CELL CONTROL AND PRODUCTION SYSTEM FOR MANAGING THE WORKING SITUATION OF A NUMBER OF MANUFACTURING MACHINES IN A MANUFACTURING CELL

Abstract

Cell control (12B) which is connected to a host computer (13) with a database and is set up to control a manufacturing cell (11) comprising a plurality of manufacturing machines (14, 15, 16), wherein the cell control (12B) comprises: a multitude of sensors (19) which is set up to determine the state of each of the multitude of manufacturing machines (14, 15, 16); a state storage unit (20) which is set up to store the state of each of the manufacturing machines (14, 15, 16) as determined by the sensors (19); a fault information acquisition unit (21) which is configured to acquire the fault information from each of the production machines (14, 15, 16) from the status information of each of the production machines (14, 15, 16) which is stored in the status storage unit (20); an input unit (22) which is set up to allow input of recovery process information at the cell control when each of the manufacturing machines (14, 15, 16) stops; a recovery operation information storage unit (23) configured to store the recovery operation information from each of the manufacturing machines (14, 15, 16); and a correlation data generation unit (24) which: - is set up to generate correlation data obtained by correlating the fault information acquired by the fault information acquisition unit (21) with the recovery operation information stored by the recovery operation information storage unit (23), and is set up to transfer this to the database, and - a learning unit (27) comprising a learning unit configured to learn the correlation data using the alarm activation rate in or the operating rate of each of the manufacturing machines (14, 15, 16) as a reward, wherein the learning unit (27) is configured to optimize the correlation data by allocating the reward such that the suitability of a recovery operation for the manufacturing machine (14, 15, 16) increases when the detected alarm activation rate decreases or the operating rate increases; and a data discrimination unit (30) which is configured to compare the recovery operation information output by the input unit (22) to the recovery operation information storage unit (23) with the multitude of correlation data accumulated in the database, whereupon, If the data differentiation unit (30) finds a recovery operation information similar to the recovery operation information in the multitude of correlation data accumulated in the database, the data differentiation unit (30) is configured to send the similar recovery operation information and the fault information associated with the recovery operation information to the output unit (26).

Description

BACKGROUND OF THE INVENTION

[0001] The present invention relates to a cell control system for controlling a manufacturing cell and a production system for managing the working situation of a plurality of manufacturing machines in the manufacturing cell.

[0002] In a machine factory, a production cell, such as a production line, consists of a variety of production machines, like machine tools or robots. When such a production cell is used to manufacture products, a decrease in the operating rate of the production machines reduces the production volume. Therefore, if a problem occurs with one of the production machines in the cell, an operator must identify the cause in order to get the machine back up and running as quickly as possible. However, if the operator is not a specialist, it can take a considerable amount of time to find a suitable way to deal with the problem, thus delaying the machine's recovery.

[0003] JP 2 934 026 B2 discloses a production system that identifies a defective manufacturing machine while referring to operational data from each manufacturing machine, derives a procedure for dealing with the defective manufacturing machine, and performs an automatic correction based on the derivation.

[0004] JP 4 873 267 B2 discloses a production system that analyzes the previous state of a manufacturing machine when the manufacturing machine stops in order to classify it into a variety of types, and that displays information for identifying the state of the manufacturing machine on a screen according to each type.

[0005] JP 5 436 460 B2 discloses a production system equipped with a teaching unit that automatically teaches a troubleshooting process for a robot that is in a fault condition, based on an operational history of teaching the robot as a manufacturing machine, and a library unit for storing the content of the troubleshooting process.

[0006] However, the production system disclosed in JP 2 934 026 B2 lacks a function for storing the procedure for handling a defect in the production machine as data and updating it. Therefore, in the production system disclosed in JP 2 934 026 B2, even if a defect similar to the previous one occurs in the production machine, the procedure for handling the defect must be derived again.

[0007] The production system disclosed in JP 4 873 267 B2 merely provides an operator with information specifying the state of the stopped production machine and lacks a function for storing and outputting the contents of a troubleshooting procedure performed for a malfunction of the stopped production machine. Thus, if the operator has difficulty finding a procedure to deal with the reported state of the production machine, the machine's recovery is delayed.

[0008] The production system disclosed in JP 5 436 460 B2 is designed for an operator-free environment and lacks a function to inform the operator, when a fault occurs in the production machine, about the most appropriate troubleshooting procedure for that fault. Therefore, in the production system disclosed in JP 5 436 460 B2, the recovery of the production machine is delayed in some cases involving tasks requiring operator intervention.

[0009] German patent application DE 102 30 895 A1 discloses a transaction data communication system and a method that communicates information within a company with a process control system and a plurality of information technology systems, which are communicatively coupled to the process control system via a web service interface with a transaction information server. The system and method generate transaction process control information and format this information based on an extensible markup language input schema to create formatted transaction process control information.The system and procedure send the formatted transaction process control information to the transaction information server via the web services interface and map the formatted transaction process control information to an extensible markup language output schema, which is assigned to one of the majority of information technology systems, to form mapped transaction process control information. The system and procedure send the mapped transaction process control information to one of the majority of information technology systems.

