System and method for minimizing unproductive idle time within an automation process

Abstract

The invention relates to a system (1) and a method for minimizing non-productive idle times within an automation process performed by an automation facility, the system (1) comprising a model memory (7) storing a probability model (PM) of the distribution of object replacement times of objects used or consumed in the automation process; and an optimizer (13) adapted to calculate an optimal allocation of objects to library locations of an object storage library (3) depending on the probability model (PM) and on the productive non-idle time sequences of objects used or consumed in process steps of the automation process.

Description

[0001] The present invention relates to systems and methods for minimizing unproductive idle time within an automation process performed by an automation facility.

[0002] An automated facility, in particular a factory, includes a plurality of machines suitable for executing a process step of an automated process, in particular a manufacturing or production process. During the automated process, different kinds of objects may be used or consumed. These objects may include, for example, a turning tool (machine tool) and / or a workpiece. The machine of the facility executing the automated process utilizes the turning tool or the workpiece. The object may also include materials consumed in the automated process, such as raw materials. The object may also include a container for transporting such materials. The machine of the factory may use the turning tool, for example, by milling, drilling, or other operations for removing material from the workpiece or adding material to the workpiece to process the workpiece and / or material. During the process step of the automated process, the turning tool may be utilized. In order to use the turning tool, the turning tool may be fixed to the spindle of the machine. Different kinds of turning tools may be stored in a turning tool repository, in particular a so-called rack-type object repository. Whenever a different turning tool is needed for executing a process step of the automated process, the turning tool previously used is removed from the spindle and stored in the repository, for example, by a robot.

[0003] The robot may include a robot arm for taking out and returning an object, such as a turning tool and / or a workpiece or any other object required for performing a process step. The robot does not necessarily have to include a robot arm, but may use any entity, particularly a turning tool, that moves the required object, including, for example, a conveyor belt or any other type of handling equipment. During the manufacturing process, the turning tool may be taken out from an object storage library and then transported to the spindle of the turning tool, where it is fixed. Some object storage libraries may include object carriers, particularly tool carriers. The object carrier allows pre-taking out and post-storage of objects when the machine of the facility performs a process step with the current turning tool. Object carriers may be provided for different types of objects. For example, a turning tool carrier may include two slots at the junction between the spindle for fixing the turning tool and the object storage library of the system. However, even when using these types of object storage libraries including object carriers, non-productive idle time may also be caused due to short process steps.

[0004] This is Figure 1 As shown in the schematic diagram of .

[0005] Figure 1 A machine using sequentially different kinds of turning tools MT is shown, wherein an object carrier having two slots SL1, SL2 is provided to take out the required turning tools MT from the object storage OSM and to return the turning tools MT to the object storage OSM after the corresponding process steps have been completed. Figure 1In the illustrated example, the machine M uses a first turning tool MT1 in a first process step of the automation process at time t0, while a second turning tool MT2 required for the next process step is still stored in the object storage system OSM. The next turning tool MT2 is pre-taken out by the object carrier at time t1 and temporarily stored in the second slot SL2 of the tool carrier, as shown in FIG. Figure 1 At time t2, the process step requiring the first turning tool MT1 is terminated, and the turning tool MT1 is moved to Figure 1 . The object carrier can then, for example, be rotated so that the second slot SL2 carrying the next turning tool MT2 faces the machine M, while the first slot SL1 carrying the first turning tool MT1 that is no longer needed faces the object storage repository OSM. At time t3, the second turning tool MT2 is moved to the machine M and, for example, fixed in the spindle of the machine M for the next process step. The machine M uses the received turning tool MT2 for the next process step. At the same time, the first turning tool MT1 is returned to the object storage repository OSM. At time t4, the second turning tool MT2 is placed back into the first slot SL1 of the object carrier, while the next turning tool MT3 is placed into the other slot SL2 of the object carrier at time t5. At time t6, the third turning tool MT3 is fixed in the spindle of the machine tool M, while the second turning tool MT2 that is no longer needed is placed back into the object storage repository OSM. Figure 1 As illustrated in FIG. 1 , the fourth turning tool MT4 is placed in the second slot SL2 of the object carrier at time t7, and the turning tool MT3 is returned to the first slot SL1 of the object carrier at time t8. Finally, the turning tool MT3 is returned to the object library at time t9. Figure 1 The library idle time MIT and the optimization potential of the object storage library OSM are further illustrated. The illustrated optimization potential OP time period results from the fact that the time for utilizing the second turning tool MT2 is insufficient to store the first turning tool MT1 and pre-take out the third turning tool MT3. In the case that the first turning tool MT1 and the third turning tool MT3 are stored in the library location of the object storage library OSM so that the transport time is reduced, the non-productive idle time can be minimized. The turning tool transport time does depend mainly on the tool path in the object storage library OSM.

