Method for implementing a process configuration and control unit for optimizing a cycle in implementing the method

Abstract

The invention relates to a method for implementing a process configuration of a transformable process system, wherein the process system has stator objects and moving objects, wherein in an optimization loop, in a network generation step, a transport network is generated, wherein the transport network comprises a graph for connecting the stator objects, and in a strategy generation step, a control strategy agent is generated, wherein the control strategy agent is configured to implement a choreography of the moving objects in the transport network, wherein the transport network and the control strategy agent constitute a process configuration for the transformable process system, wherein in a process configuration evaluation step, the process configuration is evaluated, and depending on the evaluation of the process configuration, the optimization loop is executed again or the process configuration is provided, wherein in an implementation step, the process configuration is implemented in the process system.

Description

Technical Field

[0001] The present invention relates to a method for implementing process configuration of a variable process system, an apparatus for implementing the method, a control unit for implementing the optimized loop in the method, and a computer program. Background Technology

[0002] Document DE 10 2016 216 206 A1 discloses a manufacturing system comprising a lane, at least one evaluation unit, and at least one robotic vehicle, wherein the lane comprises a plurality of modular lane elements, wherein at least two lane elements form a lane segment of the lane, wherein the lane has optical markings, wherein the optical markings include position information, wherein at least one robotic vehicle is movably arranged on the lane, wherein the robotic vehicle has an optical sensor unit for detecting at least one optical marking as position data, and wherein at least one evaluation unit is configured to determine the position and / or orientation of at least one robotic vehicle on the lane and / or in the manufacturing system based on the position data. Summary of the Invention

[0003] The subject of this invention is a method for configuring a process system for implementing a variable process system, having the features of claim 1; an apparatus for implementing the method, having the features of claim 13; a control unit, having the features of claim 14; and a computer program, having the features of claim 15. Preferred or advantageous embodiments of the invention are derived from the dependent claims, the following description, and the accompanying drawings.

[0004] This invention relates to a method for implementing a process configuration for a transformable process system. This process system can implement any process, process task, process sequence, and / or combination of processes in manufacturing and / or logistics. The process system particularly includes a transportation system.

[0005] The process system includes a stator object and a moving object.

[0006] Moving objects are understood in particular as components that change their position within a process system during its operation. Moving objects include, in particular, active objects such as mobile robots, mobile transport equipment, especially FTS (Automated Transport Systems), FTF (Automated Transport Vehicles), AGVs (Automated Guided Vehicles), etc. Active objects can be non-track-constrained and, for example, have wheels for propulsion. Alternatively, they are track-constrained and / or require a stator structure, and are based, for example, on magnetic levitation technology, tracks, etc. Active objects can also be designed as levitated and / or flying objects, such as drones. Alternatively, moving objects can be configured as passive objects, such as transport containers, tool holders, or workpiece carriers, which can be moved, in particular, by means of active objects.

[0007] A stator object is particularly understood as a component of a process system that is stationary during operation. This includes, for example, process stations, manufacturing stations, or logistics stations (e.g., storage locations, storage stations, and / or storage units, such as high-bay warehouses, robotic warehouses, etc.). A stator object also includes possible motion areas for moving objects. These motion areas can be configured as free areas, or alternatively, as lanes, driving modules, or stator structures, such as the magnetic infrastructure for magnetic levitation technology used for moving objects. A stator object particularly includes positioning points that allow for positioning within the process system. The stator object and its positioning points constitute the structural configuration of the process system.

[0008] In this method, an optimization loop is executed once or preferably multiple times. Within the optimization loop, the process system is optimized.

[0009] The optimization loop includes a network generation step for generating the transportation network. The transportation network comprises a graph for connecting stator objects. Specifically, this graph has nodes connected by edges. Nodes can be configured as location points of stator objects, particularly POIs (Points of Interest), or as path points. Path points are particularly used for motion execution and motion orchestration (e.g., navigation points similar to those in aviation flight) and as azimuth points for buffering moving objects. Edges connect the nodes and can form unidirectional or bidirectional connections. The transportation network is spatially defined for hardware configuration and can form a specific transportation system topology.

[0010] The optimization loop further includes a strategy generation step for generating a control strategy agent, which is configured to implement and / or define the orchestration of moving objects in the transportation network. Orchestration refers to the coordinated control and monitoring of all participating moving objects to ensure the efficient, safe, and synchronized operation of processes in the process system. This may optionally include assigning tasks to individual moving objects, dynamically adjusting the routes and schedules of moving objects, avoiding collisions, and integrating with higher-level systems such as ERP or MES platforms. The control strategy agent, in particular, forms the basis for system controllers (e.g., programmable logic controllers) to perform operable control tasks. Control tasks execute the temporally coordinated movement of moving objects to perform motion tasks, especially transportation tasks.

[0011] A transport network with control policy agents constitutes the process configuration of a transformable process system. With the transport network, the process configuration includes all available routes or trajectories for the moving objects. With the control policy agents, the process configuration includes verifying (and especially controlling) the time-varying motion of the moving objects in order to execute the processes of the process system during operation.

