,locating the goods being produced in the production environment,analyzing the load of the manufacturing stations or the production environment on the basis of the position data and analyzing the manufacturing orders located in the order buffer by means of a machine learning model,prioritizing the manufacturing jobs located in the job buffer by the machine learning model on the basis of the analysis; andprocessing the production orders for producing the goods according to the prioritization.Thus, a method for controlling and automatically optimizing the flow of goods through the production environment may be provided.The load on the manufacturing stations or the production environment is intended to include in particular the transit time of the goods through the manufacturing stations, the transit time of the goods through all manufacturing stations of the production environment provided for production, the buffer dwell time, and / or the dwell time of the goods during the transition between two manufacturing stations. In this case, both the goods of individual production orders and an average value over a plurality of production orders or another viewing together can be considered.The machine learning model preferably comprises minimizing a filling degree of the job buffer and / or minimizing a buffer dwell time. Long waiting times of production orders can thus be avoided and the cycle time from the input of the production orders to the complete cycle through the production environment can be minimized.In some configurations, it is further preferred that the machine learning model comprises the minimization of resting times of the goods to be produced during the transition of the goods between successive production stations and / or the minimization of the total transit time of the goods to be produced through the production environment. The cycle time can thus be further reduced by avoiding bottleneckes during the successive cycle of the manufacturing stations.Preferably, the machine learning model includes a reinforcement learning model. The configuration as model-reinforcing learning has the advantage that the method can analyze the run times, buffer dwell times and / or other dwell times resulting from the prioritization of the production orders that has been carried out and can further optimize itself on the basis of the analysis.The models of machine learning, or of reinforcement learning, can be designed in particular as a computer-implemented method. For example, the models can be designed as an artificial neural network. The artificial neural network can be designed, in particular, to prioritize the production orders held in the order buffer or to output an optimized buffer list. In some configurations, the artificial neural network is therefore designed to minimize a cost function, as a result of which optimization of the production control towards a minimum cycle time of the goods can be achieved.If the artificial neural network is designed as model-reinforcing learning, a software agent can be implemented which is designed to optimize the utilization of the production environment, in particular the throughput times, buffer dwell times or other dwell times. The software agent can be designed to perform an action, in particular an optimization of the cycle time, in particular by making a prioritization of the production orders of the order buffer. As feedback, the software agent may receive the round trip times, buffer live times, and other live times resulting from the actions taken. Furthermore, a reward for the software agent can be calculated from this information. The software agent can then adjust its policy before performing a subsequent optimization step.In some embodiments of the method, it can furthermore be provided that the software agent performs a new optimization if one of the following events occurs: the omission or addition of a production order from / to the order buffer, the prioritizing of a production order by a user or a system unit, the maintenance or malfunction of a production station of the production environment, the actual or prospective exceeding of a maximum throughput time of a production order, the reaching of a time, the expiration of a predefined time interval, a change in an environmental factor, for example the reaching of a threshold temperature, the registering of a staff failure.In some refinements of the method, it may be provided that status information of the manufacturing stations is supplied to the model-enhancing learning via a process control network and is taken into account by the latter in the optimization of the run times, buffer dwell times and other dwell times.In some configurations of the method, the production of the goods can comprise at least two substeps, each of the substeps comprising an order buffer. The machine learning model is then preferably designed to minimize the average buffer laying time averaged over all order buffers and / or the transit time of the goods to be produced through all substeps.Additionally or alternatively, the machine learning model can then be designed to minimize the respective buffer laying time and / or the transit time of the goods to be produced through the respective substep. In some refinements, the method for this can comprise a model-reinforcing learning with a software agent for each of the order buffers, wherein the software agents are each designed to minimize the respective buffer dwelling time and / or the transit time of the goods to be produced through the respective substep.For example, a first buffer can be provided for one or more set-up