[0006] Given that Figure 1 Taking into account the optimization potential illustrated in , the object of the present invention is to provide a method and a system for minimizing unproductive idle times within an automation process performed by an automation facility.

[0007] According to a first aspect of the invention, this object is achieved by a system comprising the features of the first embodiment.

[0008] According to a first aspect, the present invention provides a system for minimizing unproductive idle time within an automation process performed by an automation facility, the system comprising: a model memory storing a probabilistic model of object replacement time distributions of objects used or consumed in the automation process;

[0009] An optimizer adapted to calculate an optimal allocation of objects to repository locations of object storage repositories depending on a probabilistic model and depending on productive non-idle time series of objects used or consumed in process steps of the automated process.

[0010] According to a first aspect of the invention, the probability model stored in the model memory of the system does incorporate random variables and probability distributions into the model of the event or phenomenon. While the deterministic model gives a single possible outcome of the event, the probability model used by the present invention gives a probability distribution as a solution.

[0011] The system according to the first aspect of the invention may be used in a wide range of different automated facilities in which objects are processed or transported.The automated processes performed by the automated facilities may include production processes that use and / or consume objects to manufacture products.

[0012] The automated process may also include the logistics of storing and / or transporting the objects.

[0013] The objects may include different kinds of objects, in particular turning tools and / or workpieces utilized by a machine of the facility performing an automated process.

[0014] Additionally, objects may include materials that are consumed or moved during the automation process.

[0015] The object may include a container suitable for transporting an object such as a turning tool, a workpiece, or material.

[0016] Different kinds of objects can be stored in an object repository of the system. The object repository can include a plurality of repository locations adapted to receive and store objects used in the automation process. The physical size of the object storage locations within the object repository can correspond to the size of the different kinds of objects used in the automation process. The system according to the present invention can include one or more object repositories for the same or different kinds of objects used in the automation process.

[0017] Thus, the system of the invention can be used in a wide range of different kinds of automated processes, in particular production and / or logistics processes.

[0018] In a possible embodiment of the system according to the first aspect of the invention, the object replacement time of the object comprises a fetching time required by the object carrier to fetch the object from a source location in the object storage repository to a target location for use in the automation process, and

[0019] The storage time for an object to be returned by an object transporter from a target location to the same or a different source location in the object storage repository.

[0020] In a possible embodiment, each object repository of the system that may be provided for a particular type of object may include an associated object handler for exchanging the object between the automation facility and the corresponding object repository.

[0021] In a further possible embodiment of the system according to the first aspect of the invention, the probabilistic model stored in the model memory is constructed and / or learned by a model updater of the system based on observed object change times.

[0022] In further possible embodiments, the probabilistic model stored in the model memory is constructed and / or learned depending on observed structural changes of the object repository, as indicated in the repository knowledge graph.

[0023] Providing a model updater has the advantage that the potential of the optimizer to reduce unproductive idle time can be evaluated without running explicit measurement campaigns. Experiments can be used to obtain information about the object repository.

[0024] In a possible embodiment of the system according to the first aspect of the invention, the system is deployed on an edge device on the premise of an automation facility that performs an automation process.

[0025] This offers the advantage that the waiting time for replacing information or data can be reduced, so that the reaction time of the system to a specific event can be minimized.

[0026] In another possible embodiment of the system according to the first aspect of the present invention, the system may be deployed on a cloud platform which is connected via a data network to a site of an automation facility that performs the automation process.

[0027] This embodiment has the advantage that certain services can be provided by the cloud platform provider to the operator of the automation facility.

[0028] In another possible embodiment of the system according to the first aspect of the present invention, the probability model stored in the model memory comprises an artificial intelligence model, a regression model, a Gaussian process model or a Bayesian model.

[0029] Thus, the present invention can utilize a wide variety of different probability models suitable for corresponding use cases. The system can use different probability models employed in the model memory for different kinds of automation processes. Thus, the system according to the first aspect of the present invention can be easily adapted to different automation environments and / or use cases.

[0030] In another possible embodiment of the system according to the first aspect of the invention, the system comprises an instantiator adapted to automatically derive a deterministic model from a probabilistic model stored in a model memory, the deterministic model being applied to an optimizer which calculates an optimal allocation of objects to library locations of object repositories based on the derived deterministic model.

[0031] In a further possible embodiment of the system according to the first aspect of the invention, the system comprises a memory storing a repository knowledge graph comprising information about the structure of the object repository and / or about physical properties of objects stored in the object repository.

[0032] This provides the advantage that the system can adapt or flexibly use different kinds of object repositories and can even take into account changes in the physical structure of the object repositories used.

[0033] In another possible embodiment of the system according to the first aspect of the invention, the structure of the object repository and / or information about the physical properties of the objects stored in the object repository can be automatically obtained by measuring the time required to move objects between different library locations of the object repository, directly or via the system's object handlers, during an automation process or during library idle time when the system's object handlers neither take out nor return objects.

