DE102018109252A1 - AI system - Google Patents

Abstract

The invention relates to an AI system (10) with several machines (13). Each machine (13) is set up to process and / or transfer a workpiece (14). The machining of a workpiece (14) is understood to mean both a machining and / or a non-machining and / or a forming and / or a measuring and / or a testing of the workpiece (14). The AI system (10) has a central AI node (11) and in each machine (13) each have a local AI node (12). All local AI nodes (12) are communicatively connected to the central AI node (11). The local AI nodes (12) are each assigned to one or more subsystems (21, 22, 23, 24). A common subsystem (21, 22, 23, 24) are those local AI nodes (12) belonging to identical or similar machines (13) or whose machines (13) have identical or similar machine assemblies (19). Within the AI system (10), data is transferred from each local AI node (12) to other AI nodes (11, 12). During operation, dynamically changing process sequence data (P) of the machines (13) are transmitted only between the local AI nodes (12) of a common subsystem (21, 22, 23, 24). As a result, the amounts of data to be transmitted and evaluated within the AI system (10) are reduced and within a subsystem (21, 22, 23, 24) a real-time transmission is enabled.

Description

The invention relates to an AI system with a plurality of communication nodes connected via paths of a network AI node. For example, an AI node may belong to a machine for handling workpieces. "KI" is the abbreviation for "Artificial Intelligence". In connection with this application, the term "AI" or "artificial intelligence" should be understood to mean that the AI system or an AI node is set up to independently learn from ascertained and / or received data and, in particular, to do so Deep learning structure has, for example, a neural network.

Many attempts have been made to predict the behavior and condition of machines. It is customary to collect and evaluate sensor signals of the machine. Such a system is for example off DE 44 47 288 B4 known. There is described a diagnostic module with a parameter model module, a modified ARTMAP neural network, a fuzzy logic module and an expert system. This should make the diagnostic options faster and more accurate and the weaknesses of the included individual modules can be remedied by their combination in the diagnostic module.

However, such a system has the disadvantage that a prediction of certain conditions of the machine, in particular failures of machine components or errors in the operation, is limited.

It may therefore be considered an object of the present invention to provide an AI system for improved operation of the machine.

This object is achieved by an AI system with the features of claim 1.

The KI system according to the invention has a plurality of machines which are each set up for processing and / or for transferring a workpiece. The machines are in particular machines for forming a workpiece and / or pressing. The machine may be, for example, a press for deep drawing and / or extrusion and / or Abstreckgleitziehen and / or punching and / or forging and / or cutting and / or embossing. The machine for transferring workpieces may, for example, be a gripping device or another transfer device, which is set up in particular for feeding and / or removing a workpiece to a processing machine or from a processing machine.

Each machine of the AI system has a local AI node, with each local AI node providing computing power to control and / or monitor the respective associated machine. The AI system also has a central AI node that provides central processing power and can perform higher-level functions. The central AI node is communicatively connected to all the local AI nodes and, unlike the local AI nodes, is not associated with only one machine. The central AI node can, for example, be set up to define the transmission protocol between the AI nodes in the AI system and / or to manage or control the communication in the AI system.

Preferably, each local AI node and / or the central AI node is configured to execute an algorithm for automatic learning. In particular, the particular AI node may act as a deep learning algorithm. Preferably, each local AI node and / or the central AI node has a neural network, for example a convolutional neural network (CNN). For example, each local node 12 and / or the central node may be configured to learn automatically based on pattern recognition algorithms and / or pattern prediction algorithms.

The local AI nodes are at least partially communicative and, in particular, a network allows communication among all of the local AI nodes. For example, all of the AI nodes of the AI system may be communicatively coupled to each other via a local area network, a local area network, or other communication device, such that there is the potential for each AI node to communicate with any other AI node. For example, all KI nodes may be communicatively connected to each other via a corporate Intranet and / or the Internet. Preferably, the central AI node is located remotely from the local AI nodes. The local AI nodes may be in close proximity to each other or at remote geographic locations (eg, different cities or countries).