stations. A second buffer can be provided, for example, for the passage through a component placement system and different post-processing stations. Thus, the passage of the goods to be produced can be optimally controlled for each of the substeps by means of the model-reinforcing learning.It is particularly advantageous if the model reinforcement learning is designed and set up for exchanging data between the at least two software agents. It can thus be provided that a first software agent optimizes the pass through the first substep and a second software agent optimizes the pass through the second substep, wherein both software agents interact. In particular, the interaction of the two software agents and the data exchange connected thereto can be configured such that the first software agent and / or the second software agent further optimize the pass through the respective substep on the basis of the data of the respective other software agent. Furthermore, all software agents can be designed such that the total flow rate is optimized taking all substeps into account. For example, a correspondingly designed, superordinate software agent can be provided for this purpose.The software agent can be designed, in particular, to interact with the localization system or the production environment. In particular, the software agent can analyze effects of the prioritization of the production orders output by it or by the artificial neural network and maximize rewards obtained therefrom. A self-learning model can thus be implemented for optimizing the run times, buffer dwell times and other dwell times.Advantageously, the model reinforcement learning is based on a model-free, a model-based or a strategy-based algorithm. The model can take account of the peculiarities and boundary conditions of the production sequences. In particular, a gradient approximation of the expected gain can be implemented. Combinations of different models are also possible. The reinforcement learning model may be configured to maximize the future expected gain. In the present case, this can be, in particular, a future minimum transit time, buffer dwelling time and / or other dwelling time of the goods.In such embodiments, in which a superordinate software agent is provided, this is preferably based on a strategy-based algorithm. The subordinate software agent or agents are preferably based on a model-based algorithm.Furthermore, run times, buffer dwell times, and / or other dwell times calculated by the machine learning model may be communicated to the localization system. Thus, a graphical representation of the load to be expected can be displayed within the localization system.In some preferred embodiments, the machine learning model can be initially trained on the basis of training data. The training data can comprise, for example, past priorities of the production orders and the resulting run times. Training the machine learning model with the training data can ensure that the method also carries out a meaningful prioritization during the first execution, so that even then short run times are already reached.In the context of the method, the following data can be provided in particular to the machine learning model: a job number of the relevant production job, the previous buffer life of the production job, a material or product number, a number of pieces to be produced. Furthermore, the data can comprise different time data, in particular: a sum of setup times, a sum of the scheduled productive time during production in the production stations, a sum of the required working time. In addition, further data, for example a prioritization carried out by a user, can be provided.As output, the machine learning model can output in particular one or more of the following data: a rank number for prioritizing the relevant production order, the order number of the production order, the material or product number, the number of pieces to be produced, the previous and / or the expected buffer storage time, the expected cycle time, the expected time of termination of the production process of all the products of the production order. In addition, the job buffer can be output in the form of a list. The list can be displayed on a display device or output to a process automation network for further process automation.In some embodiments, the localization system can be designed to optically capture the position data of the goods. In particular, it can be provided that the goods or objects carried with the goods are provided with optical signatures, such as bar codes or QR codes, which are detected within the framework of the method for localization. For this purpose, the manufacturing stations, including the waiting areas or buffer bed areas, can comprise corresponding detectors. The optical signatures may then be automated or scanned by a person. In particular, the optical signatures can comprise ID data of the detector, of the goods and / or of the production orders. Alternatively or additionally, the optical detection of the goods can also be carried out with the aid of a method for object detection.The ID data can comprise in particular data for identifying the goods and / or a production order associated with the goods. Furthermore, the ID data can comprise an identifier for identifying the detector. For example, the ID data may include a part number, a part type, a job number, a serial number or other identification number of the detector, an identifier of a manufacturing station, or similar information.Alternatively or additionally to the acquisition of the position data by means of optical