[0034] In a further possible embodiment of the system according to the first aspect of the invention, the system further comprises a risk assessor adapted to assess the risk of incurring additional non-productive idle time by using a specific repository location for storing objects in the object repository.

[0035] In a further possible embodiment of the system according to the first aspect of the invention, the distribution of replacement times of objects in the probability model comprises a continuous density function.

[0036] In a further possible embodiment of the system according to the first aspect of the invention, the process steps of the automation process are controlled by a control program executed by a controller of the automation facility, using an optimal allocation of objects to library locations of object storage repositories of the automation facility as calculated by an optimizer of the system.

[0037] In a possible embodiment of the system according to the first aspect of the invention, the object storage repository comprises a plurality of library locations, each library location being suitable for storing one or more physical objects used or consumed in at least one process step of an automation process performed by an automation facility, or being suitable for storing at least one container for physical objects used or consumed in at least one process step of an automation process performed by the facility.

[0038] In a further possible embodiment of the system according to the first aspect of the invention, productive non-idle time sequences of objects used or consumed in process steps of an automation process performed by the automation facility are predefined in the control program and / or measured by sensor components of the automation facility.

[0039] According to a second aspect, the present invention further provides a computer-implemented method for minimizing unproductive idle time within an automation process performed by an automation facility, the computer-implemented method comprising the features of the second embodiment.

[0040] According to a second aspect, the present invention provides a computer-implemented method for minimizing unproductive idle time within an automation process performed by an automation facility, comprising the following steps:

[0041] providing a probability model of the distribution of object replacement times of objects used or consumed in the automation process, calculating an optimal allocation of the objects to repository locations of the object repository depending on the provided probability model and depending on the productive non-idle time series of the objects used or consumed in the process steps of the automation process, and

[0042] Process steps of the automated process are controlled in response to the calculated optimal assignment of objects to library locations of the object storage repository.

[0043] In the following, possible embodiments of different aspects of the invention are described in more detail with reference to the accompanying drawings.

[0044] Figure 1 A timing diagram is shown for illustrating potential problems of the present invention;

[0045] Figure 2 A block diagram showing a possible exemplary embodiment of a system according to the invention;

[0046] Figure 3 A possible exemplary embodiment of a system according to the invention is shown;

[0047] Figure 4 A flow chart is shown for illustrating a possible exemplary embodiment of a computer-implemented method according to further aspects of the invention.

[0048] Figure 5 An example of a continuous distribution function used in accordance with the method and system of the present invention is shown.

[0049] exist Figure 2In the illustrated embodiment of the figure, an embodiment of a system 1 according to the invention is shown, which comprises a turning tool system 2A connected to an optimization system 2B. The turning tool system 2A is used for handling and managing turning tools MT used by machines of an automated production facility. The turning tools MT form objects used in process steps of an automated process. In the illustrated embodiment, the turning tool system 2A comprises an object repository 3, which is suitable for storing the turning tools MT as objects used by machines 5 of the production facility. Figure 1 In the illustrated schematic block diagram, an object carrier 4 is provided between the machine tool 5 and the object repository 3. The object carrier 4 can, for example, be provided for switching or replacing the machine tool MT required in the continuous production steps performed by the machine tool 5 during the production process. The object carrier 4 can take out one or more objects from the object repository 3 and supply the objects to at least one machine tool 5. The object can be, for example, a turning tool MT fixed to the spindle of the machine tool 5. The turning tool MT can be, for example, a tool for milling a workpiece W or for drilling a hole in the workpiece W. After the automation process step has been completed and the corresponding turning tool MT is no longer needed, the object carrier 4 can take out the turning tool MT from the machine tool 5 and put it back into the object repository 3. Figure 2 In the illustrated embodiment, the turning tool system 2 includes a single turning tool magazine 3. The number and type of object repositories 3 may vary depending on the specific use case or automation process. For example, in a production facility including different types of machine tools 5, there may be different types of object repositories 3, which have different types of objects required for different production steps. Each object repository 3 may include an associated object carrier 4 for transporting objects between the production or automation facility and the corresponding object repository 3. By using actuators, in particular robot arms or conveyor belts, transportation of objects between the object repository 3 and the object carrier 4 and between the object carrier 4 and the machine tool 5 can be provided. The object repository 3 of the turning tool system 2 is provided for storing and retrieving turning tools MT required in the production steps of the production process. The machine tool 5 uses the received turning tool MT for performing different types of production steps, such as milling and drilling. The machine tool 5 can receive other objects from other object repositories 3, for example workpieces or raw materials required for performing certain production steps. Figure 1 The object carrier 4 shown in the figure is optional and is used to speed up the switching, i.e. to reduce the object replacement time. The object replacement time includes the fetching time and the storage time. The fetching time is the time required by the object carrier 4 to fetch the corresponding object from the source library position in the object storage library 3 to the target position for use in the automation process, in particular the process step performed by the machine tool 5. The storage time includes the time required by the object carrier 4 to return the corresponding object from the target position to the same or different source library position in the object storage library 3.