[0012] The optimization loop includes a process configuration evaluation step for assessing the generated process configuration. Based on the evaluation of the process configuration, if the evaluation is negative, the optimization loop is executed again to improve the evaluation; or if the evaluation is positive, the process configuration is provided.

[0013] In the implementation steps, the process configuration is implemented in the process system and used by the process system to execute the process in further operation.

[0014] This invention aims to reduce the manual overhead of creating control strategies and map networks for reconfigurable process systems, particularly those with transportation systems. Furthermore, this invention aims to determine better control strategies and transportation networks by automating the currently cumbersome manual design and optimization processes. The following drawbacks can be remedied or at least partially reduced by this invention: current transportation network and control strategy planning is done manually, leading to the following problems: manual overhead and high time requirements; low overall system adaptability due to high overhead; reduction of real-world complexity to make planning easier for humans to understand (e.g., complex continuous motion dynamics are reduced to discrete arrival / departure behaviors); conservative and potentially suboptimal designs due to detailed design iterations; and the development of error-prone control strategies or transportation networks.

[0015] To date, transportation networks have only been acquired and optimized considering available motion areas, resulting in: little consideration of the load scenarios faced by the system (e.g., frequent source / sink connections to obtain multi-lane roads); little consideration of the behavior of mobile robots (i.e., motion dynamics), which may require different arrangements of orientation points or incorrect estimates of the maneuver space (i.e., orthogonal meshes, triangular meshes, irregular patterns such as spline curves); and complex road networks (i.e., large graphs with many unvisited nodes) serving as the basis for computational tasks during operation (e.g., cyclic graph search for path planning).

[0016] To date, control strategies have mostly been deterministic and defined as rules, which leads to: large sets of rules that are difficult to maintain; low optimization when defining general strategies (i.e., rule sets) for various mobile robots with their own behaviors; low optimization when ignoring the load scenarios faced by the transport or mobile robot system; the need to understand and formalize all rules used to handle errors based on system behavior and system state; and limited complexity of control strategies that can be expressed through rule sets.

[0017] To date, developed control strategies and transportation networks have typically remained static due to the overhead and uncertainty of their adjustment (e.g., complex rule sets do not change to avoid system failure); they are not adjusted based on emerging load scenarios; and they accept a low degree of optimization in order to ensure better robustness.

[0018] This method reduces the manual overhead of planning and developing process configurations. Furthermore, by combining process configuration evaluation steps and optimization loops across multiple iterations, the process configuration can be better adapted to the target preset.

[0019] In a preferred embodiment, an initial process configuration is provided as the input configuration for the optimization loop, based on either a manually configured initial process configuration or a real initial process configuration. The manually configured initial process configuration can be generated, for example, through a stochastic principle or a standard process configuration, where the nodes of the graph are uniformly distributed. The real initial process configuration can be taken over by a manually configured process system and optimized using this method. By using the initial process configuration, the optimization loop can be unified for both the initial creation of the optimized process configuration and the reconfiguration of the process configuration, as described later.

[0020] Particularly preferred is that the method includes a reconfiguration step, in which the current process configuration can be optimally generated in an optimization loop based on state data of the process system with the current process configuration. The state data particularly includes load data of the process system during operation, such as the load on edges and / or nodes of moving objects. In particular, the load data may include congestion status, waiting status, etc., of moving objects. Alternatively or supplementarily, the state data includes system changes of the process system caused by changes in the stator object and / or structural configuration. Changes in the stator object may be caused, for example, by a stator object failure or replacement. Furthermore, the state data may optionally supplementarily include changes in the processes that should be implemented by the process system.

[0021] In a preferred embodiment, a monitoring step is performed, wherein the status data of the process system is monitored. The monitoring step can be performed at time intervals or continuously in parallel with the operation of the process system. In a status assessment step, the status data, particularly changes in the status data, are evaluated, wherein a reconfiguration step is initiated based on the evaluation of the status data. For example, in the status assessment step, a small change in the status data may result in no reconfiguration step being performed, while a large change in the status data may result in the reconfiguration step being performed.

[0022] In particular, a reconfiguration step is performed when changes to the process system are manually input or identified through a monitoring step. A control strategy agent is created based on these changes, particularly in parallel with the continuous operation of the system, or at least in advance, so that the control strategy agent is already operational during commissioning, i.e., deployment of the new control strategy agent, thus eliminating the need for system runtime during commissioning. The process system, control unit, and control strategy agent (especially the control core) can be designed to be dynamically adjustable. If additional modules (especially stator objects and / or moving objects) are added to the process system, this can be identified during the monitoring step and trigger the control strategy agent generation process. After the generation process, the added inputs and outputs (especially input neurons and output neurons) of the control strategy agent are connected to the sensors and actuators of the process system (especially the stator objects and / or moving objects).

