WO2021199383A1

WO2021199383A1 - Machine learning system

Info

Publication number: WO2021199383A1
Application number: PCT/JP2020/015065
Authority: WO
Inventors: 林　英松; 督那須; 紀之尾崎; 僚柏木
Original assignee: 三菱電機株式会社
Priority date: 2020-04-01
Filing date: 2020-04-01
Publication date: 2021-10-07
Also published as: JPWO2021199383A1; JP6884287B1

Abstract

A machine learning system is provided with a PLC (4) which is a control device that controls production equipment, an IPC (2) which is an information processing device, and an AI unit (5) which is an inference device. The IPC (2) has a data collection unit (14) which collects state data indicating an operating state of the production equipment, a data processing unit (16) which processes the collected state data, a learning unit (17) which generates a learned model by means of machine learning using the state data which has undergone processing, and an output processing unit (18) which generates a program for carrying out collection and processing of the state data and outputs a learned model. The AI unit (5) outputs, to the PLC (4), inference results based on the learned model when the state data which has undergone collection and processing by execution of the program is input.

Description

Machine learning system

This disclosure relates to a machine learning system that implements machine learning.

At production sites such as factories, troubles may occur due to abnormalities in production equipment. At the production site, it is required to prevent the situation such as production stoppage due to the failure of the production equipment by grasping the occurrence of the abnormality at an early stage. In addition, preventive maintenance measures may be taken by predicting future outbreaks. Even if the production equipment is normal, the quality of the products produced may deteriorate and the yield may decrease. At the production site, measures are taken against the decrease in productivity due to phenomena such as abnormalities in production equipment or deterioration in product quality. Therefore, conventionally, various methods for diagnosing the state of production equipment by analyzing data collected from a production site have been proposed.

Machine learning methods are useful in diagnosis by analyzing data. Since machine learning requires high processing power, machine learning may be performed in the edge computing area rather than the equipment installed at the production site. In the edge computing area, machine learning for a plurality of production facilities can be efficiently performed by concentrating machine learning for a plurality of production facilities with one device. Further, it is more efficient to distribute the diagnosis of the state of the production equipment to the equipment installed at the production site than to perform the diagnosis with the equipment in the edge computing area. If a device in the edge computing area is to perform not only machine learning but also diagnosis, there is a problem that an expensive device having a higher processing capacity is required.

Patent Document 1 discloses a network system having a plurality of local systems, each of which monitors and controls a plant, and an operation support providing system that provides operation support information to each local system. According to Patent Document 1, each local system captures data indicating the operating state of the plant and detects plant abnormalities by obtaining the difference between the result of simulating the operation of the plant and the actual data captured from the plant. do. The driving support providing system collects the data captured by each local system and uses the collected data to generate a simulation model. The driving support providing system provides the generated simulation model to each local system. Each local system uses the simulation model provided by the driving assistance providing system in the simulation.

JP-A-2002-304211

The network system according to Patent Document 1 only transmits the simulation model generated by the operation support providing system to the local system, and the local system confirms whether or not the plant is operating according to the simulation. Therefore, according to the technique of Patent Document 1, it has not been possible to control a local system, which is a system at a production site, as a series of flows from data collection to an operation leading to productivity improvement.

The present disclosure has been made in view of the above, and an object of the present disclosure is to obtain a machine learning system that can control a system at a production site as a series of flows from data collection to operations leading to productivity improvement.

In order to solve the above-mentioned problems and achieve the object, the machine learning system according to the present disclosure includes a control device for controlling production equipment, an information processing device, and an inference device. The information processing device uses a data collection unit that collects status data indicating the operating status of production equipment, a data processing unit that processes the collected status data, and a trained model by machine learning using the processed status data. It has a learning unit to generate and an output processing unit to generate a program for collecting and processing state data and output a trained model. The inference device outputs the inference result based on the trained model to the control device by inputting the state data that has been collected and processed by executing the program.

The machine learning system according to the present disclosure has the effect of being able to control the system at the production site as a series of flows from data collection to operations leading to productivity improvement.