[0050] like Figure 1As illustrated in the figure, the object storage 3 or the object carrier 4 and the machine 5 can be controlled by means of an NC program executed by a controller 6 of the turning tool system 2. The process steps of the automation process performed by one or more machines 5 of the automation facility can be controlled by a control program executed by the controller 6. The control program knows the position or orientation of the different objects in the object storage 3.

[0051] Figure 2 The optimization system 2 shown in comprises a memory 7 which stores a probability model PM of the distribution of object change times for objects used or consumed in the automation process. The probability model PM defines a probability distribution. The probability model PM incorporates random variables and probability distributions into the model of events. In conventional legacy systems, tool change times can be measured by dedicated measurement activities. For such measurement activities, the experiments to be performed need to be specified based on the control NC program. The experiments correspond to a sequence of tool movements from a source library orientation to a target library orientation. These experiments must be performed for various tools MT because the change time also depends on the tool properties of the tool MT. Tool properties may include the weight and geometry of the tool MT. The geometry of the tool MT may also affect which paths are collision-free. Because not all tools MT can be placed in all orientations, experiments must be selected on a tool-by-tool basis. Finally, if the vicinity of the library orientation within the object storage repository 3 is allocated due to collision avoidance, there may be increased movement and / or change times. After performing these experiments, if not all data sets have been collected, the results need to be interpolated. The data set may include a four-tuple (source position, target position, object ID, change time). For a large object repository 3 with multiple locations for storing multiple different objects, collecting the change times for all triplets (source position, target position, object ID) is unduly time consuming. Furthermore, the experiment may need to be performed multiple times to handle statistical uncertainties. The approach used in conventional legacy systems to construct a change time curve while avoiding dedicated measurement activities is based on observing the machine tool 5 and simply collecting statistics on tool changes. However, due to the amount of possible combinations, this process will also take an unduly long time. The probability model PM used in the system 1 according to the invention, stored in the model memory 7, replaces the change time in the four-tuple (source position, target position, object ID, change time) by a distribution over the change times. The distribution may, for example, specify that for a particular triple (source position, target position, object ID), all change times between one minute and two minutes are equally likely, so that the times are distributed with, for example, an average change time of 1.5 minutes and an additional distribution shaping parameter. In a possible embodiment, the distribution of the replacement time of an object such as a turning tool MT in the probability model PM may include a continuous density function CDF, also as Figure 5As shown in . This continuous density function can represent how likely a specific replacement time is. In a simple case, each triple (source position, target position, object ID) has its own assigned continuous density function CDF. The distribution can be, for example, a Weibull distribution or another distribution such as a Gaussian distribution. A so-called Gaussian process can be used. In order to accelerate the generalization of unseen data and dates with little observed evidence in the past, a hierarchical prior of a prior / distribution on the entire row or column of the decomposition / contraction object repository 3 can usually be used to construct a representation. For example, for a rack-type object repository 3, the prior above the average is sampled from the distribution of rows, columns, and specific library positions in the object repository 3. Each prior is a weight. Typically, the prior for a specific position can be initially set to zero, which has a much higher confidence than the prior associated with the rack row and rack column, where the confidence reflects the prior that the positions in the same row have similar replacement times and the positions in the same column have similar additive factors. Information about the physical layout of the object repository 3 can be stored in the library knowledge graph MKG. The library knowledge graph MKG may store information about the physical layout of the object repository 3, the type of the object repository 3, and physical information about the objects stored in the object repository 3. The optimization system 2B may include a memory 8 of the library knowledge graph MKG. Furthermore, the library knowledge graph MKG may include information or data about the physical structure of the object repository 3 and / or about the physical properties of the objects stored in the object repository 3. These objects may include, for example, a turning tool MT or a workpiece W. In a possible embodiment, information about the structure of the object repository 3 and / or about the physical properties of the objects stored in the object repository 3 is automatically obtained by measuring the time required to move objects between different library locations of the object repository 3 via the object carrier 4 during an automation process and / or during a library idle time when the object carrier 4 of the system 1 neither takes out nor returns an object.

[0052] The probability model PM stored in the model memory 7 of the optimization system 2A can be constructed and / or learned by the model updater 9 of the optimization system 2B based on the observed object change times, and / or learned depending on the observed structural changes of the object repository 3 as indicated by the library knowledge graph MKG stored in the memory 8. The library knowledge graph MKG can contain, for example, the structure and information about the rack of the object repository 3. This can include the geometric position of the library orientation, but also the physical characteristics of the objects stored in the rack, such as the characteristics of the turning tool MT stored in the rack. In a possible embodiment, the library knowledge graph MKG can also include information about which object can be stored at which position in the object repository 3. The model updater 9 of the optimization system 2B can observe the object change times when they actually occur, that is, during the runtime of the automation process. The distribution is updated accordingly. The model updater 9 can use different algorithms, such as variational EM, variational algorithms or sampling-based algorithms such as MCMC. The model updater 9 can receive information from the runtime evaluator 10 of the turning tool system 2A.