[0023] In a preferred embodiment, the network generation step is based on the structural configuration of the process system. The structural configuration includes, in particular, the positioning of the stator object, such that the structural configuration includes the positioning points of the stator object, especially POIs. These positioning points are fixed and preset during the optimization loop and cannot be changed during the optimization loop, instead constituting fixed boundary conditions. Thus, the network generation step simplifies to the generation of path points and edges therefrom. When processing path points and edges, other boundary conditions of the moving object can be considered, such as the distances between edges or path points or to the positioning points of the stator object, to consider, for example, the geometry of the moving object. Furthermore, the motion characteristics of the moving object can be considered, such as the minimum curve radius during motion or the transport conditions to reach the stator object. The network generation step can be implemented, for example, through informed or uninformed sampling methods, extended search methods, or a combination of both.

[0024] In the policy generation step, a control policy agent is generated based on the transportation network and structural configuration. The generation of the control policy agent includes the initial generation of the control policy agent and / or the reconfiguration of the control policy agent.

[0025] In principle, a control policy agent can be constituted as a set of rules for defining and / or implementing control policies.

[0026] However, it is preferable to configure the control policy agent as an agent using reinforcement learning methods, wherein the control policy agent is trained to optimize logistics performance indicators related to the configuration of the transportation network and structure. The control policy agent may, in particular, comprise or be composed of neural networks.

[0027] Logistics performance indicators (KPIs) are particularly concerned with the timely execution of the movement of objects to the correct nodes in the transportation network. KPIs include key performance indicators (KPIs) such as: average delay in movement tasks (especially transportation tasks); throughput of movement tasks (especially transportation tasks); time intervals between errors in movement tasks (especially transportation tasks); and the overall effectiveness of the process system.

[0028] To generate the control policy agent, a process system with a transportation network is simulated. The simulation techniques used are kinematic simulation and discrete-event simulation. The control policy agent implements the control policy in the simulation. The simulation demonstrates the multimodal behavior (i.e., kinematics, errors) of the process system and enables the calculation of desired key performance indicators (KPIs). These KPIs are used to reward or penalize the control policy agent during the learning process. The control policy agent then adjusts the control policy to increase rewards and decrease penalties. Therefore, the generated or learned control policy, preset by the control policy agent during the operation of the process system, can directly support the desired motion characteristics of moving objects in the process system.

[0029] Preferably, the control strategy agent is implemented as a transport controller. The control strategy agent, acting as a transport controller, is configured to provide actuator commands to the process system (especially the stator and / or moving object). The stator and / or moving object is configured to execute these actuator commands. In particular, the actuator commands and their execution are implemented in real time. Thus, the control strategy agent is capable of real-time control (especially open-loop and / or closed-loop control) of the process system.

[0030] Particularly preferred is that the control policy agent includes a control core, wherein the control core is configured as a neural network or the aforementioned neural network. This neural network can, in particular, be a convolutional and / or deep neural network (CNN, DCNN).

[0031] In a preferred implementation, the neural network has an input layer with input neurons (especially input nodes). A control strategy agent, particularly a control core, is data-connected to the process system such that state values ​​of the process system are applied to the input neurons. These state values ​​are, in particular, sensor values ​​recorded in the process system and representing the state of moving objects and / or stator objects. Therefore, sensor values ​​can be, for example, binary states, presence signals of moving objects, on / off states (open / closed; locked / released) of components of the moving object and / or stator object, the velocity of the moving object, etc. State values ​​can be constituted as sensor values ​​or derived from sensor values. For example, fused sensor values, which aggregate multiple single sensor values, can be used. Alternatively or supplementarily, the neural network has an output layer with output neurons (especially output nodes), where actuator instructions are output as the output of the neural network at the output neurons, wherein the actuator instructions are applied to the process system, particularly in real time. Thus, the neural network obtains real physical inputs as state values ​​and generates real physical outputs with actuator instructions. In particular, sub-methods including recording state values, processing state values ​​by a neural network, outputting executor instructions, and implementing executor instructions are configured to have real-time capabilities.

[0032] The control core should specifically represent the control logic of the transportation system. The control core is created during the development phase prior to deployment and operation.

[0033] In particular, it can achieve the following advantages or a selection of these advantages: - Reduce the manual overhead of developing rule sets, writing them as code, and implementing the code on controllers.

[0034] - Integrate the currently separated stages: (1) Transportation strategy development; (2) Coding of transportation logic; (3) Deployment and debugging of transportation system; (4) Runtime optimization of process system.

[0035] - Able to update transportation strategies in real time under constantly changing transportation loads.

[0036] - Compared to manually developed, rule-based logic, the control core's control strategy offers higher performance and better robustness.

[0037] In a preferred embodiment, the process configuration is evaluated holistically in the process configuration evaluation step. This can be an independent evaluation or an evaluation that considers the logistics performance indicators of the control strategy agent. In this design, it is considered that specifying the transportation network in the network generation step may limit the optimization potential of the control strategy agent, since this specification occurs before the generation of the control strategy agent. Based on this background, the optimization loop can be restarted in the process configuration evaluation step, whereby the transportation network is first regenerated or generated in an improved and / or modified form, and the control strategy agent is created based on the new transportation network, thereby improving the process configuration holistically.

[0038] In a preferred embodiment of the present invention, the reconfiguration step includes: providing expected transport load information / transport load specifications and the current configuration of the process system (structural configuration and process configuration) for creating a reconfigured process configuration.