[0010] The patent application US 2011 / 0 264 440 A1 discloses methods and devices for displaying localized resources in a process control system. The method comprises receiving a locale identifier and a resource identifier from an application in a server, identifying a language file in a database corresponding to the locale identifier via the server, determining a resource contained in memory that corresponds to the resource identifier via the server using the language file, accessing the resource from memory via the language file, and sending the resource via the server to the application to display the resource in a format associated with the locale identifier.

[0011] The Wikipedia™ entry “User Guide” as of September 16, 2015 reveals a user manual that includes a dedicated troubleshooting chapter. SUMMARY OF THE INVENTION

[0012] The present invention provides a cell control and production system that enables an operator to easily determine the most suitable recovery process for the manufacturing machine when a manufacturing machine stops in a manufacturing cell.

[0013] According to the disclosure, a cell control system and a production system are provided according to the independent claims. Developments are described in the dependent claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] These tasks, features and advantages of the present invention and further tasks, features and advantages will become more apparent from the detailed description of the typical embodiments shown in the accompanying drawings. Fig. Figure 1 shows a block diagram of the configuration of a production system according to one embodiment. Fig. 2 shows a view of an example configuration of from a in Fig. Correlation data generated by the correlation data generation unit shown in section 1 and stored in a database on a host computer. Fig. Figure 3 shows a view of another embodiment of correlation data. Fig. Figure 4 shows a flowchart of a process for generating correlation data, encompassing the fault information and troubleshooting information of a manufacturing machine and storing the correlation data in a database of a host computer. Fig. Figure 5 shows a flowchart of a process for retrieving troubleshooting process information according to the fault information of a manufacturing machine from the database of a host computer. Fig. Figure 6 shows a block diagram of the configuration of a first modification of the in Fig. 1 production system shown. Fig. Figure 7 shows a block diagram of the configuration of a second modification of the in Fig. 1 production system shown. Fig. Figure 8 shows a block diagram of the configuration of a third modification of the in Fig. 1 production system shown. Fig. Figure 9 shows a schematic diagram of a neuron model. Fig. Figure 10 shows a schematic diagram of a three-layer model of a neural network. DETAILED DESCRIPTION

[0015] The following are embodiments of the present invention explained with reference to the drawings. In all figures, similar elements or functional elements are designated by the same reference numerals. The figures have been scaled accordingly to facilitate understanding. Furthermore, the embodiments shown in the drawings are examples of carrying out the present invention, and the present invention is not limited to the embodiments shown.

[0016] Fig. Figure 1 shows a block diagram of an embodiment of a production system according to the present invention.

[0017] As in Fig. Figure 1 shows a production system 10 with at least one manufacturing cell 11, a cell control 12 for controlling the manufacturing cell 11 and a host computer 13 equipped with a database.

[0018] The manufacturing cell 11 is located in a factory for the production of goods. The cell controller 12 and the host computer 13, however, are located in a building separate from the factory. For example, the cell controller 12 could be located in a different building at the same factory site as the manufacturing cell 11. In this case, the manufacturing cell 11 and the cell controller 12 are preferably connected so that they can communicate via a communication device 18, for example, an intranet.

[0019] The host computer 13 can, for example, be located in an office outside the factory. In this case, the cell controller 12 and the host computer 13 are preferably connected so that they can communicate via a communication device 17, for example, the internet. According to the present embodiment, the host computer 13 is preferably a computer located in the office that is a production planning device for managing the work status of a plurality of manufacturing cells 11 or manufacturing machines.

[0020] Manufacturing cell 11 is a group created by flexibly combining a variety of manufacturing machines to produce products. Manufacturing cell 11 includes, as described in... Fig. Figure 1 shows a first production machine 14, a second production machine 15, and a third production machine 16; however, the number of production machines in the production cell 11 is not limited. The production cell 11 can also be a production line in which a workpiece is successively processed by a plurality of production machines to produce a finished product. Alternatively, the production cell 11 can be a production line in which two or more workpieces (parts) processed in each of the two or more production machines are combined by another production machine in the middle of the process to produce a finished product. Alternatively, in the invention of this application, two or more workpieces processed in two or more production cells 11 can be combined to produce a finished product.

[0021] Each of the manufacturing machines 14 to 16 is, for example, an NC machine tool or an industrial robot. Of course, each manufacturing machine used in the present invention is not limited to an NC machine tool or an industrial robot. Examples of each manufacturing machine could include a PLC, a transfer machine, a measuring instrument, a press machine, a press fitting machine, a printing press, a die-casting machine, an injection molding machine, a food processing machine, a packaging machine, a welding machine, a washing machine, a painting machine, an assembly machine, a woodworking machine, a sealing device, or a cutting machine.

[0022] Furthermore, the cell controller 12 and the production machines 14 to 16 are equipped with computer systems (not shown) comprising CPUs, memory such as ROMs or RAMs, and communication control units connected via a bus line. These communication control units manage the control data exchanged between the cell controller 12 and the production machines 14 to 16. Preferably, the functions or processes of the cell controller 12 and the production machines 14 to 16 are effected by programs stored in the ROMs, which are executed by the corresponding CPUs.