[0053] exist Figure 2In the illustrated embodiment, the optimization system 2B further includes an instantiator 11. The instantiator 11 is suitable for automatically deriving a deterministic model that can be stored in a model memory 12 from a probability model PM stored in a model memory 7. The deterministic model stored in the memory 12 is applied to the optimizer 13, and the optimizer 13 calculates the optimal allocation of objects to the library locations of the object storage library 3 based on the derived deterministic model stored in the memory 12. Although the probability model PM can represent the replacement time as a distribution in the form of a quadruple (source location, target location, object ID, replacement time), the optimizer 13 does expect a specific time. The instantiation of the probability model PM to the deterministic model can be performed by the instantiator 11 using different instantiation strategies. The first strategy may include a risk-neutral strategy in which the mean of the distribution is selected. This refers to the risk of selecting a wrong deterministic replacement time due to lack of knowledge. The second strategy that can be used to instantiate the probability model PM into a deterministic model may include, for example, a pessimistic strategy that selects a value longer than the mean. The distance to the mean depends on the variance of the distribution. For example, a Gaussian distribution is chosen, and under the assumption that an appropriate prior has been chosen, for example n=1 and σ=68%, a value of μ+nσ (where μ refers to the mean, n refers to the pessimism level, and σ refers to the variance) results in a time less than this value. This strategy is pessimistic because it expects the mean to be overly certain while respecting its knowledge of the variance. A third optimistic instantiation strategy is similar, but with nσ subtracted. Another possible fourth strategy is to sample deterministic replacement times from the distribution. The prior distribution of replacement times may also be initialized from similar other object repositories. In a possible embodiment, the instantiator 11 may receive a control signal for selecting a specific instantiation strategy, which is used to instantiate the probabilistic model PM stored in the memory 7 to obtain a deterministic model stored in the memory 12 and used by the optimizer 13.

[0054] exist Figure 2 In the illustrated embodiment of , the optimizer 13 is a deterministic model. In an alternative embodiment, a probabilistic optimizer can also be used to replace the instantiator 11, the memory 12 and the deterministic optimizer 13. The probabilistic optimizer can directly use the probabilistic model PM stored in the model memory 7 to optimize in terms of replacement time.

[0055] exist Figure 2 In the illustrated embodiment of FIG. 2 , the optimization system 2B further comprises a risk assessor 14. The risk assessor 14 may be used to assess the risk of incurring additional non-productive idle time by using a specific library location for storing objects in the object storage library 3 of the turning tool system 2A. In a possible implementation, the risk assessor 14 may run all object changes through the following process. For each object change (e.g., turning tool change), first the time to perform the process step tp with the current turning tool MT is retrieved.

[0056] Then, the continuous distribution function CDF for storing the last turning tool MT is retrieved by the risk assessor 14. In the next step, the continuous distribution function CDF for retrieving the next turning tool MT is retrieved by the risk assessor 14. The risk assessor 14 forms a convolution on the two retrieved continuous distribution functions CDF. In order to retrieve the continuous distribution function CDF of the unproductive idle time of this particular turning tool MT, the probability mass of all values ​​below the current processing step time to zero is retrieved and subtracted from the current processing time, also as Figure 5 As shown in the figure.

[0057] like Figure 5 The convolution of the continuous distribution function CDF shown in does indeed represent the distribution over unproductive idle time. Figure 5 The probabilities of the change times required for changing two turning tools MT in two consecutive steps of the manufacturing process are shown.

[0058] exist Figure 2 In the embodiment illustrated in FIG. 2 , the optimization system 2B includes an experiment designer unit 15 that can automatically retrieve knowledge or information about the object repository 3 of the turning tool system 2A using the library idle time MIT. Figure 1 As illustrated in the timing diagram of , the library idle time MIT may be caused by a longer production step. If the processing step is significantly longer than the time required to store the previous turning tool MT and to take out the next turning tool MT, in a possible embodiment, the system 1 can use this time to perform measurements or experiments. The purpose of these measurements is to obtain a deeper understanding of the object repository 3 and to update the probability model PM stored in the model memory 7 accordingly. The experiment can typically consist of taking out the object, moving the object to the object carrier 4, and moving the object back to its original orientation in the object repository 3. Alternatively, the object can also be moved to a different library orientation if this does not interfere with the optimizer 13. The design of the experiment can be based on observed lack of knowledge. Knowledge about the library orientation can be retrieved from the observed distribution provided by the experiment.