systems, a localization system based on RFID (radio-frequency identification) can also be used. For example, the production orders, the goods or components used for production, and / or other production-relevant objects can be associated with an RFID tag. Accordingly, the manufacturing stations may comprise RFID detectors. After the detection of the respective RFID tags or the optical signatures, the detected positions and ID data may be supplied to the machine learning model.In some embodiments, the localization system includes an ERP (Enterprise Resource Planning) system. The location of the goods can then be effected at least partially via the ERP system. For example, bookings made in the ERP system, such as the setting of a manufacturing order, the successful processing of a manufacturing order, the availability of components or materials necessary for producing the goods, inventory, booking processes for working steps at the manufacturing stations, and similar information for locating the goods, can be used. For example, the goods can be automatically detected at one of the manufacturing stations during production or can be communicated to the system by a user. Thus, the ERP system can be made aware of the position of these goods within the production environment and can be supplied to the machine learning model.The localization system is preferably designed as a real-time localization system (RTLS for short). This has the advantage that events occurring during production, such as changes or malfunctions in the production sequence, can be recognized directly. Thus, the machine learning model can also directly calculate a recalculation of the prioritization of the manufacturing orders. A "real-time localization system" can be understood to mean, for example, a localization system in which the localization of the goods being produced takes place with a temporal resolution of up to 30 minutes. The localization is preferably carried out in real time with a temporal resolution in the range up to one minute. The localization can be carried out particularly advantageously in real time with a temporal resolution in the seconds range or milliseconds range.To locate the goods during the manufacturing process (including buffer residence time and other residence times), the real-time localization system may include one or more RTLS tags. The RTLS tags can be attached to a production-relevant object or carried along with it in order to carry out the method, such that the RTLS tags remain in the spatial vicinity of the goods to be produced during the entire production process. The method then preferably comprises the steps of:associating a manufacturing order for manufacturing a plurality of identical goods with an RTLS tag,attaching the RTLS tag to a manufacturing-relevant object, which is carried to each of the manufacturing stations during the manufacturing of the goods,locating the goods of a manufacturing order by means of the real-time locating system by locating the corresponding RTLS tag.The invention further relates to a system for controlling the passage of goods to be produced through an industrial production environment, comprising a system control unit and a localization system, wherein the localization system is connected to the system control unit or is formed integrally therewith for the exchange of data. The system control unit comprises a data memory and is designed and configured to execute a computer program stored on the data memory, which is configured to carry out a method according to the invention.The localization system is preferably designed as a real-time localization system having at least one RTLS control unit, at least one RTLS anchor and at least one RTLS tag. The RTLS tag is designed and configured to transmit ID data to the RTLS anchors by means of a radio signal, and the RTLS anchors are designed and configured to detect a distance of the RTLS tags to the respective RTLS anchor on the basis of the radio signal. The RTLS anchors may then send position data to the controller based on the ID data and the detected distance. The RTLS control unit can furthermore comprise a system control unit or be connected to it for the exchange of data.The system control unit can therefore execute a computer program for carrying out the method. The RTLS control unit can be configured separately and can likewise execute a corresponding computer program. For transmitting the position data to the system control unit, a data connection and a corresponding interface between the two control units can be provided. Alternatively, both computer programs may be executed by the system controller. It can thus be provided that both control units are integrated in the system control unit. Likewise, the computer programs can be realized as a single computer program.The real-time localization system can be designed in particular as a sensor network and can comprise an RTLS control unit, which has a first communication interface, at least one RTLS anchor, which has a second communication interface, and at least one RTLS tag, which has a third communication interface. The RTLS tag or tags can be configured and configured to send ID data via the third communication interface, wherein the second communication interface and the third communication interface are configured for wireless communication. The RTLS anchor or anchors may be configured and configured to receive the ID data of the RTLS tags via the second communication interface and to detect a distance of the RTLS tag from the RTLS anchor. The RTLS anchor capability is further configured to send position data, which