[0023] The structure of cell control 12 is described in detail below.

[0024] As in Fig. As shown in Figure 1, the cell controller 12 consists of sensors 19, a state storage unit 20, a fault information acquisition unit 21, an input unit 22, a recovery process information storage unit 23, a correlation data generation unit 24, a recovery process information retrieval unit 25, and an output unit 26. The functions of these components are described below in sequence.

[0025] Each sensor 19 is arranged in the corresponding machine of the production machines 14 to 16 and determines the state of the corresponding machine of the operating production machines 14 to 16 at a predetermined time interval. The state storage unit 20 sequentially stores the states of the production machines 14 to 16 that are determined by the sensors 19 at a predetermined time interval. Preferably, a plurality of sensors 19 are arranged in each production machine to determine different states of each of the production machines 14 to 16. Furthermore, the cell controller 12 is preferably designed to simultaneously process the state information obtained from the sensors 19 arranged in the production machines.

[0026] The fault information acquisition unit 21 acquires the fault information from each production machine from the status information of each of the production machines 14 to 16, which is stored in the state storage unit 20. The fault information acquisition unit 21 also transmits the fault information to the correlation data generation unit 24 after receiving it from the latter.

[0027] Examples of state information include information on the value of the current supplied to a motor for driving a robot arm (if the production machines are 14 to 16 robots), information on the output of a position sensor attached to the motor, etc. In this case, the fault information is a group of information comprising a variety of information pieces relating to, for example, the reduction in the motor's current value when the robot stops due to faults, and a fault in the position sensor output. Preferably, such fault information includes not only the state information of a production machine at the time the production machine stops, but also the state information of the production machine during a predetermined period before the time the production machine stops.This allows the fault information to include a predictive state of the manufacturing machine, indicating that it will stop due to a fault.

[0028] The input unit 22 has a function for entering recovery operation information into the cell control 12. This information represents the recovery operations performed for stopping production machines 14 to 16 due to a malfunction. The recovery operation information consists, for example, of operation codes in a list of recovery operations. The list of recovery operations is a table describing the correlation between a variety of different recovery operations and operation codes, each consisting of a multi-digit number representing the corresponding recovery operation. Examples of input devices for the input unit 22 for entering such operation codes include a keyboard, a touchscreen, etc. The input unit 22 can be located in any of the production machines 14 to 16.In this case, preferably a teaching module for performing a teaching process for a robot or a control module of an NC machine tool is used as the input unit 22.

[0029] The recovery process information storage unit 23 stores the recovery process information entered by the input unit 22 and transmits it to the correlation data generation unit 24. The correlation data generation unit 24 generates correlation data to represent the correlation between the fault information acquired by the fault information acquisition unit 21 and the recovery process information stored in the recovery process information storage unit 23 in each of the production machines 14 to 16.

[0030] When the correlation data is generated, the identification data (ID) of each production machine, the fault information transmitted by the fault information acquisition unit 21, and the recovery process information transmitted by the recovery process information storage unit 23 after the fault information are preferably correlated with each other. Alternatively, the correlation data generation unit 24 generates the correlation data, while a learning unit 27 is caused to learn this data as described in a first modification (see Fig. 6), which is described later. The correlation data generated as described above are transferred from the correlation data generation unit 24 to the host computer 13 and accumulated in the database of the host computer 13.

[0031] The recovery process information retrieval unit 25 retrieves the recovery process information from the database of the host computer 13, based on the fault information acquired by the fault information acquisition unit 21. The output unit 26 outputs the recovery process information retrieved by the recovery process information retrieval unit 25, together with the fault information acquired by the fault information acquisition unit 21. Examples of the output unit 26 include a display device for showing recovery processes based on the recovery process information, or a printing device for printing the recovery processes based on the recovery process information on a sheet of paper. The output unit 26 can be located in any of the manufacturing machines 14 to 16.

[0032] Fig. Figure 2 shows an example of the configuration of the correlation data generated by the correlation data generation unit 24 as described above and accumulated in the database of the host computer 13. As in Fig. Figure 2 shows the correlation data generated based on the correlation between the fault information of each production machine and the recovery process information in the ID of each production machine. Such correlation data is considered a Fig. The data table shown in section 2 is configured, and the correlation between the ID of each manufacturing machine, the fault information, and the recovery process information is performed in the corresponding row of the data table.

[0033] In particular, the identification numbers (e.g., WF0005, RD002, SF011) of production machines 14 to 16 are entered into the table of production machine IDs in the Fig. The correlation data shown in Figure 2 is written. “WF005” is the identification number of the first production machine 14, which is a transfer robot. “RD002” is the identification number of the second production machine 15, which is a drilling robot. “SF011” is the identification number of the third production machine 16, which is a screwing robot.