[0059] Exploration is needed to retrieve additional information about the object repository 3. Otherwise, the system 1 will either focus on library locations with known replacement times and underutilize the potential of the object repository 3, or may take unreasonable risks by using library locations with unknown replacement times. The system 1 according to the present invention can keep risks under control with the help of a risk assessor unit 14. Two sources of exploration can be built into the system 1 implicitly or explicitly. In a possible embodiment, the exploration to retrieve information about the object repository 3 can be based on an experiment designer unit 15. The second unit or component that can be used for exploration is the instantiator unit 11. Sampling based on the instantiation strategy with the greatest risk is risk neutral while leading to a higher degree of experimental experience.

[0060] Figure 2 The system 1 illustrated in the embodiment of can be used to minimize non-productive idle times within an automation process performed by a machine 5 of an automation facility. The system 1 comprises a model memory 7 storing a probability model PM of the distribution of object change times of objects such as turning tools MT used in the automation process and an optimizer 13 adapted to calculate an optimal allocation of objects to library locations of an object storage repository 3 depending on the probability model PM and depending on the productive non-idle time series of the objects used in the process steps of the automation process. In a possible embodiment, the probability model PM stored in the model memory 7 may comprise an artificial intelligence model. Other probability models PM may also be used, in particular regression models, Gaussian process models or Bayesian models.

[0061] like Figure 2 The system 1 shown in the figure can be deployed on an edge device at a site of an automation facility that performs an automation process, or on a cloud platform that is connected to a site of an automation facility that performs an automation process via a data network.

[0062] Figure 3 1 shows a block diagram of an embodiment of a system 1 according to the present invention, wherein the system 1 is deployed on an edge device. Figure 3 As can be seen in FIG. 1 , the optimization system 2B is implemented on the edge device and forms an edge computing system. The turning tool system 2A can be implemented as follows: Figure 3 The machine interface 17 shown in is connected to the edge computing system 2B. The turning tool system 2A includes at least one numerical control unit NCU. The numerical control unit NCU may further include multiple modules, such as a numerical control kernel NCK that performs motion calculations and actually controls the machine process according to the control program. Further modules of the numerical control unit NCU may include a PLC, which can be used to control peripheral devices such as an external object repository 3. The numerical control unit NCU may include further components, such as a drive, a human-machine interface HMI or a communication processor. The numerical control unit NCU of the turning tool system 2A can be connected to the edge computing system 2B via one or more data interfaces, such as a network interface. The interface set including different drivers and data interpreters is in Figure 3Depicted as a generic machine interface 17. Processing components (e.g. applications) hosted on the edge computing system 2B can use different types of information or data about the turning tool system 2A, and can use processed data received via the machine interface 17. The NCU can provide current library information data MID about the object repository 3. The edge computing system 2B can receive and process additional high-precision information about the orientation and speed of the axes, information about the active turning tool MT, the selected turning tool MT, NC codes and other machine configurations and settings. In addition, the interface 17 can be used by the edge computing system 2B to load and execute numerical control programs NCP on the numerical control unit NCU, and possibly send commands to the PLC. Figure 3 In the embodiment illustrated in , the optimization system 2B is deployed on an edge computing device. In a possible embodiment, the process monitoring unit 16 can parse high-frequency process data pd and process events pe during the execution of the numerical control program NCP, including monitoring the duration of machine processing steps and the triggering and completion of tool change events. This data can be made available to the model updater 9 to update the probability model PM in response to the observed turning tool change time. Similarly, the experiment designer unit 15 can extract and use the productive non-idle machining time of a specific NC control program (NCP) to find a suitable time to insert the experiment in order to improve the quality of the probability model PM. The extraction of machining time or processing time can be performed either from the same high-frequency data stream or alternatively from a copy of the running NC control program NCP' obtained via the machine interface 17. These commands can be executed on the device by means of deploying the NC control program or by issuing the turning tool positioning command MT-CMD directly to the PLC and measuring the response time.

[0063] A possible sequence executed by an optimization application running on an edge computing system 2B connected to a numerical control unit NCU of a turning tool system 2A can optimize the orientation of turning tools in a turning tool magazine 3 to reduce the idle time for a given NC control program and tool configuration. The optimization component can obtain the current tool setup of the turning tools MT, including the turning tools in the magazine 3 and their magazine orientations and the overall magazine configuration of the object storage library 3.

[0064] Optionally, it may be chosen to perform a series of change time measurements before running the NC control program. This may be formulated as the NC control program itself and deployed on the numerical control unit NCU of the turning tool system 2A. Alternatively, the selected series may be defined as a series of commands for the PLC. Furthermore, change time measurements are performed and, in a possible embodiment, may be decided based on the current knowledge of the probability model PM and the knowledge required to perform the optimization. Thereafter, the NC control program may be run for which the optimal magazine orientation will be determined. During the execution of the NC program, the process monitoring unit 16 may calculate the sequence of turning tools MT and the duration of the corresponding machining phases as well as the time executed to change the turning tool MT from the turning tool magazine 3.