The central AI node and / or at least one of the local AI nodes is adapted to form at least one subsystem based on the available data of the local AI nodes. The data available is the data that the particular AI node receives from other AI nodes and, if appropriate, the data that a local AI node receives from the machine having the local AI node.

A subsystem has only local AI nodes. The subsystem is formed either by the local AI nodes that are contained in similar or identical machines or by the local AI nodes of the machines with similar or identical machine assemblies.

For example, a subsystem may have all of the local KI nodes of the existing deep drawing presses in the KI system. A subsystem may thus comprise presses of the same press type, for example all deep-drawing presses or all extrusion presses or all ironing presses or all stamping presses or all cutting presses or all forging presses or all embossing presses. A subsystem may also include machines of different machine types, such subsystem including the machines each having a same or similar machine assembly. For example, machines of the AI system may have the same or similar hydraulic pumps, drive motors, actuators, etc. Thus, it is also possible that a local AI node belongs to multiple subsystems.

The local KI nodes of a common subsystem are adapted to transmit dynamically changing process flow data of the machines, bypassing the central KI node, preferably in real time. This allows very fast communication between machines or local AI nodes of the same subsystem. The amounts of data that are transferred within a subsystem are significantly smaller than the data volumes of the entire AI system. By forming one or more subsystems, the data is already filtered at the sending local AI node so that only the data is transferred to each local AI node for them to learn, control the machine, or customize its control are advantageous or necessary. The artificial intelligence of the AI system is thus not only available in the central AI node, but also in the local AI nodes. This can reduce communication to and from the central AI node.

For the assignment of the local KI nodes to a subsystem, for example, a one-time identification of the machine having the local KI node can be carried out and then an assignment to a corresponding subsystem can take place. For example, the central AI node may receive the tag of each newly added machine to the AI system from the respective local AI node and assign it to at least one relevant subsystem. Alternatively, this task can be taken over decentrally by one or more of the existing local AI nodes. Alternatively, it is also possible to evaluate the data of a newly added machine to the AI system and compare it with the data of other local AI nodes or with predetermined comparison data and then perform an assignment to one or more subsystems. The analysis of the data and the assignment to the subsystem may be performed by the central AI node and / or one or more of the local AI nodes.

It is advantageous if the machines are set up to cyclically repeat a process sequence during the machining or during the transfer of the workpiece. Such a cyclical process is typical for the machining or handling of workpieces in series production. In the case of a press, for example, one cycle consists of moving the tool starting from a starting position into an end position for forming the workpiece and then moving it back to the starting position. Typically, the cycle data corresponds to the period for a complete stroke of a press ram.

It is advantageous if each local AI node is arranged to distinguish the dynamically during operation changing process data from stationary data of the machine having the local AI node. In particular, the process flow data is the data that changes over time during a single cycle. Preferably, all other data is stationary data.

In this context, it should be noted that the stationary data are not completely constant over the life of a machine. For example, the stationary data may also change over a longer period of several cycles, such as the temperature of a machine or machine assembly, the ambient temperature, etc. Thus, unlike the process flow data, the steady state data is substantially constant with respect to a single cycle thus static or quasi-static. The steady-state data changes within a given tolerance range within a single cycle, while the process flow data changes beyond a predetermined minimum within a single cycle.

Each local AI node may be configured to identify the data as process flow data that temporally varies by at least a respective predetermined minimum threshold during a single cycle and / or each local AI node is configured to identify the data as process flow data that change in the same or similar time in at least two consecutive cycles. These features can be very easy to recognize the process flow data and used to distinguish the process flow data from the stationary data. Preferably, all data that is not process flow data is stationary data.

It is preferred if each local AI node is arranged to transmit only the stationary data to the central AI node. It may also be advantageous if each local AI node is adapted to communicate only the process flow data of the machine having the local AI node to the at least one other local AI node of the common subsystem.

In particular, each local AI node is adapted to automatically adjust the control of the machine having the local AI node based on the data from other local AI nodes of the same subsystem. The local AI nodes are therefore self-learning. In this context, it is also possible to transmit the learning algorithm of a local AI node to the other local AI nodes of the same subsystem.