The figure which shows the structure of the machine learning system which concerns on Embodiment 1. The figure which shows the structure of the information processing apparatus included in the machine learning system which concerns on Embodiment 1. The figure for demonstrating the program executed in the PLC system among the machine learning system which concerns on Embodiment 1. A flowchart showing an operation procedure of the information processing apparatus included in the machine learning system according to the first embodiment. A flowchart showing an operation procedure of the PLC system included in the machine learning system according to the first embodiment. The figure for demonstrating the modification of the program executed in the PLC system among the machine learning system which concerns on Embodiment 1. The figure for demonstrating the 1st modification of the data processing flow in the machine learning system which concerns on Embodiment 1. The figure for demonstrating the 2nd modification of the data processing flow in the machine learning system which concerns on Embodiment 1. The figure for demonstrating the 3rd modification of the data processing flow in the machine learning system which concerns on Embodiment 1.

The machine learning system according to the embodiment will be described in detail below based on the drawings.

Embodiment 1.
FIG. 1 is a diagram showing a configuration of a machine learning system according to the first embodiment. The machine learning system 1 according to the first embodiment is a data processing platform that collects data from equipment installed at a production site and executes a series of processes for feeding back the processing result of the collected data to the equipment at the production site. Has. The machine learning system 1 includes an industrial PC (Industrial Personal Computer: IPC) 2 and a PLC (Programmable Logic Controller) system 3. IPC2 is a device that constitutes the above data processing platform. The PLC system 3 includes a PLC 4 which is a control device for controlling production equipment, and an AI (Artificial Intelligence) system 5 which is an inference device.

IPC2 is an information processing device located in the edge computing area. The edge computing area is an area on the data generation side with respect to the cloud computing area where centralized data processing is performed, and is a conceptual area where advanced data processing is performed. In the first embodiment, the edge computing area is an area in the factory. The machine learning system 1 implements machine learning in IPC2.

PLC system 3 is a system installed at the production site. The PLC4 controls the production equipment by executing a sequence program. The AI unit 5 outputs the inference result based on the trained model to the PLC4. The trained model is the result of machine learning by IPC2.

In the first embodiment, the AI unit 5 diagnoses the abnormality of the production equipment by inferring the presence or absence of the abnormality of the production equipment. The diagnosis performed by the AI unit 5 may be either a diagnosis of an abnormality in the entire production equipment or a diagnosis of an abnormality in each component of the production equipment. The AI unit 5 outputs the diagnosis result, which is the inference result, to the PLC 4. When the diagnosis result indicating that there is an abnormality is input from the AI unit 5, the PLC 4 takes measures such as stopping the operation of the production equipment or decelerating the drive source in the production equipment. The PLC system 3 can prevent a situation such as production suspension due to a failure of the production equipment by such measures for the diagnosis result. As a result, the machine learning system 1 can improve the productivity.

The AI unit 5 may perform a life diagnosis of the production equipment or a life diagnosis of each part by inferring the life of the parts constituting the production equipment. The diagnosis performed by the AI unit 5 is not limited to the diagnosis targeting the production equipment, and may be the diagnosis targeting the product manufactured by the production equipment. For example, the AI unit 5 may perform a quality diagnosis of a product. The diagnosis performed by the AI unit 5 may be any diagnosis useful for improving productivity or product quality at the production site.

In the first embodiment, the AI unit 5 may propose the production conditions by inferring the production conditions for improving the productivity of the production equipment. The AI unit 5 may output inference results useful for improving productivity at the production site or improving product quality. In the following description, the case where the AI unit 5 diagnoses an abnormality in the production equipment will be taken as an example.

IPC2 executes data processing including each process of collecting state data indicating the operating state of the production equipment, processing the collected state data, and learning using the processed state data. The operating state is the state of the components of the production equipment or the state of the substance in the production equipment at the time of operation of the production equipment, and can be quantified. A state is an electrical state, a mechanical state, a thermodynamic state, or a hydrodynamic state. The state data is data that quantifies the operating state. The state data includes data taken out from the inside of the production equipment and data obtained by detection by a sensor provided outside the production equipment. Further, the state data includes internal data representing the state of PLC4. The internal data is data stored in the internal memory of the PLC4.