[0065] Alternatively, the experiment planner unit 15 may send a request to the numerical control unit NCU to execute a turning tool change during a machining phase when the object storage repository 3 storing the turning tool MT is idle.

[0066] After executing the NC program (NCP), the library optimizer 13 may be called with an instantiation of the most recent probability model PM of tool change times, together with information collected by the process monitoring unit 16 about tool sequences and active productive non-idle times and retrieved overall library configuration information.

[0067] In a possible embodiment, the resulting optimization can be presented to a human operator either via a human-machine interface or, alternatively, automatically implemented on the object repository 3 via the machine interface 17. This can be used in the form of an NC program to be run on the NCU to sort the turning tool MT in the best orientation to minimize idle time.

[0068] The system 1 according to the invention provides the possibility to evaluate the potential of the optimizer 13 to reduce the unproductive idle time without running necessary explicit measurement activities. This can be achieved mainly due to the provision of the model updater 9. Furthermore, the experiment designer unit 15 can be used to obtain information about the object repository 3 faster. This allows to predict a certain productivity increase of the automated facility.

[0069] The system 1 further provides a reduced number of experiments or measurements due to a more focused assessment of the library change time. This is a result of an experimental strategy that implicitly takes the lack of knowledge into account. The required generalization to unseen data is due to the library knowledge graph MKG and the decomposed probabilistic knowledge. Therefore, the operator of the automation facility will be more willing to deploy the optimization system, since neither production time is lost due to measurement activities nor any losses are incurred due to such measurement activities.

[0070] A further advantage of the system 1 according to the invention is that no manual interpolation from partial measurement campaigns has to be performed. This is because the system 1 can generalize the acquired knowledge by means of a decomposed distribution using a stored base knowledge graph MKG.

[0071] Another advantage of the system 1 according to the invention is that the risk of errors due to lack of knowledge can be evaluated in a statistically reasonable way. This is mainly due to the provision of the risk assessor unit 14 of the optimization system 2B. In addition, the system 1 according to the invention can learn online and adapt itself to new objects, such as new turning tools MT. This is mainly due to the provision of the model updater 9. The system 1 uses model-based reinforcement learning to achieve object repository optimization. The system 1 according to the invention can improve the ratio of productive time to non-productive time in turning tools by minimizing the non-productive idle time caused by object changes, especially turning tool changes. The system 1 according to the invention can be used for any kind of optimization system in which idle time is minimized. This can also include flexible conveyor belts in production facilities, logistics, material handling, robots and airport logistics. In a possible implementation, the system 1 can be deployed on an edge device such as Sinumerik Edge or SIMATIC Edge. In an alternative embodiment, the system 1 according to the invention can also be run as part of a cloud optimization system. The system 1 according to the invention can measure object change time without inducing additional non-productive time. In a possible embodiment, the probability model PM stored in the model memory 7 can be updated or learned by the model updater 9 during the runtime of the automation system. In a possible embodiment, the probabilistic model PM comprises an artificial neural network ANN that can undergo a reinforcement learning process. The artificial neural network ANN can be learned in a machine learning process using a training data set including object change times. The artificial neural network ANN may include several layers to process data received by the process monitoring unit 16. In a possible embodiment, the model updater unit 9 may be used to continuously train the artificial neural network ANN stored in the model memory 7. In a possible embodiment, the sequence of productive non-idle times of objects used or consumed in process steps of the automation process can be observed using sensor components of the automation facility and / or information derived from the executed control program.

[0072] Figure 4 A flow chart of a possible embodiment of a computer-implemented method for minimizing unproductive idle time within an automation process performed by an automation facility is shown. In the illustrated exemplary embodiment, the computer-implemented method includes three main steps.

[0073] In a first step S1 , a probability model PM of the object replacement time distribution of objects used or consumed in the automation process is provided. The probability model PM can be stored in a model memory 7 .

[0074] In a further step S2 , an optimal allocation of objects to the library locations of the object storage repository 3 is calculated depending on the probability model PM and on the sequence of productive non-idle times of the objects used or consumed in the process steps of the automation process.

[0075] In a further step S3 , process steps of the automation process are controlled in response to the calculated optimal assignment of objects to library locations of the object storage repository 3 .

[0076] In a possible embodiment, the calculation of the optional allocation is performed by an optimizer unit, such as Figure 2 The optimizer 13 illustrated in the embodiment of the embodiment is executed. In a possible embodiment, the automated process controlled in step S3 may include a production process for manufacturing a product using an object such as a turning tool MT. The object repository 3 may include multiple library locations. The object repository 3 may include one or more racks, each rack having a predetermined number of rows and columns to store objects. In a possible embodiment, the library location can be indicated by a triple (rack number, column number, row number). By using the knowledge or information stored in the library knowledge graph MKG, these library locations can be automatically converted into the physical coordinates (x, y, z) of the corresponding library location. In a possible embodiment, the conversion of the library location to the physical coordinates can be performed separately for each object repository 3 depending on the known physical structure indicated in the library knowledge graph MKG of the corresponding library. This allows the flexible use of different types of object repositories 3 in the system 1 according to the present invention. In a possible embodiment, the probability model PM and the library knowledge graph MKG can be loaded from a cloud platform such as that used by the optimizer 13 to calculate the optimal allocation of objects to the library locations within the object repository 3.