[0039] Preferably, during the monitoring step, reconfiguration requirements arising from inadequate process system performance or identified structural configuration (especially hardware configuration) changes are identified as status data. Information regarding process system performance and structural configuration is collected from the process system (i.e., from the real-time system) using sensors and associated software systems (e.g., transportation task systems and hardware configuration management systems, or manufacturing control systems).

[0040] In the condition assessment step, it is preferable to determine whether reconfiguration is needed based on condition data. This decision can be based on current observations (e.g., low throughput) and / or anticipated future events / behaviors (e.g., an expected increase in the number of motion tasks). If reconfiguration is required, information / specifications regarding the expected transport load and the current configuration of the process system (structural and technological) are provided to create the reconfigured process configuration.

[0041] In particular, new requirements regarding the motion task can be introduced during the state assessment step. These new requirements may involve changes to the motion route and / or trajectory. Alternatively or additionally, they may involve variations in the load on the motion route and / or trajectory. In this way, the new requirements can be introduced into the optimization loop as updated target presets to adapt the process configuration to these new requirements.

[0042] Another aspect of the invention relates to an apparatus for implementing the methods described above. The apparatus has a process system configured to perform the implementation steps. Furthermore, the apparatus has a control unit for implementing an optimization cycle. Optionally, the control unit is configured to implement a process configuration evaluation step, a reconfiguration step, a monitoring step, and / or a status evaluation step as part of the optimization cycle.

[0043] Another aspect of the invention comprises a control unit for implementing the optimized loop and the steps described above. This control unit can be configured, in particular, as a digital data processing device, a cloud, a computer, etc.

[0044] Another subject of the invention comprises a computer program having program code means, wherein, when the computer program is executed on a control unit and / or on a process system and / or on a digital data processing device, the computer program is configured to perform the optimization loop as described above.

[0045] This method thus relates to an automated control strategy and a method for generating reconfigurable process systems (especially as transportation systems) and transportation networks, preferably based on artificial intelligence.

[0046] This computer-aided approach is coupled with the physical process system and serves the associated generation and operation phases. The approach can also be used independently to support the development phase and subsequent implementation.

[0047] In light of the shortcomings discussed above, the present invention aims to provide the following advantages: Reduce manual overhead in planning and developing control strategies and transportation networks.

[0048] By optimizing control strategies and reducing the transportation network under specific load scenarios, the complexity of computational operations is reduced. Better, more robust control strategies are developed through "search and learning," rather than using human-preset strategies that may not be fully optimized due to associated overhead. The optimized control strategies and transportation networks are automatically updated during system operation to better handle emerging load scenarios.

[0049] This invention relates particularly to adaptive process systems (especially transportation systems) that should be reconfigured. Therefore, these process systems include a modular architecture that can be divided into stator infrastructure (e.g., road modules, magnetic infrastructure, conveyor belts, etc.) as stators and movers (e.g., wheeled vehicles, magnetic movers, pallets on conveyor belts, etc.) as moving objects. The movers are active (e.g., vehicles with their own drive systems) or passive (e.g., boxes) objects and can carry payloads to be transported (e.g., assembly components). The movers can also move on different types of stator infrastructure, resulting in hybrid systems (e.g., wheeled movers that, due to their structural design, can also move on conveyor belt infrastructure).

[0050] Control strategies and transportation networks define specific hardware or structural configurations. These configurations describe a particular infrastructure layout and a defined set of mobile devices. A transportation network can be formalized as a node-edge network, consisting of nodes (i.e., points of interest (POIs) and waypoints (WPs)) and edges (i.e., one-way or two-way transitions). POIs can be source / sink points of materials, transition points in specific areas, etc. WPs are orientation points used for motion execution and coordination (similar to navigation points in aviation) and for buffering mobile devices. The transportation network spatially defines the hardware configuration, thus forming a specific transportation system topology.

[0051] Based on the topology of a transportation system, control strategies can be applied to coordinate the movement of vehicles. When specifying an optimized control strategy, many different objectives (such as throughput, path redundancy, etc.) and constraints (i.e., collision-free movement or the ability of a vehicle to traverse only specific edges) must be considered. The control strategy forms the basis of the operational control tasks executed by the system controller (such as a programmable logic controller). These control tasks execute the time-controlled movement of vehicles to perform transportation. Together, the transportation network and the control strategy constitute the operational program configuration of the process system, i.e., the process configuration.

[0052] This invention is applied to the stage of generating program configurations during the program configuration management process of reconfigurable process systems (especially their transportation systems). This process identifies reconfiguration requirements based on the real-time system, generates new program configurations with optimized attributes, and deploys the new program configurations in the real-time system's controller. Attached Figure Description

[0053] Other features, advantages, and effects of the present invention will become apparent from the following description of preferred embodiments of the invention and the accompanying drawings. The drawings show: Figure 1 A block diagram of an apparatus with a process system is shown, wherein a method for process configuration for implementing the process system is performed on the apparatus. Figures 1-6 Representative process systems with different transportation networks are shown; Figure 7 A schematic diagram of a control strategy intelligent agent with a control core is shown. Figure 8 It shows Figure 7 Control strategy intelligent agent and Figure 6 A schematic diagram illustrating the interactions between process systems. Detailed Implementation

[0054] Figure 1 A schematic diagram illustrates an apparatus 1 as an embodiment of the present invention. Apparatus 1 includes a process system 2 and a control unit 3. The control unit 3 is configured as a digital data processing device. A method for configuring the process system 2 is implemented in apparatus 1.