[0034] An information group or data column comprising a multitude of state information pieces to represent a multitude of states (state 1, state 2, etc.) detected by the sensors 19 in the production machines are entered into the fault information table in the Fig. 2 correlation data shown. Furthermore, an information group or data columns comprising a variety of recovery operation information to represent a variety of recovery operations (Operation 1, Operation 2, etc.) that are performed when each manufacturing machine is recovered are written to the recovery information table in the Fig. The correlation data shown in Figure 2 is written. For example, in the transfer robot with the manufacturing machine ID "WF005", the fault information is a group of information comprising a variety of state information, such as a decrease in motor current or a fault in the position sensor output. The recovery process information, corresponding to the stop process information, is a group of information comprising a variety of recovery process information, such as removing a workpiece or adjusting a force sensor. That is, in the transfer robot "WF005", the state information representing a decrease in motor current (state 1), the state information representing a fault in the position sensor output (state 2), and so on, are simultaneously recorded as fault information for the transfer robot.Based on these parts of the fault information, an operator performs a variety of recovery operations (Operation 1, Operation 2, etc.) such as removing a workpiece, adjusting a force sensor, etc., and all parts of the recovery information representing these recovery operations are entered into input unit 22. Thus, a variety of parts of the status information representing status 1, status 2, etc., are entered into the input unit. Fig. The table shown in Figure 2 contains fault information. Furthermore, a variety of recovery process information is entered into the table to represent Process 1, Process 2, etc. Fig. 2. Table of recovery process information shown.

[0035] Fig. Figure 3 shows another mode of correlation data. A state in which stopping is possible (stop state) is extracted from the state of each production machine, and the correlation between a cause for each stop state and a corresponding recovery operation is established. In this mode, the stop state occurring in a specific production machine and the recovery operation can be applied to another machine in a different line.

[0036] In Fig. Figure 2 shows that each part of the state information contained in the fault information is written using language; however, for example, error codes consisting of multi-digit numbers to represent different types of states that can occur in any production machine can also be written instead of language. Regarding the recovery process information, process codes consisting of multi-digit numbers, which are entered by the input unit 22, are preferably used instead of the ones shown in Figure 2. Fig. 2 described language written.

[0037] The typical operation of the cell controller 12 according to the present embodiment is described below.

[0038] The cell controller 12 according to the present embodiment stores the correlation data, comprising the fault information and the recovery process information, in the database of the host computer 13. This process is described below with regard to Fig. 1 and Fig. 4 described.

[0039] Fig. Figure 4 shows a flowchart of the process for generating the correlation data, including the fault information and the recovery process information, and for storing this data in the database of the host computer 13.

[0040] The cell controller 12 controls the production machines 14 to 16 in the production cell 11, thus initiating product production. During product production, the states of the production machines 14 to 16 are determined by the sensors 19 at predefined time intervals and stored sequentially in the state storage unit 20 (step S11).

[0041] Subsequently, the fault information acquisition unit 21 determines in real time whether a specific production machine has stopped due to a fault from the state information of each of the production machines 14 to 16, which is stored in the state storage unit 20 (step S12). For example, the fact that the output of the sensors 19 mounted in a specific production machine is lower than a predefined value causes the production machine to stop due to a detected fault. Through this determination, the fault information acquisition unit 21 retrieves the fault information of the production machine that stopped due to a fault from the state information of each of the production machines 14 to 16, which is stored in the state storage unit 20 (step S13).

[0042] The system then determines whether the recovery process information has been entered from input unit 22 to recovery process information storage unit 23 (step S14). Step S14 only proceeds to the next step if it is determined that the recovery process information has been entered in step S14. Thus, if the fault information acquisition unit 21 acquires the fault information from the production machine as described above, the cell control device 12, for example, causes a light source device or a sound source device to execute an external alarm due to the missing input of the recovery process information.

[0043] If it is determined that recovery process information was entered in step S14, the correlation data generation unit 24 generates correlation data (step S15). The correlation data is a set of data generated by correlating the identification information of the stopped production machine, the fault information as described previously, and the entered recovery process information. The correlation data generation unit 24 then transmits the correlation data to the host computer 13, where it is stored in the host computer 13's database (step S16).

[0044] Step S17 then determines whether a stop command has been issued to cell controller 12. If a stop command has been issued, the process for generating and storing the correlation data ends. If, however, no stop command has been issued, the process returns from step S17 to step S11, and the processes in steps S11 to S16 are executed again. Repeating these processes results in the correlation data being accumulated in the database of host computer 13.

[0045] Furthermore, according to the present embodiment, the cell controller 12 can, upon acquiring the fault information, retrieve the recovery process information corresponding to the fault information from the database of the host computer 13 and report it to an operator. This process is described below with regard to Fig. 1 and Fig. 5 described.

[0046] Fig. Figure 5 shows a flowchart of the procedure for retrieving the recovery process information according to the fault information from the database of the host computer 13 and reporting this information.

[0047] The in Fig. The 5 steps S21 to S23 shown are identical to steps S11 to S13 in Fig. 4. Thus, executing steps S21 to S23 results in the acquisition of fault information from the production machine, which stops due to a fault.