IPC2 generates a program for PLC4 to perform the same process as each process of collection and processing in the data processing flow in the order of collection, processing and learning. The program generated by IPC2 may be referred to as a PLC program in the following description. IPC2 outputs the generated PLC program to PLC4. IPC2 also generates a trained model that is the result of machine learning. IPC2 outputs the generated trained model to AI unit 5.

The PLC4 collects the state data and processes the collected state data by executing the PLC program. The PLC 4 outputs the processed state data to the AI unit 5. The AI unit 5 has an inference engine for making a diagnosis based on the trained model. The AI unit 5 makes a diagnosis based on the trained model by inputting the processed state data. The AI unit 5 outputs the diagnosis result to the PLC4.

In this way, the machine learning system 1 causes the PLC4 to perform the same collection and processing as the IPC2, and also causes the AI unit 5 to perform the diagnosis based on the learned model generated by the IPC2. The machine learning system 1 can operate the PLC 4 with high responsiveness to the diagnosis result.

FIG. 2 is a diagram showing a configuration of an information processing device included in the machine learning system according to the first embodiment. A processing program for realizing the above data processing flow is installed in IPC2. FIG. 2 shows the hardware configuration of the IPC2 and the functional configuration realized by using the hardware configuration.

The IPC 2 includes a processor 10 that executes various processes, a memory 11 that is a built-in memory, a communication device 12 that communicates with an external device of the IPC 2, and a storage device 13 that stores various information.

The processor 10 is a CPU (Central Processing Unit). The processor 10 may be a processing device, an arithmetic unit, a microprocessor, a microcomputer, or a DSP (Digital Signal Processor). The memory 11 is a RAM (Random Access Memory), a ROM (Read Only Memory), a flash memory, an EPROM (Erasable Programmable Read Only Memory) or an EEPROM (registered trademark) (Electrically Erasable Programmable Read Only Memory). The storage device 13 is an HDD (Hard Disk Drive) or an SSD (Solid State Drive). The above processing program is stored in the storage device 13. The processor 10 reads the processing program stored in the storage device 13 into the memory 11 and executes it.

The processing program may be stored in a storage medium that can be read by a computer system. The IPC2 may store the processing program recorded in the storage medium in the memory 11. The storage medium may be a portable storage medium that is a flexible disk, or a flash memory that is a semiconductor memory. The processing program may be installed in a computer system from another computer or server device via a communication network.

The IPC2 has one or a plurality of data collection units 14 for collecting state data, a control unit 15 for controlling data processing in the IPC2, and a data processing unit 16 for processing the collected state data. Although two data collection units 14 are shown in FIG. 2, the IPC2 is provided with an arbitrary number of data collection units 14 according to the contents of the data processing flow executed by the IPC2.

The IPC2 generates a learning unit 17 that generates a trained model by machine learning using the processed state data, and a PLC program for causing the PLC4 to collect and process the state data, and outputs the PLC program to the PLC4. It has an output processing unit 18 and an output processing unit 18. Hereinafter, the PLC program is simply referred to as a program. The output processing unit 18 has a program generation unit 21 that generates a program. Further, the output processing unit 18 outputs the trained model to the AI unit 5.

The control unit 15 has a distribution unit 19 that controls data distribution between each data collection unit 14, a data processing unit 16 and a learning unit 17, and a setting management unit 20 that manages setting information. The IPC2 has a setting information storage unit 22 that stores setting information. The setting information will be described later.

Each function of each data collection unit 14, control unit 15, data processing unit 16, learning unit 17, and output processing unit 18 is realized by a combination of the processor 10 and software. The functions of the data collection unit 14, the control unit 15, the data processing unit 16, the learning unit 17, and the output processing unit 18 may be realized by a combination of the processor 10 and the firmware, and may be realized by the combination of the processor 10, the software, and the firmware. It may be realized. The software or firmware is described as a program and stored in the storage device 13. The function of the setting information storage unit 22 is realized by using the storage device 13.

The external device 6 is a device installed outside the PLC 4 and is a device such as a sensor that detects the operating state of the production equipment. Examples of the sensor include a sensor attached to a drive source of production equipment to detect vibration, a sensor to detect temperature, and the like. The sensor may be either a sensor installed inside the production equipment or a sensor installed outside the production equipment.