[0055] Figure 2 The accompanying drawings show a plan view as a representative example of the process system 2 and its structural configuration. The process system 2 has a stator object 4 that defines its structural configuration. The stator object 4 can be designed as a station 5, particularly a processing station and / or a logistics station. The stator object 4 also consists of free regions 6 for motion trajectories. Optionally, the process system 2 can be combined by infrastructure modules 7 serving as the stator object 4, each infrastructure module 7 providing the free regions 6 and further or other structural details. The sum of the stator objects 4 as fixed components and their spatial distribution constitute the structural configuration.

[0056] The process system 2 has a moving object 8, wherein in this embodiment, the moving object 8 is configured as a transport robot 9.

[0057] exist Figure 2 The diagram illustrates a transportation network 10, which has path points 11 and edges 12, and positioning points 13 for stator objects 4 (especially station 5). Points 11 and 13, together with the edges 12 connecting them, constitute a graph that defines the possible movement routes and / or trajectories of the moving object 8 (especially the transport robot 9). The transportation network 10 is at least largely configured as a virtual transportation network 10 with respect to the path points 11 and edges 12, which can be freely configured in principle.

[0058] The orchestration (i.e., the control and coordination of the moving object 8) is achieved through a control strategy agent 18, which may be configured as a neural network, for example. The control strategy agent 18 controls and verifies the moving object 8 in real time.

[0059] By observing process system 2, it can be determined that the transportation network 10 and the control strategy agent 18 are implemented solely in software technology, allowing them to be freely adapted to the corresponding process tasks. The combination of the control strategy agent 18 and the transportation network 10 is therefore referred to as process configuration.

[0060] Figure 1 The generation of process configurations for process system 2 was visualized, especially the generation of regeneration or reconfiguration.

[0061] The main component is the optimization loop 15 that generates the process configuration. In the optimization loop 15, for the first process configuration that serves as input configuration, an estimated or otherwise generated artificial initial process configuration or a real (e.g., manually set) initial process configuration is used.

[0062] In network generation step 100, transportation network 10 is generated.

[0063] In strategy generation step 200, a control strategy agent 18 is generated based on the transportation network 10. In process configuration evaluation step 300, the process configuration including the transportation network 10 and the control strategy agent 18 is evaluated. If the evaluation is negative, the optimization loop 15 with steps 100 and 200 is executed again. If the evaluation is positive, the transportation network 10 and the control strategy agent 18 are delivered to the process system 2 and used there in the process.

[0064] The operation of process system 2 is monitored in monitoring step 400. In particular, status data of process system 2 is acquired, for example, through sensors or enterprise business systems, and collected in monitoring step 400. Monitoring step 400 forms the basis of reconfiguration branch 17.

[0065] In the state assessment step 500, the collected state data is evaluated. Optionally, changes in the structural configuration of the process system 2 or its functions (especially those related to motion tasks) can be entered as state data in this monitoring step 500, either automatically or manually. In the state assessment step 500, the state data is evaluated, and a reconfiguration step 600 is initiated based on the evaluation.

[0066] In reconfiguration step 600, status data, if necessary, including changes in the structural configuration and / or functionality of process system 2, is imported into optimization loop 15 as optimization input 19. This functionality may specifically involve new requirements for changed motion routes and / or trajectories. Alternatively or supplementarily, the new requirements may involve changes in the load on the motion routes and / or trajectories. In this way, the new requirements can be introduced into optimization loop 15 as updated target presets to adapt the process configuration to these new requirements. Alternatively or supplementarily, performance indicators that serve as targets for optimization loop 15 can also constitute part of the status data.

[0067] There, the optimization loop 15 with steps 100 and 200 is executed multiple times as necessary, and the reconfigured process configuration is checked again in the process configuration evaluation step 300, and then delivered to process system 2 again for operation based on the evaluation.

[0068] In step 700, the process configuration is delivered to process system 2.

[0069] Optimization loop 15 executes optimization iterations until it is stopped. Generation stops when the desired characteristics associated with the optimization objective are achieved. The optimization objective of the transportation system in process system 2 is primarily its logistics performance characteristics (i.e., transporting goods and arriving at the expected destination on time), which are formalized as KPIs (Key Performance Indicators) (e.g., average transport task delay, throughput, mean time between failures, overall equipment efficiency, etc.). The determination of these KPIs is based on simulation. The simulation applies the transportation network and control strategies developed during the optimization loop. Furthermore, the simulation considers the multimodal behavior (i.e., motion characteristics, failure behavior) of specific elements of the transportation system, as well as transportation load scenarios (the "transportation load information" specified in the demand identification phase configured earlier for the generation process). The simulation techniques used include kinematic simulation and discrete event simulation.