[0048] Subsequently, the recovery process information retrieval unit 25 retrieves the recovery process information from the database of the host computer 13, based on the fault information captured in step S23 (step S24). Then, the output unit 26 outputs the recovery process information retrieved by the recovery process information retrieval unit 25 (step S25). In the output unit 26, the fault information of the production machine that was stopped and the recovery process information, based on the fault information, are printed on paper or displayed on a screen, for example. The processes in steps S24 and S25 can be performed between steps S13 and S14 in Fig. 4 will be executed.

[0049] Step S26 then determines whether a stop command has been issued to cell controller 12. If a stop command has been issued, the process for instructing output unit 26 to output the recovery process information ends. If, however, no stop command has been issued, the process returns from step S26 to step S21, and the processes in steps S21 to S26 are executed again.

[0050] As previously described, the cell controller 12, according to the present embodiment, is configured to generate correlation data obtained in each production machine by correlating the fault information representing the state of a production machine that has stopped due to a fault with the recovery process information of the production machine. Such correlation data is transferred to and accumulated in the database of the host computer 13 as soon as the correlation data is generated. Thus, if a particular production machine stops due to a fault, the operator can extract the recovery process information from the database of the host computer 13, representing the content of the previously performed recovery process for stopping that particular production machine.This means that the cell control 12 according to the present embodiment enables the operator to easily grasp the content of the most suitable recovery process for the fault information of each manufacturing machine in the manufacturing cell 11.

[0051] In particular, in the cell control 12 according to the present embodiment, when fault information from a production machine is detected, the recovery process information corresponding to the fault information of the production machine is retrieved from the database of the host computer 13 and sent to the output unit 26. This enables the operator, especially a layperson, to quickly grasp the content of the recovery process for the production machine that has stopped and to carry out the recovery process accordingly.

[0052] The following describes a first modification of the cell control 12. Fig. Figure 6 shows a block diagram of the configuration of the first modification of the in Fig. 1. Cell control shown 12.

[0053] As in Fig. Figure 6 shows a cell controller 12A in its first modification compared to the one in Fig. The cell control unit 12 shown in Figure 1 is further equipped with a learning unit 27, a production performance information storage unit 28, and a calculation unit 29. The functions of these components are described below in sequence.

[0054] In the Fig. In the cell controller 12A shown in Figure 6, the learning unit 27 is arranged in the correlation data generation unit 24. The learning unit 27 learns the correlation data generated by the correlation data generation unit 24 using the alarm trigger rate or operating rate of each of the manufacturing machines 14 to 16 as a reward.

[0055] Specifically, after the recovery process has been performed for the production machine that stopped due to a fault, learning unit 27 records the alarm trigger rate or the operating rate over a predefined period from the time the production machine was recovered. It can be determined that the suitability of the recovery process for the production machine increases when the recorded alarm trigger rate decreases or the operating rate increases. Thus, learning unit 27 optimizes the correlation data by assigning a reward for the correlation data encompassing the fault information and the recovery process information, according to the alarm trigger rate in or the operating rate from the production machine.

[0056] The alarm trigger rate or operating rate is recorded by the production performance information storage unit 28 and the calculation unit 29. That is, the production performance information storage unit 28 stores the production performance information for each of the production machines 14 to 16, which are operated according to the control commands from the cell controller 12A. The calculation unit 29 calculates the alarm trigger rate or operating rate of each of the production machines 14 to 16 based on the production performance information stored in the production performance information storage unit 28.

[0057] Production performance information is used, for example, to perform quality control of the products manufactured in production cell 11 or for process control. Examples of production performance information include current values ​​of motors of production machines during production, times required for production, alarm information generated during production, programs or parameter values ​​used for production, identification numbers of tools used for production, ambient temperatures during production, production defects measured by measuring instruments after production, etc.

[0058] In the present embodiment, the production performance information includes the name of the production machines, production time, the number of workpieces processed, the number of alarms, etc. The identification numbers specific to the production machines are used as the names of the production machines. The production time is the time required by a specific production machine to process a workpiece. The number of workpieces processed is the number of workpieces processed by a specific production machine. The number of alarms is the number of alarms generated while a specific production machine is processing workpieces. The output of, for example, a sound source or light source located in each production machine to signal a malfunction of the corresponding production machine is used as an alarm.

[0059] In the computation unit 29, for example, the alarm activation rate is calculated as follows. The alarm activation rate can be determined by dividing the number of alarms by the production time or the number of production machines in the production cell 11. Alternatively, the alarm activation rate can be determined by dividing the time during which a production machine stops and resumes operation due to an alarm activation by the time during which the production machine should be substantially in operation. Alternatively, the time during which a production machine stops and resumes operation due to an alarm activation can be treated directly as the alarm activation rate. These methods for calculating the alarm activation rate are examples, and the present invention is not limited to these methods. The computation unit 29 can therefore calculate the operating rate instead of the alarm activation rate.The operating rate is determined by dividing the time a production machine is actually in operation by the time the production machine should essentially have been in operation.

[0060] According to the first modification described above, learning unit 27 can improve the relevance between the fault information and the recovery process information for each manufacturing machine in the correlation data accumulated in the database of the host computer 13.