The communication device 12 is communicably connected to each of the PLC 4, the AI unit 5, and the external device 6. Each of the PLC 4, the AI unit 5, and the external device 6 is connected to the communication device 12 via a network by wireless communication or a network by wired communication. The external device 6 transmits the state data, which is the result of detecting the operating state of the production equipment, to the IPC2. The PLC4 transmits the state data, which is the internal data of the PLC4, to the IPC2.

The setting information stored in the setting information storage unit 22 includes the execution order of processing in each process of collection and processing, the timing of execution of each process, and the conditions for executing each process in the data processing flow executed by IPC2. Information set for the item. The setting management unit 20 manages the setting information stored in the setting information storage unit 22. Further, the setting management unit 20 sends the setting information read from the setting information storage unit 22 to the distribution unit 19 or the program generation unit 21. The distribution unit 19 controls data distribution between each data collection unit 14, the data processing unit 16, and the learning unit 17 based on the setting information acquired from the setting management unit 20.

The learning unit 17, which is a machine learning device, generates a trained model by, for example, supervised learning according to a neural network. Supervised learning is learning in which a large number of sets of data, which are an input and a label as a result, are given to a machine learning device to learn the characteristics of the data set and estimate the result from the input. The state data processed by the data processing unit 16 and the label which is the information of the diagnosis result are input to the learning unit 17. The learning unit 17 generates learning data which is data in which the state data and the label are associated with each other. The learning unit 17 generates a learned model for inferring the optimum diagnostic result from the state data and the label.

The learning unit 17 generates a learned model for each production facility by learning using the state data for each production facility. The learning unit 17 may generate a trained model for each function by learning using the state data for each function in the production equipment. The learning unit 17 may generate a learned model for each part by learning using the state data for each part constituting the production equipment. The learning unit 17 outputs the generated learned model to the output processing unit 18.

The program generation unit 21 generates a program for causing the PLC 4 to collect and process state data based on the setting information acquired from the setting management unit 20. The program generation unit 21 generates a program for realizing the same functions as "collection" and "processing" in the data processing flow in IPC2 in PLC4.

In the first embodiment, the program generated by the program generation unit 21 is a ladder program written using the ladder language. The program may be a program written in a language other than the ladder language, a program written in the structured ladder language, or a program written in the function block diagram language.

The output processing unit 18 outputs the program generated by the program generation unit 21 to the PLC 4. The output processing unit 18 outputs the trained model generated by the learning unit 17 to the AI unit 5. As a result, the IPC 2 provides the PLC system 3 with a program for collecting and processing the state data and a trained model for inferring the diagnostic result based on the state data.

FIG. 3 is a diagram for explaining a program executed in the PLC system among the machine learning systems according to the first embodiment. FIG. 3 shows a program 30 executed in the PLC 4 and a diagnostic program 31 executed in the AI unit 5.

PLC4 holds the program 30 acquired from IPC2. The PLC4 collects and processes state data in the same manner as the IPC2 by reading and executing the held program 30.

The diagnostic program 31 is a program for obtaining a diagnostic result based on a trained model from the state data collected and processed by executing the program 30 in PLC4. The AI unit 5 holds the diagnostic program 31. The AI unit 5 reads and executes the held diagnostic program 31 to diagnose an abnormality in the production equipment from the state data.

The ladder program is composed of a plurality of circuit blocks. The circuit block is composed of a combination of a condition unit, which is a group of circuits in which contacts are connected in series or in parallel, and an operation unit, which is a group of circuits in which one or more coils are connected in series. The operating unit represents the content of arithmetic processing executed when the contacts of the condition unit are conducted. Program code written using the ladder language contains circuit symbols and variables that are the basic circuit elements. Circuit symbols include contacts and coils that represent processing in PLC4. Each variable is associated with each of the plurality of data areas included in the PLC4. Arithmetic data for each circuit element is stored in each memory area. The arithmetic data includes bit data that expresses the distinction between on and off, and word data that expresses a numerical value. The variables used in the programming of the ladder program, or the data areas to which the variables are associated, are referred to as "devices".