For communication within a subsystem, it is advantageous if each local AI node is set up to determine the shortest and / or most optimal transmission link to one or all other local AI nodes of the same subsystem. As a result, the fastest possible data transmission can be ensured, in particular the transmission of the process flow data in real time. For transmission of the shortest and / or optimal transmission path, known algorithms from graph theory, such as the Moore-Bellman-Ford algorithm or the Dijkstra algorithm can be used.

The configuration of the KI nodes and in particular of the local KI nodes can be carried out as explained in German Patent Application No. 10 2018 100 424.9, which is not previously published.

Linking a large number of machines in a common AI system provides a huge amount of data that can be used to learn and adjust the machine's control. Each local AI node thus not only has the data of its own machine available, but also the data from one or more other machines from the same subsystem. As a result, the automatic learning can be performed faster and with higher accuracy. In order to be able to handle the resulting data volumes, not all available data is transmitted to all KI nodes of the KI system, but rather the process sequence data and / or learning algorithms within a common subsystem are exchanged. The transmission of stationary data in the subsystem can be omitted. Stationary data can additionally be transmitted to the central AI node.

• 1 eine schematische, blockschaltbildähnliche Darstellung eines Ausführungsbeispiels eines KI-Systems mit einem zentralen KI-Knoten und mehreren lokalen KI-Knoten,
• 2 eine beispielhafte Darstellung zweier lokaler KI-Knoten und des zentralen KI-Knotens aus 1 und
• 3 und 4 jeweils ein schematisch dargestelltes Ausführungsbeispiel einer Maschine mit einem lokalen KI-Knoten.

Advantageous embodiments of the invention will become apparent from the dependent claims, the description and the drawings. Hereinafter, preferred embodiments of the invention will be explained in detail with reference to the accompanying drawings. Show it:

• 1 1 is a schematic, block diagram-like representation of an exemplary embodiment of an AI system with a central AI node and a plurality of local AI nodes;
• 2 an exemplary representation of two local KI nodes and the central KI node from 1 and
• 3 and 4 in each case a schematically illustrated embodiment of a machine with a local AI node.

In 1 is schematically an embodiment of an AI system 10 illustrated. The AI system 10 has a central AI node 11 and several local AI nodes 12 , Every local AI node 12 belongs to a machine 13 , Every machine 13 is in the embodiment for processing and / or transfer of a workpiece 14 set up. At the machine 13 For example, it may be a press to reshape the workpiece 14 act as it is schematic in 3 is illustrated. The machine 13 For example, it can also be set up to transfer a workpiece ( 4 ). According to the press 3 it may be a press for deep drawing and / or extrusion and / or Abstreckgleitziehen and / or punching and / or forging and / or cutting and / or embossing. Also other machines for molding or forming or processing or measuring or testing a workpiece 14 can be a local AI node 12 and to the AI system 10 according to 1 belong.

As it is schematic in 2 is shown, the central AI node points 11 a central store 15 and a central AI processor 16 on. The central AI processor 16 may be formed by a processor unit or a plurality of interconnected processor units. The at least one processor unit may be a central processor unit (CPU) or a graphics processor unit (GPU).

Each local AI node points accordingly 12 a local store 17 and a local AI processor 18 on. The local AI processor 18 can be done by a single processor unit or be formed a plurality of interconnected processor units. The at least one processor unit may be a central processor unit (CPU) or a graphics processor unit (GPU).

The local AI node 12 or the local AI processor 18 is set up the machine 13 to which he belongs to control. In particular, by means of the local AI node 12 the process flow when editing or transferring a workpiece 14 controlled. In the embodiments described herein, the machine is 13 set up the process flow for machining or transferring a workpiece 14 repeat cyclically, for example in the context of a series production of certain components.

Every machine 13 has several machine components or machine assemblies 19 on. The machine assemblies 19 two machines 13 can be the same or similar. For example, two machines 13 , like two presses, have the same drive motor, in particular the same press drive. A machine component or a machine assembly may also be formed by a motor-pump unit, a hydraulic control unit, or the like. In 2 are the same or similar machine assemblies 19 schematically illustrated by the same graphic symbol. For example, the two machines each have a machine assembly symbolized by the rectangle with rounded corners 19 while the other machine assemblies of both machines 13 differ.