[0061] Examples of the state storage unit 20, the recovery operation information storage unit 23, and the production performance information storage unit 28 include storage devices such as RAMs (Random Access Memories). Alternatively, examples of the state storage unit 20, the recovery operation information storage unit 23, and the production performance information storage unit 28 include fixed drives such as hard disks or portable storage devices such as floppy disks, optical storage discs, etc.

[0062] A second modification of the cell control 12 is described below. Fig. Figure 7 shows a block diagram of the configuration of the second modification of the in Fig. 1 cell control shown.

[0063] As in Fig. Figure 7 shows a cell controller 12B in the second modification compared to the one in Fig. The cell controller 12A shown in Figure 6 is further equipped with a data differentiation unit 30 and a learning unit 31. The functions of these components are described below in sequence.

[0064] The data differentiation unit 30 compares the recovery operation information entered by the input unit 22 into the recovery operation information storage unit 23 with a variety of correlation data accumulated in the database of the host computer 13, as described previously. If the data differentiation unit 30 finds recovery operation information similar to the entered recovery operation information in the correlation data in the database, the data differentiation unit 30 sends the similar recovery operation information and the associated fault information to the output unit 26. The recovery operation information and the fault information are preferably printed on paper or displayed on a screen in the output unit 26 to communicate them to the operator.

[0065] In the second modification of cell control 12B, the operator predicts the content of a recovery operation for the stopped production machine and inputs the recovery operation information from input unit 22 to recovery operation information storage unit 23. If the recovery operation information in the database of host computer 13 is similar to the input recovery operation information, the similar recovery operation information and the associated fault information are output by output unit 26. This allows the operator to know the stopped state of the production machine for which a recovery operation similar to the predicted recovery operation is being performed.This means that the second modification of the 12B cell controller allows the operator to verify whether the predicted recovery process is suitable for the current stop state of the production machine. Furthermore, the operator can easily predict a more suitable recovery process from the fault information of the production machine, for which the recovery process information is generated similar to the predicted recovery process.

[0066] A third modification of cell control 12 is described below. Fig. Figure 8 shows a block diagram of the configuration of the third modification of the in Fig. 1 cell control shown.

[0067] As in Fig. Figure 8 shows a cell controller 12C, similar to the second modification, in the third modification compared to the one in Fig. The cell controller 12A shown in Figure 6 is further equipped with a data differentiation unit 30 and a learning unit 31. However, as described below, the functions of these components differ from those in the second modification.

[0068] As in Fig. As shown in Figure 8, the data differentiation unit 30 compares the fault information recorded by the state storage unit 20 via the fault information acquisition unit 21 with the correlation data accumulated in the database of the host computer 13, as described previously. The data differentiation unit 30 finds fault information similar to the recorded fault information in the correlation data in the database and outputs the similar fault information to the output unit 26. The data differentiation unit 30 also retrieves recovery process information associated with the similar fault information from the database and outputs it to the output unit 26. The recovery process information and the fault information are preferably printed on paper or displayed on a screen by the output unit 26 to communicate them to the operator.Alternatively, similar fault information can be found before the actual stopping occurs, meaning that a prediction of the stopping is permissible.

[0069] In the third modification of the cell controller 12C, a state similar to the fault information can be distinguished from the state information of a production machine in the state storage unit 20. The recovery process information associated with the similar fault information can be read from the database of the host computer 13. Furthermore, the output unit 26 outputs the similar fault information and the associated recovery process information. This allows the operator to ascertain the state of the production machine, similar to the specific fault information, i.e., an indication of a stoppage. Thus, the third modification of the cell controller 12C can predict a possible stoppage of a production machine based on its state.For example, the operator can prevent a stoppage or prepare a recovery process by previously capturing the predicted fault information.

[0070] However, if the accuracy of the similarity discrimination in the data discrimination units 30 in the second and third modifications is low, the fault information associated with a recovery operation that differs significantly from the recovery operation predicted by the operator will be unexpectedly output. In this case, the operator may misunderstand that the predicted recovery operation is suitable. Thus, the learning unit 31 is preferably arranged in each data discrimination unit 30. That is, the learning unit 31 of the cell controller 12B in the second modification is preferably configured to learn a criterion with respect to which the similarity between the recovery operation information contained in each of the correlation data in the database and the recovery operation information entered in the recovery operation information storage unit 23 is distinguished.In contrast, the learning unit 31 of the cell controller 12C in the third modification is preferably configured to learn a criterion with respect to which the similarity between the fault information contained in each of the correlation data in the database and the stop information of the production machine, which was entered into the state storage unit 20, is distinguished. In particular, these learning units 31 learn such similarity discrimination criteria as, for example, the concordance rate of data using the alarm triggering rate or operating rate of each of the production machines 14 to 16 as a reward.

[0071] In particular, after the operator performs the predicted recovery process as described above to restore a production machine, the learning unit 31 records the alarm activation rate or the operating rate over a predetermined period from the time the production machine was restored. It can be determined that the suitability of the predicted recovery process increases as the recorded alarm activation rate decreases or the operating rate increases. Thus, the learning unit 31 learns a criterion best suited for distinguishing similarity in the data discrimination unit 30 by assigning a reward corresponding to the alarm activation rate in or operating rate of the production machine for the similarity discrimination criterion, for example, the data concordance rate.Furthermore, similar to the first modification, the alarm trigger rate or the operating rate is preferably recorded with the production performance information storage unit 28 and the calculation unit 29.