In the program 30, for example, a device number representing a device in which a value which is internal data acquired by the data collection unit 14 is stored, an input contact for switching on / off of data input from the device, and data output to the device. The output contact that switches on and off of, and the device number that represents the device to which the value output from the output contact is input are included. The program 30 also includes a timer for managing the on / off switching of the input contact or the output contact according to the time. Such a circuit element included in the program 30 realizes a function of collecting state data at a timing according to the setting information.

The program 30 includes, for example, an FB (function block) that smoothes data, and a counter that counts the number of times the input contact or output contact is switched on and off. The PLC4 smoothes the state data by executing the FB. The PLC4 extracts the state data according to the count by the counter. Such a circuit element included in the program 30 realizes a function of processing state data.

In the diagnostic program 31, the device that is the acquisition destination of the state data used for the diagnosis and the device that is the output destination of the diagnosis result are specified. By executing the diagnosis program 31, the AI unit 5 makes a diagnosis based on the processed state data and the trained model. The AI unit 5 outputs the diagnosis result to the PLC4.

Next, the operation of IPC2 will be described. FIG. 4 is a flowchart showing an operation procedure of the information processing apparatus included in the machine learning system according to the first embodiment. Upon receiving the command to start the generation of the trained model, the IPC2 collects the state data in step S1. In step S2, IPC2 processes the state data collected in step S1. In step S3, IPC2 generates a trained model by machine learning using state data.

In step S4, IPC2 generates program 30, which is a ladder program for collecting and processing state data. The IPC2 generates the program 30 in the program generation unit 21 based on the setting information.

In step S5, IPC2 transmits the program 30, which is the ladder program generated in step S5, and the trained model generated in step S3 to the PLC system 3. The IPC2 transmits the program 30 from the communication device 12 to the PLC4. The IPC2 transmits the trained model from the communication device 12 to the AI unit 5. As a result, the IPC 2 provides the program 30 and the trained model to the PLC system 3. As described above, the IPC2 ends the operation according to the procedure shown in FIG.

Note that in step S5, the IPC2 may transmit the trained model to the PLC4 instead of the AI unit 5. In this case, the PLC 4 outputs the received trained model to the AI unit 5. In this way, the IPC2 may provide the trained model to the AI unit 5 via the PLC4.

Next, the operation of the PLC system 3 will be described. FIG. 5 is a flowchart showing an operation procedure of the PLC system included in the machine learning system according to the first embodiment. In step S11, the PLC system 3 receives the program 30 which is a ladder program and the trained model. The PLC system 3 receives the program 30 in the PLC 4. The PLC system 3 receives the trained model in the AI unit 5. As described above, the PLC system 3 may receive the program 30 and the trained model in the PLC 4. The PLC system 3 receives the program 30 and the trained model at the time of the first diagnosis. The PLC system 3 receives the updated trained model when the trained model is updated after the first diagnosis.

The PLC system 3 performs an operation for diagnosis after receiving the program 30 and the trained model. In step S12, the PLC system 3 collects state data at the PLC 4. The PLC4 collects state data by executing the program 30 received in step S11.

In step S13, the PLC system 3 processes the state data collected in step S12 in PLC4. The PLC4 processes the state data by executing the program 30 received in step S11. The PLC 4 outputs the processed state data to the AI unit 5.

In step S14, the PLC system 3 diagnoses the production equipment in the AI unit 5. The AI unit 5 obtains a diagnosis result based on the state data input from the PLC 4 and the trained model received in step S11. In step S15, the PLC system 3 inputs the diagnosis result obtained in step S14 from the AI unit 5 to the PLC 4. As described above, the PLC system 3 ends the operation according to the procedure shown in FIG.

Next, in IPC2, a method of generating the program 30 based on the setting information of the data processing flow will be described. The program generation unit 21 confirms the processing contents for each data processing element with respect to the data processing elements "collection" and "processing" included in the setting information. The program generation unit 21 selects a program element which is a circuit element for realizing processing for each data processing element. Program elements selected include contacts, coils, timers, counters and FBs.

The program generation unit 21 selects a contact point for "collection", for example. For example, for "machining", the program generation unit 21 selects an FB whose function is pre-programmed in C language. The program element selected for machining may be an FB in which a ladder program for realizing a function related to "machining" is put together. By selecting the FB as the selection target, the program generation unit 21 can easily select the program element for realizing the processing for each data processing element.