All AI nodes 11 . 12 of the AI system 10 are via a communication network 20 communicating with each other. Therefore, it is possible for all AI nodes 11 . 12 communicate with each other and transfer data between each other. The communication network 20 may for example be a local corporate network or intranet and / or a regional or national network, such as the Internet. In a modification to this, it may also be sufficient that in the AI system 10 not all local AI nodes 12 communicate with each other but, for example, the communication of the local AI nodes 12 to a group or subsystem, as will be explained below, of the AI system 10 is limited.

Every local AI node 12 is set up to send data to the central AI node 11 to transfer and / or data from the central AI node 11 to recieve. Every local AI node 12 is also set up to send data to at least one other local AI node 12 to transfer and data from at least one other local AI node 12 to recieve.

The central AI node 11 and / or at least one of the local AI nodes 12 is set up based on the data available in each case from at least one local AI node 12 at least one subsystem 21 . 22 . 23 . 24 to build. At the in 1 exemplified AI system 10 are a first subsystem 21 (shown in solid lines), a second subsystem 22 (shown in dashed lines), a third subsystem 23 (shown in dotted lines) and a fourth subsystem 24 (shown in double-dashed line) formed. Every subsystem 21 - 24 Each has several local AI nodes 12 on. To a single common subsystem 21 - 24 become the local AI nodes 12 assigned to either identical or similar machines 13 belong or the local AI nodes 12 whose machines 13 identical or similar machine assemblies 19 exhibit.

At the in 1 schematically illustrated embodiment are the first subsystem 21 , the second subsystem 22 and the third subsystem 23 each by such local AI nodes 12 formed, similar or identical machines 13 belong. For example, the local AI nodes 12 of the first subsystem 21 , the second subsystem 22 and the third subsystem 23 belong respectively to presses of the same press type, for example, respectively to deep-drawing presses, forging presses, stamping presses, etc. The fourth subsystem 24 includes machines 13 different types that are not similar or identical. The fourth subsystem 24 is through the local AI nodes 12 the machines 13 each formed an identical or similar machine assembly 19 exhibit. For example, the machines could 13 each presses of different press types, but have the same press drive.

Whether two machines 13 of the AI system 10 may be identical or similar, for example, can be recognized by an identifier that each local AI node 12 contains and to another local AI node 12 or preferably the central AI node 11 transmitted. For example, the central AI node 11 using the tag to map the local AI node 12 the machine concerned 13 to a subsystem 21 - 24 make.

Alternatively or additionally, it is also possible during operation of a machine 13 from sensors of the machines 13 collected sensor data via the local AI node 12 in the local AI node 12 evaluate or transmit and through another AI node 11 . 12 evaluate, for example, through the central AI node 11 , Based on a comparison of the obtained sensor data with other sensor data or with samples or templates in concerned memory 15 . 17 can be the assignment of a machine and thus of the local AI node 12 to a subsystem 21 to 24 respectively. This assignment is always performed when a new machine 13 to the AI system 10 is connected. Usually, it is sufficient to do this assignment once during an initialization of a new machine 13 in the AI system 10 make.

Within the AI system 10 not all of a machine's sensors 13 captured data or from a local AI node 12 generated control data for the different machine assemblies 19 to all other AI nodes 11 . 12 in the AI system 10 transfer. This would result in the transmission of very large amounts of data. Therefore, according to the invention in the local KI node 12 containing the data within the AI system 10 have already made a distinction as to what kind of data is involved. Dynamic process flow data P be within the respective subsystem 21 . 22 . 23 . 24 only to the other local AI nodes 12 transfer. The dynamic process flow data P are from the local AI nodes 12 not at the central AI node 11 transmitted. Also between the different subsystems 21 . 22 . 23 . 24 become dynamic process data P not transmitted.