[0072] As previously described, in cell control 12B in the second modification and cell control 12C in the third modification, the distinction criterion for similarity in the data distinction unit 30 is optimized by arranging the learning unit 31 in the data distinction unit 30.

[0073] The following describes learning units 27 and 31 (hereinafter referred to as machine learning devices). The machine learning device has a function for analytically extracting useful rules or knowledge representations, criteria for determination, etc., from the set of data inputs to the device, and a function for outputting the results of the extraction and the learned knowledge. There are various machine learning methods, and these methods are broadly categorized as "supervised learning," "unsupervised learning," and "reinforcement learning." To achieve these learning outcomes, there is another method for learning how to extract the feature set itself, which is called "deep learning."

[0074] "Supervised learning" is a process in which a large volume of input-output (key) pair data is fed to a machine learning device so that features of this dataset can be learned, and a model for deriving an output value from input data—that is, the input-output relationship—can be inductively derived. This can be achieved with an algorithm, such as a neural network, which will be described later.

[0075] Unsupervised learning is a method in which a large volume of input data is fed to a machine learning device so that the distribution of the input data can be learned. The device can then be trained, for example, to compress, classify, and clean the input data, even without the corresponding training output data. For instance, the features of this data set can be grouped based on their similarity. The result obtained from the learning process is used to define a specific criterion, and then the output is assigned to optimize the criterion, thus predicting the output. There is another problem-solving method that lies between unsupervised and supervised learning and is called semi-supervised learning.This learning method provides a small volume of input-output pair data and a large volume of input data only.

[0076] Problems are solved in reinforcement learning as follows.

[0077] A machine learning device observes the state of the environment and decides on an action.

[0078] The environment varies within certain rules, and a user action can change the environment.

[0079] A reward signal is provided for each action.

[0080] The goal of maximization is the sum of (deduction) rewards to be received now and in the future.

[0081] Learning begins when the outcome of an action is either completely unknown or only partially known. The machine learning device can only capture the outcome as data after the actual start of operation. This means the optimal action must be found through trial and error.

[0082] Alternatively, the initial state can be defined as the state in which prior learning (for example, the previously mentioned supervised learning or inverse reinforcement learning) is carried out to emulate the action of a person, and learning can be started at a suitable starting point.

[0083] Reinforcement learning is a learning method for learning not only investigations or classifications, but also actions, whereby a suitable action is learned based on the interaction of the environment with that action—that is, an action to maximize rewards to be obtained in the future. In the present invention, this means that an action that can have an effect on the future can be identified. The explanation of reinforcement learning continues below with, for example, Q-learning; however, reinforcement learning is not limited to Q-learning.

[0084] Q-learning is a process for learning a value Q(s, a) in which an action a is selected in a given environmental state s. This means that it is only necessary to select the action a with the highest value Q(s, a) as the optimal action a in a given state s. Initially, however, the correct value of Q(s, a) for any combination of state s and action a is completely unknown. Subsequently, the agent (the subject of an action) selects various actions a in a given state s and distributes rewards to the actions a at that time. Thus, the agent learns to select a more advantageous action, that is, the correct value Q(s, a).

[0085] One desired outcome of the action is to maximize the sum of future rewards; accordingly, Q(s, a)=E[Σγ] is ultimately calculated. t[rt] is sought. (An expected value is set for the time period when the state varies according to the optimal action. The expected value is, of course, unknown and must be learned accordingly while it is being sought.) The update expression for such a value Q(s, a) is given, for example, by: O˜(st,at)←Q(st,at)+α(rt+1+γmaxaQ(st+1,a)→Q(st,at))

[0086] This is t the state of the environment at time t, and a t The action occurs at time t. During the action at, the state changes to s. t+1 r t+1 is the reward to be received when the state changes. The expression to which "max" is appended is obtained by multiplying the Q-value obtained when the action a with the highest Q-value is performed in state s at that time. t-1The selected parameter is obtained with γ. y is the parameter with a range of 0 < γ ≤ 1 and is called the discount rate. α is the learning factor and has a range of 0 < α ≤ 1.

[0087] This equation expresses a procedure for updating a valuation value Q (s t , a t ) of an action at in a state s t based on a reward delivered as a result of an attempt r t+1 out. If a valuation value Q(s t+1 , max a t+1 ) of the optimal action max a in a subsequent state, caused by the reward r t+1 + the action a is greater than the evaluation value Q(s) t , a t If the action a in state s is performed, Q(st, at) is increased. Conversely, if the evaluation value Q(s) is increased, Q(st, at) is increased. t+1 , max a t+1If the value of Q(st, at) is less than the evaluation value Q(st, at), Q(st, at) is reduced. This means that an attempt is made to approximate the value of a specific action in a given state, the reward immediately delivered as a result, and the value of an optimal action in the subsequent state caused by that specific action.