Claims

1. A system (1) for minimizing unproductive idle time within an automation process performed by an automation facility, The system (1) comprises: - a model memory (7) storing a probabilistic model (PM) of the object replacement time distribution of objects used or consumed in the automation process; as well as an optimizer (13) adapted to calculate an optimal allocation of objects to the repository locations of the object repository (3) depending on a probabilistic model (PM) and on the productive non-idle time series of objects used or consumed in the process steps of the automation process, The object replacement time of the object includes: - the object is retrieved from the source location in the object storage repository (3) to the target location by the object transporter (4) for the retrieval time required in the automation process, and - The storage time of an object returned by the object transporter (4) from a target location to the same or a different source location in the object storage repository (3).

2. The system according to claim 1, wherein the object comprises a turning tool and / or a workpiece utilized by a machine (5) of an automation facility executing the automation process and / or a material consumed in the automation process.

3. A system according to claim 1 or 2, wherein the probabilistic model (PM) stored in the model memory (7) is constructed and / or learned by a model updater (9) of the system (1) based on observed object change times and / or is learned depending on observed structural changes of the object repository (3) as indicated by the repository knowledge graph.

4. The system according to claim 1 or 2, wherein the system is deployed on an edge device at a site of an automation facility that performs an automation process, or on a cloud platform that is connected to a site of an automation facility that performs an automation process via a data network.

5. The system according to claim 1 or 2, wherein: The probability model (PM) stored in the model memory (7) includes Artificial intelligence models, Regression model, Gaussian process models, or Bayesian model.

6. The system according to claim 1 or 2, wherein: The system (1) further comprises An instantiator (11) is adapted to automatically derive a deterministic model from a probabilistic model (PM) stored in a model memory (7) for application to an optimizer (13), which calculates an optimal allocation of objects to repository locations of an object repository (3) based on the derived deterministic model.

7. The system according to claim 1 or 2, wherein the system (1) further comprises: A memory (8) of a repository knowledge graph (MKG) comprising information about the structure of the object repository (3) and / or about the physical properties of objects stored in the object repository (3).

8. The system according to claim 7, wherein: By measuring the time required to move objects between different library locations of an object storage repository (3) directly or via an object carrier (4) of the system (1) during an automation process or during library idle time when the object carrier (4) of the system (1) neither takes out nor returns objects, information about the structure of the object storage repository (3) and / or about the physical properties of the objects stored in the object storage repository (3) is automatically obtained.

9. The system according to claim 1 or 2, wherein the system (1) further comprises: A risk assessor (14) is adapted to assess the risk of incurring additional non-productive idle time by using a specific repository location for storing objects in the object repository (3).

10. The system according to claim 1 or 2, wherein the distribution of replacement times of objects in the probability model (PM) comprises a continuous density function (CDF).

11. A system according to claim 1 or 2, wherein the process steps of the automation process are controlled by a control program executed by a controller (6) of the automation facility, using an optimal allocation of objects to library locations of object storage repositories (3) of the automation facility calculated by an optimizer (13) of the system (1).

12. The system of claim 1 or 2, wherein the automated process comprises a production process that uses and / or consumes the object to make a product, or The logistical process of storing and / or transporting objects.

13. A system according to claim 1 or 2, wherein the object repository (3) includes a plurality of library locations, each library location being suitable for storing one or more physical objects used or consumed in at least one process step of an automation process performed by an automation facility, or being suitable for storing at least one container for physical objects used or consumed in said at least one process step of an automation process performed by an automation facility.

14. The system according to claim 1 or 2, wherein: The productive non-idle time sequences of objects used or consumed in process steps of an automation process executed by the automation facility are predefined in the control program or measured by sensor components of the automation facility.

15. A computer-implemented method for minimizing unproductive idle time within an automation process performed by an automation facility, The following steps are involved: - providing (S1) a probabilistic model (PM) of the distribution of object replacement times of objects used or consumed in the automation process, - calculating (S2) an optimal allocation of objects to repository locations of an object repository (3) depending on a provided probability model (PM) and on productive non-idle time series of objects used or consumed in process steps of the automation process, and - controlling (S3) the process steps of the automated process in response to the calculated optimal allocation of objects to the repository locations of the object repository (3), The object replacement time of the object includes: - the object is retrieved from the source location in the object storage repository (3) to the target location by the object transporter (4) for the retrieval time required in the automation process, and - The storage time of an object returned by the object transporter (4) from a target location to the same or a different source location in the object storage repository (3).

Images