[0070] Optimize loop 15 by performing the following steps: Network generation step 100: The transportation network 10 is generated based on the structural configuration of the transportation system of process system 2. Location points 13 can be derived from the structural configuration. The generation problem is thus reduced to defining path points 11. Path points 11 of the transportation network 10 must contain buffer positions for moving objects 8 and 9 and optimally support their movement during the execution of transportation tasks. In the first iteration, an initial solution must be provided (e.g., from the current system configuration) or determined (e.g., through trial and error). Determining the transportation network 10 is a geometric problem, taking into account the motion possibilities of moving objects 8 and 9 (e.g., continuous XY coordinates / nodes for wheel-constrained elements, or start / end / buffer coordinates / nodes for conveyors). In addition to the nodes 11 and 13 of the network 10, edges 12 specify possible transitions. These transitions include compatibility between conveyors and infrastructure and may also have geometric effects (e.g., specific trajectories resulting from the motion characteristics of conveyors between nodes). The formal representation of the transportation network 10 as its foundation is a graph. The methods used to construct path points 11 include: random or structured methods (e.g., isometric, pattern methods), global or local refinement / optimization methods (e.g., force-directed graph plotting, such as Fruchterman-Reingold), and motion-related trajectory definition methods. The specific application of these methods is part of the network generation step 100.

[0071] Strategy generation step 200: Based on the transportation network and structural configuration, a control policy can be created. Since rule-based definitions are cumbersome and often have limited expressive power, an artificial intelligence-based (i.e., reinforcement learning) approach is adopted. For this purpose, a simulation of process system 2 (especially its transportation system) is used as the environment, which is controlled by a control policy agent that learns and represents the control policy. The trained control policy agent can be directly applied to the real transportation system controller of process system 2. The simulation environment demonstrates the multimodal behavior (i.e., kinematics, errors) of the transportation system of process system 2 and enables the computation of desired key performance indicators (KPIs). KPIs are also used to reward the control policy agent during the learning process. Therefore, the control policy directly supports the desired transportation characteristics of the system. However, the formalization of transportation network 10 may limit the optimization potential of the control policy, which explains the sequential construction of optimization loop 15. After the learning process, the generated control policy agent 18 can be tested using test scenarios (e.g., module testing) to prevent undesirable behavior. The simulation environment and reinforcement learning are largely based on the behavior of the node / edge representations 11, 13, 12 of transportation network 10 and the moving objects 8, 9.

[0072] Subsequently, in the phase of providing program or process configuration, the process configuration is delivered to process system 2 in implementation step 700. After creating a new process configuration, it is necessary to provide the process configuration with optimized features in the real-time system for it to take effect. To this end, a new transportation network and associated control policies embodied in the control policy agent 18 are provided in the controller of the transportation system of process system 2. Subsequently, the program configuration management process restarts.

[0073] The phase for identifying process configuration reconfiguration requirements aims to identify program reconfiguration requirements arising from inadequate performance of the transportation system or identified structural changes in hardware configuration. Information regarding the performance of the transportation system in process system 2, as well as information regarding hardware configuration, is acquired from the real-time system using sensors and associated software systems (e.g., transportation task and hardware configuration management systems or manufacturing control systems). Based on this information, a decision is made as to whether reconfiguration is necessary. This decision may be based on current observations (e.g., low throughput) or anticipated future events / behaviors (e.g., an expected increase in the number of transportation tasks). If reconfiguration is required, anticipated transportation load information / specifications and the current configuration of the transportation system (structural and technological) are provided to create a technologically appropriate configuration.

[0074] exist Figure 2 As shown, moving object 8 becomes an obstacle due to a technical defect. In monitoring step 400, this event is collected in the state data of process system 2. In state assessment step 500, a reconfiguration requirement is identified based on the state data, triggering reconfiguration step 600. In optimization loop 15, a new transport network 10 with redundant motion tracks for the defective moving object 8 is generated. Furthermore, additional path points 11 and edges 12 are generated to restore or improve the performance of process system 2, and the control policy agent 18 is regenerated (especially retrained). This updated transport network 10... Figure 3 As shown in the image.

[0075] exist Figure 4 The diagram illustrates a change in structural configuration, in which another infrastructure module 7 in the form of a conveyor belt 20 is added. This change in structural configuration is acquired by monitoring step 400 and evaluated in state evaluation step 500 to trigger reconfiguration step 600. In the state data, the updated structural configuration with the conveyor belt 20 is passed to optimization loop 15. Subsequently, a new transportation network 10 using the conveyor belt 20 is generated, and the control policy agent 18 is regenerated (especially retrained). The updated transportation network 10... Figure 4 As shown in the image.

[0076] exist Figure 5Another change in structural configuration is shown, in which another infrastructure module 7 with free space 6 is added. This change in structural configuration is handled in the same manner as the addition of the conveyor belt. The updated transport network 10 is in Figure 5 As shown in the image.