comprises the detected distance and the ID data, via the second communication interface to the first communication interface of the RTLS control unit.Thus, RTLS tags carried with the goods to be manufactured or processed can be localized in real time. In particular, on the basis of the obtained information, an optimization of the production sequences can be carried out by means of the machine learning model.The first communication interface and the second communication interface can comprise, for example, at least one of the following interfaces: Ultra Wide Band (UWB), WLAN, Wifi, Bluetooth, Wide-Over-Narrow Band, Radio beacon, and Low-Frequency Signpost Identification.The real-time localization system is preferably designed to determine a relative position to at least one of the RTLS anchors. For example, a distance can be carried out on the basis of signal transit time measurements. Furthermore, a direction can be effected, for example, with the aid of angle of incidence and / or angle of emergence detection and / or with the aid of signal transit time measurements. In some configurations of the RTLS anchors, it can be provided that the second communication interface comprises at least two antennas spaced apart from one another in space. The sensor network is particularly preferably designed to determine the relative position of the RTLS tag to the two RTLS anchors on the basis of the distance of an RTLS tag to the respective RTLS anchor detected by at least two RTLS anchors. By measuring the distance to two or more RTLS anchors, it is possible in particular to optimize the position determination of the RTLS tags, for example by means of a triangulation method.The RTLS anchors are preferably designed and configured to determine a distance or position of RTS tags by means of at least one of the following methods: angle of arrival (AoA), angle of separation (AoD), time off light (ToF), multilateration, time of arrival (ToA), line of sight (LoS), two-way ranging (TWR), Symmetrical Symmetrical Double-Way Ranging (SDS-TWR).In some embodiments of the system for controlling the passage of goods to be produced, at least one RTLS tag can be fastened, in particular detachably fastened, to a production-relevant object, for example to a document holder. The document holder can comprise, for example, a clamping board and / or be part of a transport device for the goods to be manufactured. For example, the document holder can be provided for carrying papers of a production order along during the production of the goods. The production order can also be present in electronic form and can be displayed graphically, for example via a display device. The RTLS tags can furthermore have at least one sensor, by means of which an order number of a production order or a production order can be read in. For example, the RTLS tags can have an RFID sensor or an optical sensor, by means of which a barcode, a QR code or an RFID tag of a production order can be read in or scanned. Alternatively, it can be provided to send the order number of the production order or the production order to the RTLS tag via the third communication interface.In particular, the signals from RTLS tags to multiple manufacturing jobs may be detected simultaneously by the RTLS anchors. Thus, all goods that are manufactured or processed in the production environment can be monitored by means of the real-time localization system. Each production order can be assigned at least one or exactly one RTLS tag, for example.The display device can comprise, for example, an electronic paper display (e-paper display). For example, a clamping board with an injection-molded part fastened to the clamping board can be provided, wherein the injection-molded part is designed to fasten an RTLS tag in a detachable manner. Alternatively, an e-paper display can be provided, to which an RTLS tag is fixedly or detachably fastened. The clamping board or the e-paper display can in turn be fastened or integrated on a transport device or on another production-relevant object.Further tags and / or further sensors can furthermore be provided. For example, RTLS RFID tags can be attached to the goods to be manufactured. For detecting the RTLS RFID tags, RFID readers may be attached to one or more manufacturing stations, for example. Additionally, the real-time localization system can then comprise one or more of the following interfaces: active or semi-active RFID, active or semi-active RFID IR hybrid.The production stations can comprise, for example, a production line, a setup station, a conveyor belt, a sorting device, or a component placement installation. Alternatively or additionally, the manufacturing stations can comprise further sensors, for example optical, capacitive, inductive, optical sensors, or ultrasonic sensors for detecting a position of the goods.The invention is explained in more detail below on the basis of exemplary embodiments with reference to the drawings.They show schematically: FIG. 1 shows an exemplary embodiment of a localization system designed as a real-time localization system in an industrial production environment; FIG. 2 ashows an exemplary embodiment of a production-relevant object as a document holder with a detachably attached RTLS tag; FIG. 2 b shows a further exemplary embodiment of a production-relevant object designed as an e-paper display with a detachably fastened RTLS tag; FIG. 3 shows a schematic representation of an exemplary embodiment of a method according to the invention, which is designed to control the passage of goods to be produced through a production environment by means of a model