Next, the program generation unit 21 sets the connection relationship between the selected program elements by confirming the connection relationship between the data processing elements. The connection relationship between data processing elements includes a connection relationship such as branching or joining in a data processing flow. Further, the program generation unit 21 confirms the device that is the acquisition destination of the data input to the data processing element and the device that is the input destination of the data output from the data processing element. The program generation unit 21 sets a device that is an acquisition destination of data input to the program element corresponding to the data processing element and a device that is an input destination of data output from the program element corresponding to the data processing element. ..

When a data processing element including a processing condition such as a data sampling time or a data processing timing is included in the data processing flow, the program generation unit 21 sets the processing condition for the program element corresponding to the data processing element. Set the corresponding parameters. For example, when a condition regarding the data sampling time is set, the program generation unit 21 sets the time parameter of the timer, which is a program element, according to the sampling time condition. When a condition regarding the timing of data processing is set, the program generation unit 21 sets a parameter regarding the timing at which the contact, which is a program element, is switched according to the condition.

In the first embodiment, the case where supervised learning is applied to the learning algorithm used by the learning unit 17 has been described, but the present invention is not limited to this. In addition to supervised learning, reinforcement learning, unsupervised learning, semi-supervised learning, and the like can also be applied to the learning algorithm. Deep learning, which learns the extraction of the feature amount itself, can also be used as the learning algorithm. The learning unit 17 may perform machine learning according to other known methods such as genetic programming, functional logic programming, and support vector machines.

The machine learning device, which is the learning unit 17, is not limited to the one built in the IPC2, and may be provided outside the IPC2. The machine learning device may be a device connected to the IPC2 via a network. The machine learning device may exist on the cloud server.

The learning unit 17 may generate a trained model according to the data sets created for the plurality of machine learning systems 1. The learning unit 17 may acquire data sets from a plurality of machine learning systems 1 used at the same site, or may acquire data sets from a plurality of machine learning systems 1 used at different sites. It may be used to generate a trained model. After the learning unit 17 starts acquiring the data set, a new machine learning system 1 may be added to the target for which the data set is acquired. Further, after starting the acquisition of the data set from the plurality of machine learning systems 1, a part of the plurality of machine learning systems 1 may be excluded from the target for which the data set is acquired.

The learning unit 17 that has learned in one machine learning system 1 may be attached to a machine learning system 1 other than the machine learning system 1. The learning unit 17 attached to the other machine learning system 1 can update the trained model by re-learning in the other machine learning system 1.

When the diagnosis result is input from the AI unit 5 to the PLC4, the PLC4 causes the production equipment to be diagnosed to perform an operation according to the diagnosis result. The PLC 4 stops the operation of the production equipment when the diagnosis result indicating that there is an abnormality is input from the AI unit 5. Alternatively, the PLC 4 slows down the drive source in the production equipment. Since the production equipment for which the state data is collected is specified in the program 30, the PLC4 can identify the production equipment having an abnormality. The sequence program executed by the PLC4 to control the production equipment describes the processing when the PLC4 receives an interrupt signal due to a malfunction of the production equipment or the like. The PLC4 receives a signal which is a diagnosis result indicating that there is an abnormality as an interrupt signal, and executes a process according to the diagnosis result as a process when the interrupt signal is received. As a result, the machine learning system 1 can control the production equipment according to the diagnosis result by the AI unit 5.

Next, a modified example of the first embodiment will be described. The machine learning system 1 is not limited to the one in which the AI unit 5 makes inferences based on the trained model. The inference based on the trained model may be performed by the PLC 4, and the machine learning system 1 may not be provided with the AI unit 5. In this case, the PLC 4 is provided with a processing unit that collects and processes state data by executing the program 30, and an inference unit that is an inference device. When the state data collected and processed by the execution of the program 30 in the processing unit as the control device is input to the inference unit, the inference unit outputs the inference result based on the learned model to the processing unit. The processing unit and the inference unit are not shown.