[0077] Figure 6 Another plan view is shown as an example representing process system 2 and its structural configuration. Process system 2 has a stator object 4 that determines the structural configuration of process system 2. The stator object 4 can be configured as a station 5, particularly a processing station and / or a logistics station. The stator object 4 also consists of a free area 6 for the motion trajectory. Optionally, process system 2 can be combined by infrastructure modules 7 serving as stator objects 4, wherein each infrastructure module 7 provides the free area 6 and other structural details. In addition, a conveyor belt 20 is provided as a stator object 4. The sum of the stator objects 4 as fixed components and their spatial distribution constitute the structural configuration.

[0078] The process system 2 has a moving object 8, wherein in this embodiment, the moving object 8 is configured as a transport robot 9.

[0079] exist Figure 6 In this system, the transportation network 10 has path points 11 and edges 12, and positioning points 13 for stator objects 4 (especially station 5). Points 11 and 13, together with the edges 12 connecting these points 11 and 13, constitute a graph that defines the possible movement routes and / or trajectories of moving objects 8 (especially transport robots 9), and / or the movement routes and / or trajectories of conveyor belt devices 16. The transportation network 10 is at least largely configured as a virtual transportation network 10 with respect to path points 11 and edges 12 that can be freely configured in principle.

[0080] Therefore, the transportation network 10 includes a graph, which can be represented as a Petri net or in other suitable graphical representations. Figure 6 In this system, two points of interest (POIs) serve as location points 13, acting as the source and sink of the moving object 8, respectively. Location points 13 can have different functions. Similarly, path points 11 can also have different functions. For example, path points 11 can serve as reporting points, stopping points, or buffer locations for the moving object 8 and its transported goods. The functions of points 11 and 13 are ensured through the interaction of the electromechanical elements of the process system 2. For instance, when the moving object 8 approaches the stopping point 11, the process system, as an electromechanical infrastructure system, will stop the moving object 8 at that path point 11.

[0081] The orchestration (i.e., the control and coordination of the moving object 8) is implemented by a control strategy agent 18, which is configured, for example, as a neural network. The control strategy agent 18 controls and verifies the moving object 8 in real time.

[0082] Figure 7 A schematic diagram of a control strategy agent 18 is shown, illustrating how it is deployed in process system 2. The control strategy agent 18 has a control core 21, which is configured as a neural network, particularly a convolutional and / or deep neural network.

[0083] The control core 21 has an input layer 22 with input neurons 23 and an output layer 24 with output neurons 25. Multiple hidden layers 26 exist between the input layer 22 and the output layer 24. The control core 21 can have any number of hidden layers 26. Each layer 26 can have any number of nodes and include arbitrary connection edges to simulate synapses. The specific structure of the hidden layers 26 is chosen to achieve optimal control of the transport arrangement 2. The control core 21 can be generated and / or trained using reinforcement learning or similar techniques.

[0084] Input layer 22 serves as the input layer for control core 21, receiving state values ​​S_1, ..., S_Y from the system to be controlled (i.e., process system 2). Input neurons 23 receive state values ​​S_1, ..., S_Y from process system 2 (especially stator object 4 and / or moving object 8). State values ​​are, for example, sensor values. These state values ​​(especially sensor values) can be read directly from the sensors of stator object 4 and / or moving object 8 (“real sensors”), or can be obtained by fusing multiple sensors of stator object 4 and / or moving object 8 (“virtual sensors”) and / or by fusing through a data stack.

[0085] Some examples of sensors and / or status values ​​include: (1) The number of controlled or material flow objects present at a specific path point 11 or location point 13 (WP / POI); (2) Remaining delivery time of the material flow object; (3) The current velocity of the material flow object; (4) The current heading of the material flow object.

[0086] For example, the state value (especially the sensor value) at the path point 11 or the location point 13 at the source end of the moving object 8 or the general material flow object can indicate whether the source end is occupied by the moving object 8, or the state value (especially the sensor value) at the edge 12 can indicate whether the moving object 8 is in motion.

[0087] The number of input neurons 23 and the number of state values ​​acquired therefrom are freely selectable and are chosen to achieve optimal control over process system 2.

[0088] Output layer 24, serving as the output layer of control core 21, contains output neurons 25 that output actuator commands A_1, ..., A_Z to process system 2 (specifically, stator 4 and / or moving object 8). The actuator commands A_1, ..., A_Z cause specific material transport processes or behaviors of stator 4 and / or moving object 8. An example of this is switching stoppers on conveyor belt 20 or enabling moving object 8 to cross the edge 12 between nodes 11 and 13 in the transport network using Boolean output neurons 25.

[0089] Figure 8 It shows Figure 6 Contains Figure 7 The process system 2 is a control strategy agent 18, which has a control core 21. For details regarding the process system 2 or the control strategy agent 18, please refer to the description in the corresponding figures.

[0090] The stator object 4 and / or the moving object 8, and the process system 2 therefrom, constitute the electromechanical system of the device 1. The process system 2 determines the inputs and outputs of the control core 21 and, consequently, the inputs and outputs of the control strategy agent 18, especially the input neuron 23 and the output neuron 25.