For example, the dynamic process flow data is P such data that changes dynamically within a cycle during machining, testing, measuring, transfer, etc. of the workpiece. For example, process data of a press may be the position of the press ram, an applied pressing force, the moving speed of the press ram, the position or speed of a floating plate of a die cushion, the speed or torque of a drive motor, and so on. The process flow data P therefore describe the time-varying state of the machine 13 during a single, repetitive cycle.

In a machine 13 will also be stationary data S recorded and within the AI system 10 transfer. For example, stationary data becomes S from any local AI node 12 only at the central AI node 11 transmitted. A transmission of stationary data S between local AI nodes 12 and / or between subsystems 21 . 22 . 23 . 24 does not occur according to the example. At the stationary data S in particular, it is all sensor and / or control data of a machine 13 that is not related to the process flow data P belong. For example, stationary data S an address or identification of a machine 13 , a temperature value of an ambient temperature, a predetermined parameter, for example, a limit value for a maximum pressing force or a maximum ram speed, etc. The stationary data S are not necessarily permanently time-invariant, but do not change within a cycle or only below a minimum threshold that can be predetermined and defined for the individual signals or data. In the period of a single cycle are the stationary data S static or quasi static. Thus, already in the local AI node 12 made a selection as to which data to the other local AI nodes 12 same subsystem 21 . 22 . 23 . 24 be transmitted and what data (stationary data S ) at the central AI node 11 be transmitted.

This allows communication within the AI system 10 and the required bandwidth can be minimized. Every AI node 12 receives only those data that it needs and that for the independent learning and adjusting the control of the respective machine 13 are advantageous or necessary.

Every local AI node 12 is set up to execute a learning algorithm for automatic learning. In particular, each has a local AI node 12 or any local AI processor 18 a neural network 25 which can be formed by a folding neural network (CNN). Every local node 12 or local AI processor 18 may be configured to execute a deep learning algorithm. For example, each local node 12 or local AI processor 18 and to learn automatically based on pattern recognition algorithms and / or pattern prediction algorithms. Due to the data of identical or similar machines 13 or identical or similar machine assemblies 19 within a common subsystem 21 . 22 . 23 . 24 , stand each local AI node 12 all relevant process data P for self-learning. The learning speed and the adaptability of the control of a respective machine 13 in every subsystem 21 . 22 . 23 . 24 is therefore very high. At the same time it avoids all AI nodes 11 . 12 in the AI system 10 Unnecessary data is transmitted for learning and customizing a local AI node 12 have no meaning. In this way it is possible the process flow data P in a common subsystem 21 . 22 . 23 . 24 in real time. The work processes in the machines 13 of a subsystem can thus be adjusted in real time and, for example, during a cycle of the workflow.

In the central AI node 11 can the stationary data S be evaluated. This allows for a machine 13 longer-term Determine predictions such as maintenance times, repair times, failure probabilities for a specific machine assembly 19 , etc. The stationary data S do not affect the dynamic process within a cycle when machining or handling a workpiece 14 but relate to periods of time that are, for example, several orders of magnitude greater than the cycle time duration. The cycle time is typically in the range of a fraction of a second to several minutes, or at most a few hours, during the periods of prediction and evaluation of maintenance and / or repair needs of a machine 13 usually in the range of weeks or months or years.

The AI system 10 can also be set up within the AI system 10 used to synchronize and / or synchronously change transmission protocols used. For example, the transmission or the transmission protocol to be used within the AI system 10 through the central AI node 11 be specified. The optimization of the transmission protocol, for example to minimize transmission times or bandwidths, can be part of the self-learning, so that an adaptation of the transmission during the operation of the AI system 10 is possible. In particular, the transmission protocols are within a subsystem 21 . 22 . 23 . 24 as short as possible, so that a fast transmission, especially in real time, is possible.

To be within a common subsystem 21 . 22 . 23 . 24 the shortest possible and / or optimal transmission link between the associated local AI nodes 12 can be found in any local AI node 12 a suitable algorithm, for example a Bellman-Ford-Moore algorithm or another suitable algorithm, as are known in particular from graph theory.