reinforcement learning; and FIG. 4 shows further, exemplary aspects of the method.In the following description of the preferred embodiments, like reference numerals designate like or comparable components.FIG. 1 shows an exemplary embodiment of a localization system 1 designed as a real-time localization system in an industrial production environment. The localization system 1 shown comprises two RTLS tags 3, a plurality of RTLS anchors 5 and an RTLS control unit 7. the RTLS tags 3 each have a first communication interface 3C and the RTLS anchors 5 each have a second communication interface 5C. Furthermore, the RTLS control unit 7 has a third communication interface 7C. In the example, all communication interfaces 3C, 5C, 7C are designed as wireless communication interfaces. In the example, the production environment has four production stations 9 ato 9 d, which can be different stations of a production line, for example. Furthermore, goods 10 aand 10 bare shown, which are processed or produced at the production facilities 9 ato 9 d. The goods 10 aand 10 bare manufactured, for example, within the scope of a first manufacturing order Fa (goods 10 a) and a second manufacturing order Fb (goods 10 b). Incoming production orders Fa, Fb are collected in an order buffer 9 and supplied to the production devices 9 ato 9 daccording to a prioritization. At least one RTLS tag 3 is carried along with the goods 10 a, 10 bfor each production order Fa, Fb. Thus, the goods 10 a, 10 bassociated with a specific production order Fa, Fbcan be localized. For example, it is possible to determine in which of the production facilities 9 ato 9 dthe goods 10 a, 10 bare located. Furthermore, further parameters can be determined which relate, for example, to at least one of the following aspects: a transit time through one of the manufacturing stations 9 ato 9 d, a transit time, buffer dwell time and / or a dwell time of the goods 10, 10 bbetween the manufacturing stations 9 ato 9 d, a position within a manufacturing station 9 ato 9 d.The RTLS tags 3 can be attached, for example, to a production-relevant object, such as a transport device or a document holder 21. The fastening can advantageously be designed to be releasable. In this embodiment, the production-relevant object is carried along with the goods 10 a, 10 band is placed, for example, at or at the respective production station 9 ato 9 d.The goods 10 a, 10 bmay further comprise further RTLS tags 3, for example RFID tags. Furthermore, further sensors, for example RFID readers or optical sensors for object detection, can be installed at the production stations 9 ato 9 dor in their surroundings, so that each individual one of the goods 10 a, 10 bcan be monitored and localized particularly accurately when passing through the production station 9 ato 9 d.The RTLS tags 3 are configured to transmit radio signals received by the RTLS anchors 5. A time stamp can be transmitted as well, or it can be generated by the respective RTLS anchor 5 when a radio signal is received. The RTLS anchors 5 are furthermore designed to determine a distance and / or a relative position of the RTLS tag 3 to the RTLS anchor 5. In particular, it can be provided that the position of an RTLS tag 3 is calculated on the basis of the distance and / or relative position detected by two or more RTLS anchors 5. The position can be calculated in this case by the RTLS control unit 9 or by the RTLS anchors 5. For example, the RTLS anchors 5 can be designed and configured for data exchange with one another. Furthermore, the detected distances and / or positions are transmitted from the RTLS anchors 5 to the RTLS control unit 9. The RTLS control unit 9 comprises a display device 8 for displaying the positions of the RTLS tags 3 or of the goods 10 a, 10 b.FIGS. 2 aand 2 bschematically show two exemplary embodiments of a production-relevant object with an RTLS tag 3. A holder 23 designed as an injection-molded part is fastened to the document holder 21. By means of the holder 23, an RTLS tag 3 can be detachably fastened to the document holder 21. An order document 27 of a production order, for example, can be fastened to the document holder 21, which is carried along with the goods 10 a, 10 bduring production.FIG. 2 bschematically shows an embodiment of the production-relevant object as electronic paper (e-paper) 31 having an e-paper display 33. Analogously to the embodiment of FIG. 2 a, a holder for the detachable fastening of the RTLS tag is also provided here.The RTLS tags 3 can furthermore have one or more further sensors, by means of which information relating to the production orders Fa, Fb can be detected or read in. For example, the sensors may comprise an RFID sensor, and / or an optical sensor such as a barcode or QR scanner. In such embodiments, the production order Fa, Fb or a identification number can be read in or detected simply, so that the RTLS tag 3 is assigned to a specific production order Fa, Fb. Thus, the RTLS tags 3 can be easily associated with a manufacturing order Fa, Fb.FIG. 3 schematically shows an exemplary embodiment of a system 41 for carrying out a method according to the invention in the form of a computer-implemented method. The system 41 comprises a system control unit which is designed and configured to execute a control computer program for optimizing passage times, buffer dwell times and / or other dwell times of the goods. The control computer program includes an artificial neural network 42 having a buffer controller 43. the buffer controller 43 forms an input layer and output layer of the artificial neural network 42 and includes an