[0091] Input neuron 23 is connected to a sensor in process system 2 to obtain state values ​​as sensor values ​​directly from process system 2. Alternatively, the state values ​​as sensor values ​​are pre-fused to form a virtual sensor, the state values ​​of which are then applied to input neuron 23. These state values ​​are provided and / or processed in real time.

[0092] In the example shown, state values ​​S_1 and S_3, indicating that location point 13 is occupied and path point 11 is free, are transmitted from location point 13 and path point 11. Furthermore, state value S_3, indicating that edge 12 is just available for transport, is transmitted from edge 12 connecting points 11 and 13 to input neuron 23 of control core 21. As previously described, state values ​​are acquired from sensors as sensor values, or derived from these sensor values.

[0093] Output neurons 25 are connected to the actuators or their PLCs (Programmable Logic Controllers) of the process system 2 (especially the stator object 4 and / or the moving object 8), and implement the actuator instructions A_1, ..., A_Z provided by the output neurons 25, especially in real time. In response to received state values ​​S_1, S_2, S_3, the control core 21 generates an actuator instruction A_1 at one of the output neurons 25, stating that the material flow object (in this case, the moving object 8 located at positioning point 13) should be transported via edge 12. This actuator instruction A_1 is implemented, especially in real time.

[0094] Path point 11 and location point 13 are connected by edge 12, which allows the moving object 8 to traverse between path point 11 and / or location point 13. The movement of the moving object 8 is triggered by a “traverse()” instruction, which is received in the process system 2 as an actuator instruction A_1 from the control strategy agent 18 (especially the control core 21). This actuator instruction A_1 is then interpreted and implemented by the stator object 4 and the moving object 8. Thus, for example, at the mechanical stop at the stop path point 11 of the conveyor belt 20, the stop is released to achieve the movement.

Claims

1. A method for implementing a process configuration of a transformable process system (2), wherein the process system (2) has a stator object (4) and a motion object (8). in, In optimizing loop (15): In the network generation step (100), a transport network (10) is generated, wherein the transport network (10) includes a graph for connecting the stator objects (4), and In the strategy generation step (200), a control strategy agent (18) is generated, wherein the control strategy agent (18) is configured to implement the arrangement of the moving objects (8) in the transportation network (10). The transportation network (10) and the control strategy agent (18) constitute a process configuration for the transformable process system. In the process configuration evaluation step (300), the process configuration is evaluated, and based on the evaluation of the process configuration, the optimization loop (15) is executed again or the process configuration is provided. In the implementation step (700), the process configuration is implemented in the process system (2).

2. The method according to claim 1, characterized in that, Based on a manual initial process configuration or a real initial process configuration, the initial process configuration is provided as the input configuration for the optimization loop (15).

3. The method according to any one of the preceding claims, characterized in that, In the reconfiguration step (600), an updated process configuration is generated in the optimization loop (15) based on state data, wherein the state data includes load data and / or system changes of the process system (2).

4. The method according to claim 3, characterized in that, In the monitoring step (400), the status data of the process system (2) is monitored, and in the status evaluation step (500), the status data is evaluated, and the reconfiguration step (600) is initiated based on the evaluation of the status data.

5. The method according to any one of the preceding claims, characterized in that, The network generation step (100) is performed based on the structural configuration of the process system (2).

6. The method according to any one of the preceding claims, characterized in that, The strategy generation step (200) is performed based on the structural configuration of the process system (2) and the transportation network (10).

7. The method according to any one of the preceding claims, characterized in that, The control strategy agent (18) is implemented as a transport controller or constitutes a transport controller, wherein the control strategy agent (18) provides actuator instructions (A1, ..., A_Z) to the process system (2).

8. The method according to claim 7, characterized in that, The control strategy agent (18) includes a control core (21), which is configured as a neural network or includes a neural network.

9. The method according to claim 8, characterized in that, The neural network of the control core (21) has an input layer (22) with input neurons (23), wherein the input neurons (23) receive the state values ​​(S_1, ..., S_Y) of the process system (2), and / or the artificial neural network has an output layer (24) with output neurons (25), wherein actuator instructions (A1, ..., A_Z) are output at the output neurons (25) as the output of the neural network, wherein the actuator instructions (A1, ..., A_Z) are applied to the process system (2).

10. The method according to any one of the preceding claims, characterized in that, The process configuration evaluation step (300) includes an evaluation of the control strategy agent (18) and the transportation network (10).

11. The method according to any one of the preceding claims, characterized in that, The reconfiguration step (600) includes providing the current configuration of the process system (2) for creating the process configuration.

12. The method according to any one of the preceding claims, characterized in that, In the monitoring step (400), the performance of the process system (2) is monitored for inadequacy or changes in the structural configuration of the process system (2) are identified.

13. An apparatus (1) for carrying out the method according to any one of the preceding claims, comprising: The process system (2) for implementing the implementation step (700), and Control unit (3) for implementing the optimized loop (15).

14. A control unit (3) configured to implement the optimized loop (15) for the method of any one of claims 1 to 10.

15. A computer program comprising program code means, when the computer program is executed on a control unit (3) according to claim 12 and / or on a device (1) according to claim 11 and / or on a digital data processing device, for implementing an optimized loop (15).

Images