[0088] Examples of methods for expressing Q(s, a) on a computer include a method for obtaining the values ​​of all state-action pairs (s, a) as a table (action value table) and a method for preparing a function to approximate Q(s, a). In the latter method, the previously described update equation can be obtained by fitting a parameter of the approximation function using a method such as the stochastic gradient method. Examples of the approximation function include a neural network, which is described later.

[0089] A neural network can be used as an approximation algorithm for a value function in supervised learning, unsupervised learning, and reinforcement learning. The neural network consists, for example, of an arithmetic device and memory, which a neural network uses to simulate a neural model, such as in Fig. 9 realize. Fig. Figure 9 shows a schematic diagram to represent a neuron model.

[0090] As in Fig. Figure 9 shows a neuron outputting an output y in response to a variety of inputs x (here, for example, inputs x1 to x3). Weights w (w1 to w3) are applied to the corresponding inputs x1 to x3. This causes the neuron to output the output y, which is expressed by the preceding equation. The inputs x, the output y, and the weights w are vectors. y=fk(∑i=1nxiwi−θ)

[0091] where θ is the distortion and fk This is the activation function.

[0092] A three-layer weighted neural network consisting of a combination of neurons as previously described is subsequently compared with respect to Fig. 10 described. Fig. Figure 10 shows a schematic diagram illustrating a weighted neural network with three layers D1 to D3.

[0093] As in Fig. Figure 10 shows a multitude of inputs x (for example, input x1 to input x3) being entered on the left side of the neural network and results y (for example, result y1 to result y3) being output on the right side of the neural network.

[0094] Specifically, the inputs x1 to x3, to which the corresponding weights have been applied, are entered into the three neurons N11 to N13. These weights applied to the inputs are collectively referred to as w1.

[0095] Neurons N11 and N13 each output z11 to z13. z11 to z13 are collectively referred to as a feature vector z1 and can be treated as a vector obtained by extracting a feature set from an input vector. This feature vector z1 is a feature vector between the weights w1 and w2.

[0096] The feature vectors z11 to z13, to which the corresponding weights have been applied, are input to two neurons N21 and N22. These weights applied to the feature vectors are collectively referred to as w2.

[0097] Neurons N21 and N22 each output z21 and z22, respectively. z21 and z22 are jointly represented by a feature vector z2. This feature vector z2 is a feature vector between weight w2 and weight w3.

[0098] The feature vectors z21 and z22, to which the corresponding weights have been applied, are input to the three neurons N31 to N33. These weights applied to the feature vectors are collectively referred to as w3.

[0099] Finally, neurons N31 to N33 each output the results y1 to y3.

[0100] The operation of the neural network comprises a learning mode and a value prediction mode. A training dataset is used to learn the weights w in the learning mode, and parameters obtained from this learning are used to determine the action of the processing engine in the prediction mode. (For simplicity, the term "prediction" is used here; however, various tasks can be performed, including acquiring, classifying, deriving, etc.)

[0101] Learning can take two forms: online learning, where data collected through the actual operation of the processing machine is directly learned in prediction mode and mirrored in a subsequent action, and batch learning, where previously collected data is learned together with a group of data, followed by a data acquisition mode with parameters obtained from the learning process. An additional learning mode can be interposed each time a predetermined amount of data has been collected.

[0102] The weights w1 to w3 can be learned through an error feedback process. Error information is introduced from the right side to the left. The error feedback process is a method for adjusting (learning) each weight to reduce the difference between the output y when input x is applied and the actual output y (teacher) in each neuron.

[0103] In such a neural network, three or more layers can be provided. (This is called "deep learning".) An arithmetic device that incrementally extracts features from input data to produce a result can be automatically generated from teacher-only data.

[0104] Reinforcement learning, a machine learning method such as Q-learning, is applied to the in Fig. 6 shown learning unit 27 and the one in Fig. 7 and Fig.The machine learning method applicable to learning units 27 and 31 is not limited to Q-learning. For example, when supervised learning is applied to learning units 27 and 31, the value functions correspond to learning models and the rewards correspond to errors. IMPACT OF THE INVENTION

[0105] According to one aspect of the present invention, saving data when a manufacturing machine stops in the production cell, including a recovery process best suited to the manufacturing machine, and updating the recovery process to an optimal state, enables the operator to easily grasp the content of the recovery process. According to another aspect of the present invention, the operator, particularly a layperson, can quickly and appropriately perform the recovery process for the stopped manufacturing machine.

[0106] According to a further aspect of the present invention, the learning unit can improve the relevance between the fault information and the recovery process information of each manufacturing machine in the correlation data accumulated in the database of the host computer.

[0107] According to a further aspect of the present invention, the operator can, based on the recovery process information output by the correlation data and the corresponding fault information of the manufacturing machine, confirm whether the predicted recovery process is suitable for restoring the manufacturing machine or not.

[0108] According to a further aspect of the present invention, the operator can predict a stoppage of a production machine and prepare a recovery process for this based on the state information of the production machine and the fault information of the production machine, which is similar to the state information.

[0109] According to a further aspect of the present invention, arranging the learning unit in the data discrimination unit optimizes the similar discrimination criterion in the data discrimination unit.

Images