interface for receiving the RTLS data. The RTLS data comprise the position data of the goods 10 a, 10 bto be produced determined by the localization system 1 or data derived therefrom. In the example, the RTLS data also include the production orders Fa, Fb held in the order buffer 9 and / or incoming production orders Fa, Fb, or data derived therefrom. Thus, the RTLS data can be supplied to the artificial neural network 42 and by means of which an optimum passage of the goods 10 a, 10 bthrough the production environment can be set. In the embodiment of FIG. 3, the artificial neural network 42 also comprises a model-reinforcing learning with a simulation routine 45 and a software agent SA.By means of the simulation routine 45, run-through, buffer and / or lying times for a current prioritization of the order buffer 9 can be calculated in cooperation with the current load of the manufacturing devices 9 ato 9 d. The prioritization of the job buffers 9 can then be optimized with the aid of the software agent SA. An analysis of the previously made decisions of the software agent SA and their effects on the run times, buffer dwell times and / or other dwell times can be taken into account in this case. The system is thus designed to optimize itself.The software agent SA is now designed to output an optimized prioritization of the production orders Fa, Fb to the buffer controller 43 for further processing. In particular, it can be provided that the job buffer 9 is updated according to the calculated prioritization and the production jobs Fa, Fb stored in the job buffer 9 are supplied to the production process according to the prioritization.The job buffer 9 can be organized, for example, as a list and / or be displayed on a display device. Alternatively or additionally, the calculated prioritization can be output via the buffer controller 43 to the localization system 1 and visualized by means of the latter. Furthermore, it can be provided that the prioritization is output directly to a process controller (not shown), so that the production orders Fa, Fb can be supplied to the production devices 9 ato 9 din an automated or partially automated manner in the calculated, optimized sequence.In the example, training data 51 are also shown, which can be supplied to the model-strengthening learning. In particular, it can thus be ensured that the model reinforcement learning can perform an optimization of the buffer list from the beginning. The software agent SA may then analyze the effects (rewards obtained) of the decisions made and implement self-optimizing control of the manufacturing process.FIG. 4 illustrates further aspects of the method by way of example. In the embodiments shown, the software agent SA is designed as a superordinate software agent. Furthermore, the model reinforcement learning here comprises a (subordinate) software agent SA 1 for optimizing the prioritization of a first job buffer AP 1, and a (subordinate) software agent SA 2 for optimizing a second job buffer SA 2. The software agent SA can specify, for example, changes to the optimization strategy or the optimization goal. The two software agents SA 1 and S 2 can then each calculate a new prioritization of the respective job buffers AP 1, AP 2. For example, the production orders of the order buffer AP1 are processed in the production stations 9a and 9b, while the production orders Fb of the order buffer AP2 are processed in the production stations 9c and 9d. In this case, the positions of the goods are detected by the localization system 1 and fed to the machine learning model. The localization system 1 can advantageously again be designed as a real-time localization system. The corresponding position data can be supplied to the higher-order software agent SA and / or to the lower-order software agents SA 1, S 2. As soon as a predetermined criterion is present, the prioritization of the two job buffers AP 1, AP 2 can be recomputed. The criterion can in particular comprise a change of the job buffers AP 1, AP 2. Furthermore, the criterion can comprise further data 61, such as status information or events, which are input, for example, by a user or an external process control system. The higher-level software agent SA can, for example, be based on a strategy-based algorithm. The subordinate software agents SA 1 and SA 2 can in particular be based on a model-based algorithm.Alternatively, the prioritization of the two job buffers AP 1, AP 2 can be effected directly by the software agent SA, so that the two software agents SA 1 and SA 2 can be omitted. The configuration can likewise be modified to the effect that the superordinate software agent SA is dispensed with. In this case, it can optionally be provided that the software agents SA 1 and SA 2 exchange data with one another.List of reference characters1 Localization system 3 RTLS tag 3C Communication interface of RTLS tag 5 RTLS anchor 5C Communication interface of RTLS anchor 7 RTLS control unit 7C Communication interface of RTLS control unit 8 Display device 9 Job buffer 9 ato 9 d Fertigungs stations 10 a, 10 bWare 21 Document holder 23 Holder 25 Clamping device 27 Job document 31 Electronic paper (E paper) 33 Display device / E paper display 41 Control system 42 Artificial neural network 43 Buffer control / data interface 45 Simulation routine 51 Training data 61 External data Fa, Fb Fertigungs jobs AP, AP 1, AP 2 Job buffer SA, SA 1, SA 2 Software agent

Claims