FIG. 6 is a diagram for explaining a modified example of a program executed in the PLC system among the machine learning systems according to the first embodiment. FIG. 6 shows an example of a program generated by the program generation unit 21 when the PLC4 infers a diagnosis result based on the state data. The program generation unit 21 generates a program including a program 30 for collecting and processing state data and a diagnostic program 32 which is a ladder program for diagnosis. The diagnostic program 32 includes FB33, which is a functional unit for performing a diagnosis based on the trained model. The inference unit outputs the diagnosis result based on the learned model to the processing unit by executing the diagnosis program 32.

The machine learning system 1 is not limited to one that shares functions such that IPC2 is responsible for learning and PLC system 3 is responsible for data collection for diagnosis to diagnosis. FIG. 7 is a diagram for explaining a first modification of the data processing flow in the machine learning system according to the first embodiment. In the first modification, the machine learning system 1 performs from data collection to diagnosis in both the IPC 2 and the PLC system 3. That is, both the IPC2 and the PLC system 3 are responsible for collecting, processing, and diagnosing functions. When both the IPC 2 and the PLC system 3 perform the diagnosis, the machine learning system 1 can compare and verify the diagnosis result by the IPC 2 and the diagnosis result by the PLC system 3. Further, the machine learning system 1 can continue the diagnosis even if one of the IPC 2 and the PLC system 3 has a problem.

FIG. 8 is a diagram for explaining a second modification of the data processing flow in the machine learning system according to the first embodiment. In the second modification, in the machine learning system 1, the IPC2 is responsible for collection and processing, and the PLC system 3 is responsible for diagnosis. FIG. 9 is a diagram for explaining a third modification of the data processing flow in the machine learning system according to the first embodiment. In the third modification, in the machine learning system 1, the PLC system 3 is responsible for collection, and the IPC2 is responsible for processing and diagnosis. In the second modification and the third modification, the machine learning system 1 shares the collection, processing, and diagnosis functions between the IPC 2 and the PLC system 3. By sharing collection, processing and diagnosis, the machine learning system 1 can reduce the processing load in each of the IPC 2 and the PLC system 3.

According to the first embodiment, the IPC 2 outputs the trained model generated by the learning unit 17 and the program 30 generated by the output processing unit 18 to the PLC system 3. The PLC system 3 collects and processes state data by executing the program 30 in the PLC 4. The PLC system 3 outputs the inference result based on the trained model from the AI unit 5 to the PLC 4 by inputting the processed state data to the AI unit 5. As described above, the machine learning system 1 has the effect of being able to control the system at the production site as a series of flows from data collection to operations leading to productivity improvement.

The configuration shown in the above embodiments is an example of the contents of the present disclosure. The configurations shown in the embodiments can be combined with other known techniques. The configurations shown in the embodiments may be combined as appropriate. A part of the configuration shown in the embodiment may be omitted or changed without departing from the gist of the present disclosure.

1 machine learning system, 2 IPC, 3 PLC system, 4 PLC, 5 AI unit, 6 external device, 10 processor, 11 memory, 12 communication device, 13 storage device, 14 data collection unit, 15 control unit, 16 data processing unit , 17 learning unit, 18 output processing unit, 19 distribution unit, 20 setting management unit, 21 program generation unit, 22 setting information storage unit, 30 program, 31, 32 diagnostic program, 33 FB.

Claims

A control device that controls production equipment and
A data collection unit that collects state data indicating the operating state of the production equipment, a data processing unit that processes the collected state data, and a learning unit that generates a trained model by machine learning using the processed state data. An information processing device having an output processing unit that generates a program for collecting and processing state data and outputs the trained model.
A machine learning system including a reasoning device that outputs a reasoning result based on the learned model to the control device by inputting state data that has been collected and processed by executing the program.
The output processing unit has a program generation unit that generates a ladder program, which is the program for causing the control device to collect and process state data.
The machine learning system according to claim 1, wherein state data that has been collected and processed by executing the program in the control device is input to the inference device.
The second aspect of claim 2, wherein the program generation unit generates the ladder program based on the setting information set for the processing in each process of the collection and the processing executed by the information processing apparatus. Machine learning system.
The machine learning system according to claim 2 or 3, wherein the program generation unit generates the ladder program including a function block for processing.