The invention relates to an AI system 10 with several machines 13 , Every machine 13 is for editing and / or transferring a workpiece 14 set up. Under the machining of a workpiece 14 is both a machining and / or non-machining and / or forming and / or measuring and / or testing the workpiece 14 Understood. The AI system 10 has a central AI node 11 and in every machine 13 each a local AI node 12 , All local AI nodes 12 are with the central AI node 11 communication connected. The local AI nodes 12 are each one or more subsystems 21 . 22 . 23 . 24 assigned. A common subsystem 21 . 22 . 23 . 24 form those local AI nodes 12 that are identical or similar machines 13 belong or their machines 13 identical or similar machine assemblies 19 to have. Within the AI system 10 be from any local AI node 12 Data to other AI nodes 11 . 12 transfer. During operation, the process data changes dynamically P the machines 13 only between the local AI nodes 12 a common subsystem 21 . 22 . 23 . 24 transfer. As a result, the amount of data to be transmitted and evaluated within the AI system 10 reduced and within a subsystem 21 . 22 . 23 . 24 a real-time transmission is enabled.

LIST OF REFERENCE NUMBERS

10

AI system

11

central AI node

12

local AI node

13

machine

14

workpiece

15

central store

16

central AI processor

17

local memory

18

local AI processor

19

engineering group

20

Communication network

21

first subsystem

22

second subsystem

23

third subsystem

24

fourth subsystem

25

neural network

P

Prozessblaufdaten

S

stationary data

QUOTES INCLUDE IN THE DESCRIPTION

This list of the documents listed by the applicant has been generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Cited patent literature

• DE 4447288 B4 [0002]

Claims (11)

AI system (10) with several machines (13) for processing and / or transferring a workpiece (14), each having a local AI node (12), a central AI node (11) communicatively connected to all the local AI nodes (12) of the machines (13), wherein the local AI nodes (12) are at least partially communicating with each other, wherein the central AI node (11) and / or at least one of the local AI nodes (12) is adapted, based on the available data of the local AI nodes (12), at least one subsystem (21, 22, 23, 24 ), wherein a subsystem (21, 22, 23, 24) is formed either by the local AI nodes (12) of identical or similar machines (13) or by the local AI nodes (12) of machines (13 ) is formed with identical or similar machine assemblies (19), wherein the local KI nodes (12) of a common subsystem (21, 22, 23, 24) are adapted to dynamically changing process flow data (P) of the machines (13) To bypass the central AI node (11). AI system after Claim 1 , characterized in that the machines (13) are adapted to cyclically repeat a process flow during the processing or transfer of the workpiece (14). AI system after Claim 1 or 2 Characterized in that each local KI-node (12) is adapted to dynamically changing during operation process flow data (P) of stationary data (S) to distinguish the local KI-node (12) having machine (13) , AI system after Claim 2 and 3 , characterized in that each local AI node (12) is adapted to identify the data as process flow data (P) that temporally varies by one predetermined minimum threshold during each cycle and / or that each local AI node (12 ) is arranged to identify the data as process flow data (P) which changes in time or equal to at least two consecutive cycles. AI system after Claim 4 , characterized in that each local AI node (12) is arranged to identify the data as stationary data (S) which are not process flow data (P). AI system after one of Claims 3 to 5 , characterized in that each local AI node (12) is adapted to transmit the stationary data (S) only to the central AI node (11). AI system after one of Claims 3 to 6 characterized in that each local AI node (12) is adapted to present the process flow data (P) of the machine (13) having the local AI node (12) only to the at least one other local AI node (12) of the common subsystem (21, 22, 23, 24). A KI system according to any one of the preceding claims, characterized in that each local KI node (12) is adapted to control the machine (13) having the local KI node (12) based on data from other local KI nodes (12) of the same subsystem (21, 22, 23, 24). KI system according to one of the preceding claims, characterized in that each local AI node (12) and / or the central AI node (11) is adapted to execute an algorithm for automatic learning. KI system according to one of the preceding claims, characterized in that each local KI node (12) and / or the central KI node (12) comprises a neural network (25). KI system according to one of the preceding claims, characterized in that each local KI node (12) is set up within a subsystem (21, 22, 23, 24) the shortest transmission link to another local KI node (12). to investigate.

Images