Method for controlling the passage of goods (10a, 10b) to be produced through an industrial production environment, wherein in each case a multiplicity of identical goods (10a, 10b) are combined in a production order (Fa, Fb) and a plurality of production stations (9a to 9d) are passed through, and wherein the production environment comprises a localization system (19) for capturing position data of the goods (10a, 10b) in production, the method comprising: - holding a first number of production orders (Fa, Fb) in an order buffer (9), - locating the goods (10a, 10b) in production environment, analyzing the current load of the manufacturing stations (9a to 9d) of the production environment on the basis of the position data and analyzing the manufacturing orders (Fa, Fb) located in the order buffer (9) by means of a machine learning model, prioritizing the manufacturing orders (Fa, Fb) located in the order buffer (9) by the machine learning model on the basis of the analysis, and processing the manufacturing orders (Fa, Fb) for manufacturing the goods (10a, 10b) in accordance with the prioritization.The method of claim 1, wherein the machine learning model comprises a reinforcement learning model.Method according to claim 1 or 2, wherein the machine learning model comprises minimizing a filling degree of the order buffer (9) and / or minimizing a buffer liveness.Method according to one of claims 1 to 3, wherein the machine learning model comprises the minimization of resting times of the goods (10a, 10b) to be produced during the transition of the goods (10a, 10b) between successive production stations (9a to 9d) and / or the minimization of the total transit time of the goods (10a, 10b) to be produced through the production environment.Method according to one of claims 2 to 4, wherein the production of the goods (10a, 10b) comprises at least two substeps, wherein each of the substeps comprises an order buffer (9), and wherein the reinforcement learning model is designed to minimize the respective buffer laying time and / or the transit time of the goods to be produced through the respective substep.Method according to claim 5, wherein the reinforcement learning model provides for each of the order buffers a software agent (SA1, SA2) for minimizing the respective buffer laying time and / or the transit time of the goods to be manufactured through the respective substep.Method according to one of claims 5 or 6, wherein the model reinforcement learning is designed to minimize the average buffer laying time averaged over all order buffers (9) and / or the transit time of the goods to be produced through all substeps.Method according to claim 6 or 7, wherein the model reinforcement learning is designed and configured for exchanging data between the at least two software agents (SA, SA1, SA2).Method according to one of Claims 2 to 8, wherein the model-enhancing learning calculates a new prioritization of the production orders (Fa, Fb) located in the order buffer (9) when at least one of the following events occurs: a change in the production orders (Fa, Fb) located in the order buffer (9), a prioritizing of a production order (Fa, Fb) by a user or another system unit, a maintenance or fault of a production station (9a to 9d) of the production environment, the anticipated or actual exceeding of a maximum throughput time of a production order (Fa, Fb), the reaching of a time, the expiry of a predefined time interval, a change in an environmental factor.The method according to claims 1 to 9, wherein the machine learning model is initially trained on the basis of training data (51).Method according to any of claims 1 to 10, wherein the locating system (1) is a real-time locating system and the goods (10a, 10b) being manufactured are located in real-time.Method according to claim 11, wherein the real-time locating system (1) comprises a plurality of RTLS tags (3), the method comprising the steps of: - associating a manufacturing order (Fa, Fb) for manufacturing a plurality of identical goods (10a, 10b) with an RTLS tag (3), - attaching the RTLS tag (3) to a manufacturing-relevant object which is carried to each of the manufacturing stations (9a to 9d) during the manufacturing of the goods (10, 10b), - locating the goods (10a, 10b) of a manufacturing order (Fa, Fb) by means of the real-time locating system (1) by locating the corresponding RTLS tag (3).A system for controlling the passage of goods (10a, 10b) to be produced through an industrial production environment, comprising a system control unit and a localization system (1), wherein the localization system (1) is connected to the system control unit or is formed integrally therewith for the exchange of data, and wherein the system control unit comprises a data memory and is formed and configured to execute a computer program stored on the data memory, which is configured to carry out a method according to one of claims 1 to 12System for controlling the passage of goods (10a, 10b) to be produced through an industrial production environment according to claim 13, wherein the localization system (1) is designed as a real-time localization system with at least one RTLS control unit (7), at least one RTLS anchor (5) and at least one RTLS tag (3), wherein the RTLS tag (3) is designed and configured to transmit ID data to en or the RTLS anchors (5) by means of a radio signal and wherein the RTLS anchor or anchors (5) are designed and configured to detect a distance of the RTLS tags (3) to the respective RTLS anchor (5) on the basis of the radio signal and position data based on the ID data and the detected distance, to send the RTLS control unit (9), and wherein the RTLS control unit (9) is further formed integrally with the system control unit or is connected thereto for exchanging data.

Images